@article{heine_techniques_2023,
	title = {Techniques to produce and evaluate realistic multivariate synthetic data},
	volume = {13},
	copyright = {2023 The Author(s)},
	issn = {2045-2322},
	url = {https://www.nature.com/articles/s41598-023-38832-0},
	doi = {10.1038/s41598-023-38832-0},
	abstract = {Data modeling requires a sufficient sample size for reproducibility. A small sample size can inhibit model evaluation. A synthetic data generation technique addressing this small sample size problem is evaluated: from the space of arbitrarily distributed samples, a subgroup (class) has a latent multivariate normal characteristic; synthetic data can be generated from this class with univariate kernel density estimation (KDE); and synthetic samples are statistically like their respective samples. Three samples (n = 667) were investigated with 10 input variables (X). KDE was used to augment the sample size in X. Maps produced univariate normal variables in Y. Principal component analysis in Y produced uncorrelated variables in T, where the probability density functions were approximated as normal and characterized; synthetic data was generated with normally distributed univariate random variables in T. Reversing each step produced synthetic data in Y and X. All samples were approximately multivariate normal in Y, permitting the generation of synthetic data. Probability density function and covariance comparisons showed similarity between samples and synthetic samples. A class of samples has a latent normal characteristic. For such samples, this approach offers a solution to the small sample size problem. Further studies are required to understand this latent class.},
	language = {en},
	number = {1},
	urldate = {2023-10-17},
	journal = {Scientific Reports},
	author = {Heine, John and Fowler, Erin E. E. and Berglund, Anders and Schell, Michael J. and Eschrich, Steven},
	month = jul,
	year = {2023},
	note = {Number: 1
Publisher: Nature Publishing Group},
	keywords = {Applied mathematics, Computational science, Data processing, Predictive medicine, Scientific data, Statistical methods, Statistics},
	pages = {12266},
}

Data modeling requires a sufficient sample size for reproducibility. A small sample size can inhibit model evaluation. A synthetic data generation technique addressing this small sample size problem is evaluated: from the space of arbitrarily distributed samples, a subgroup (class) has a latent multivariate normal characteristic; synthetic data can be generated from this class with univariate kernel density estimation (KDE); and synthetic samples are statistically like their respective samples. Three samples (n = 667) were investigated with 10 input variables (X). KDE was used to augment the sample size in X. Maps produced univariate normal variables in Y. Principal component analysis in Y produced uncorrelated variables in T, where the probability density functions were approximated as normal and characterized; synthetic data was generated with normally distributed univariate random variables in T. Reversing each step produced synthetic data in Y and X. All samples were approximately multivariate normal in Y, permitting the generation of synthetic data. Probability density function and covariance comparisons showed similarity between samples and synthetic samples. A class of samples has a latent normal characteristic. For such samples, this approach offers a solution to the small sample size problem. Further studies are required to understand this latent class.

@misc{wen_transformers_2023,
	title = {Transformers in {Time} {Series}: {A} {Survey}},
	shorttitle = {Transformers in {Time} {Series}},
	url = {http://arxiv.org/abs/2202.07125},
	doi = {10.48550/arXiv.2202.07125},
	abstract = {Transformers have achieved superior performances in many tasks in natural language processing and computer vision, which also triggered great interest in the time series community. Among multiple advantages of Transformers, the ability to capture long-range dependencies and interactions is especially attractive for time series modeling, leading to exciting progress in various time series applications. In this paper, we systematically review Transformer schemes for time series modeling by highlighting their strengths as well as limitations. In particular, we examine the development of time series Transformers in two perspectives. From the perspective of network structure, we summarize the adaptations and modifications that have been made to Transformers in order to accommodate the challenges in time series analysis. From the perspective of applications, we categorize time series Transformers based on common tasks including forecasting, anomaly detection, and classification. Empirically, we perform robust analysis, model size analysis, and seasonal-trend decomposition analysis to study how Transformers perform in time series. Finally, we discuss and suggest future directions to provide useful research guidance. To the best of our knowledge, this paper is the first work to comprehensively and systematically summarize the recent advances of Transformers for modeling time series data. We hope this survey will ignite further research interests in time series Transformers.},
	urldate = {2023-10-17},
	publisher = {arXiv},
	author = {Wen, Qingsong and Zhou, Tian and Zhang, Chaoli and Chen, Weiqi and Ma, Ziqing and Yan, Junchi and Sun, Liang},
	month = may,
	year = {2023},
	note = {arXiv:2202.07125 [cs, eess, stat]},
	keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning, Electrical Engineering and Systems Science - Signal Processing, Statistics - Machine Learning},
}

Transformers have achieved superior performances in many tasks in natural language processing and computer vision, which also triggered great interest in the time series community. Among multiple advantages of Transformers, the ability to capture long-range dependencies and interactions is especially attractive for time series modeling, leading to exciting progress in various time series applications. In this paper, we systematically review Transformer schemes for time series modeling by highlighting their strengths as well as limitations. In particular, we examine the development of time series Transformers in two perspectives. From the perspective of network structure, we summarize the adaptations and modifications that have been made to Transformers in order to accommodate the challenges in time series analysis. From the perspective of applications, we categorize time series Transformers based on common tasks including forecasting, anomaly detection, and classification. Empirically, we perform robust analysis, model size analysis, and seasonal-trend decomposition analysis to study how Transformers perform in time series. Finally, we discuss and suggest future directions to provide useful research guidance. To the best of our knowledge, this paper is the first work to comprehensively and systematically summarize the recent advances of Transformers for modeling time series data. We hope this survey will ignite further research interests in time series Transformers.

@misc{ma_survey_2023,
	title = {A {Survey} on {Time}-{Series} {Pre}-{Trained} {Models}},
	url = {http://arxiv.org/abs/2305.10716},
	doi = {10.48550/arXiv.2305.10716},
	abstract = {Time-Series Mining (TSM) is an important research area since it shows great potential in practical applications. Deep learning models that rely on massive labeled data have been utilized for TSM successfully. However, constructing a large-scale well-labeled dataset is difficult due to data annotation costs. Recently, Pre-Trained Models have gradually attracted attention in the time series domain due to their remarkable performance in computer vision and natural language processing. In this survey, we provide a comprehensive review of Time-Series Pre-Trained Models (TS-PTMs), aiming to guide the understanding, applying, and studying TS-PTMs. Specifically, we first briefly introduce the typical deep learning models employed in TSM. Then, we give an overview of TS-PTMs according to the pre-training techniques. The main categories we explore include supervised, unsupervised, and self-supervised TS-PTMs. Further, extensive experiments are conducted to analyze the advantages and disadvantages of transfer learning strategies, Transformer-based models, and representative TS-PTMs. Finally, we point out some potential directions of TS-PTMs for future work.},
	urldate = {2023-10-17},
	publisher = {arXiv},
	author = {Ma, Qianli and Liu, Zhen and Zheng, Zhenjing and Huang, Ziyang and Zhu, Siying and Yu, Zhongzhong and Kwok, James T.},
	month = may,
	year = {2023},
	note = {arXiv:2305.10716 [cs]},
	keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning},
}

Time-Series Mining (TSM) is an important research area since it shows great potential in practical applications. Deep learning models that rely on massive labeled data have been utilized for TSM successfully. However, constructing a large-scale well-labeled dataset is difficult due to data annotation costs. Recently, Pre-Trained Models have gradually attracted attention in the time series domain due to their remarkable performance in computer vision and natural language processing. In this survey, we provide a comprehensive review of Time-Series Pre-Trained Models (TS-PTMs), aiming to guide the understanding, applying, and studying TS-PTMs. Specifically, we first briefly introduce the typical deep learning models employed in TSM. Then, we give an overview of TS-PTMs according to the pre-training techniques. The main categories we explore include supervised, unsupervised, and self-supervised TS-PTMs. Further, extensive experiments are conducted to analyze the advantages and disadvantages of transfer learning strategies, Transformer-based models, and representative TS-PTMs. Finally, we point out some potential directions of TS-PTMs for future work.

@article{sehri_university_2023,
	title = {University of {Ottawa} constant load and speed rolling-element bearing vibration and acoustic fault signature datasets},
	volume = {49},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340923004456},
	doi = {10.1016/j.dib.2023.109327},
	abstract = {The collection and analysis of data play a critical role in detecting and diagnosing faults in bearings. However, the availability of large open-access rolling-element bearing datasets for fault diagnosis is limited. To overcome this challenge, the University of Ottawa Rolling-element Bearing Vibration and Acoustic Fault Signature Datasets Operating under Constant Load and Speed Conditions are introduced to provide supplementary data that can be combined or merged with existing bearing datasets to increase the amount of data available to researchers. This data utilizes various sensors such as an accelerometer, a microphone, a load cell, a hall effect sensor, and thermocouples to gather quality data on bearing health. By incorporating vibration and acoustic signals, the datasets enable both traditional and machine learning-based approaches for rolling-element bearing fault diagnosis. Furthermore, this dataset offers valuable insights into the accelerated deterioration of bearing life under constant loads, making it an invaluable resource for research in this domain. Ultimately, these datasets deliver high quality data for the detection and diagnosis of faults in rolling-element bearings, thereby holding significant implications for machinery operation and maintenance.},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Sehri, Mert and Dumond, Patrick and Bouchard, Michel},
	month = aug,
	year = {2023},
	keywords = {Fault detection/Diagnosis, Machine condition monitoring, Signal processing, Vibration},
	pages = {109327},
}

The collection and analysis of data play a critical role in detecting and diagnosing faults in bearings. However, the availability of large open-access rolling-element bearing datasets for fault diagnosis is limited. To overcome this challenge, the University of Ottawa Rolling-element Bearing Vibration and Acoustic Fault Signature Datasets Operating under Constant Load and Speed Conditions are introduced to provide supplementary data that can be combined or merged with existing bearing datasets to increase the amount of data available to researchers. This data utilizes various sensors such as an accelerometer, a microphone, a load cell, a hall effect sensor, and thermocouples to gather quality data on bearing health. By incorporating vibration and acoustic signals, the datasets enable both traditional and machine learning-based approaches for rolling-element bearing fault diagnosis. Furthermore, this dataset offers valuable insights into the accelerated deterioration of bearing life under constant loads, making it an invaluable resource for research in this domain. Ultimately, these datasets deliver high quality data for the detection and diagnosis of faults in rolling-element bearings, thereby holding significant implications for machinery operation and maintenance.

@article{jung_vibration_2023,
	title = {Vibration and current dataset of three-phase permanent magnet synchronous motors with stator faults},
	volume = {47},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340923000707},
	doi = {10.1016/j.dib.2023.108952},
	abstract = {Permanent magnet synchronous motors (PMSM) are widely used in industry applications such as home appliances, manufacturing process, high-speed trains, and electric vehicles. Unexpected faults of PMSM are directly related to the significant losses in the engineered systems. The majority of motor faults are bearing fault (mechanical) and stator fault (electrical). This article reports vibration and driving current dataset of three-phase PMSM with three different motor powers under eight different severities of stator fault. PMSM conditions including normal, inter-coil short circuit fault, and inter-turn short circuit fault in three motors are demonstrated with different powers of 1.0 kW, 1.5 kW and 3.0 kW, respectively. The PMSMs are operated under the same torque load condition and rotating speed. Dataset is acquired using one integrated electronics piezo-electric (IEPE) based accelerometer and three current transformers (CT) with National Instruments (NI) data acquisition (DAQ) board under international organization for standardization standard (ISO 10816-1:1995). Established dataset can be used to verify newly developed state-of-the-art methods for PMSM stator fault diagnosis. Mendeley Data. DOI: 10.17632/rgn5brrgrn.5},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Jung, Wonho and Yun, Sung-Hyun and Lim, Yoon-Seop and Cheong, Sungjin and Park, Yong-Hwa},
	month = apr,
	year = {2023},
	keywords = {Condition Monitoring, Current, Fault Diagnosis, Stator Fault, Three-Phase Permanent Magnet Synchronous Motors, Vibration},
	pages = {108952},
}

Permanent magnet synchronous motors (PMSM) are widely used in industry applications such as home appliances, manufacturing process, high-speed trains, and electric vehicles. Unexpected faults of PMSM are directly related to the significant losses in the engineered systems. The majority of motor faults are bearing fault (mechanical) and stator fault (electrical). This article reports vibration and driving current dataset of three-phase PMSM with three different motor powers under eight different severities of stator fault. PMSM conditions including normal, inter-coil short circuit fault, and inter-turn short circuit fault in three motors are demonstrated with different powers of 1.0 kW, 1.5 kW and 3.0 kW, respectively. The PMSMs are operated under the same torque load condition and rotating speed. Dataset is acquired using one integrated electronics piezo-electric (IEPE) based accelerometer and three current transformers (CT) with National Instruments (NI) data acquisition (DAQ) board under international organization for standardization standard (ISO 10816-1:1995). Established dataset can be used to verify newly developed state-of-the-art methods for PMSM stator fault diagnosis. Mendeley Data. DOI: 10.17632/rgn5brrgrn.5

@article{ahern_dataset_2023,
	title = {A dataset for fault detection and diagnosis of an air handling unit from a real industrial facility},
	volume = {48},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S235234092300327X},
	doi = {10.1016/j.dib.2023.109208},
	abstract = {This dataset was collected for the purpose of applying fault detection and diagnosis (FDD) techniques to real data from an industrial facility. The data for an air handling unit (AHU) is extracted from a building management system (BMS) and aligned with the Project Haystack naming convention. This dataset differs from other publicly available datasets in three main ways. Firstly, the dataset does not contain fault detection ground truth. The lack of labelled datasets in the industrial setting is a significant limitation to the application of FDD techniques found in the literature. Secondly, unlike other publicly available datasets that typically record values every 1 min or 5 min, this dataset captures measurements at a lower frequency of every 15 min, which is due to data storage constraints. Thirdly, the dataset contains a myriad of data issues. For example, there are missing features, missing time intervals, and inaccurate data. Therefore, we hope this dataset will encourage the development of robust FDD techniques that are more suitable for real world applications.},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Ahern, Michael and O'Sullivan, Dominic T. J. and Bruton, Ken},
	month = jun,
	year = {2023},
	keywords = {Detection, HVAC data, Real data, Time series},
	pages = {109208},
}

This dataset was collected for the purpose of applying fault detection and diagnosis (FDD) techniques to real data from an industrial facility. The data for an air handling unit (AHU) is extracted from a building management system (BMS) and aligned with the Project Haystack naming convention. This dataset differs from other publicly available datasets in three main ways. Firstly, the dataset does not contain fault detection ground truth. The lack of labelled datasets in the industrial setting is a significant limitation to the application of FDD techniques found in the literature. Secondly, unlike other publicly available datasets that typically record values every 1 min or 5 min, this dataset captures measurements at a lower frequency of every 15 min, which is due to data storage constraints. Thirdly, the dataset contains a myriad of data issues. For example, there are missing features, missing time intervals, and inaccurate data. Therefore, we hope this dataset will encourage the development of robust FDD techniques that are more suitable for real world applications.

@article{jung_vibration_2023-1,
	title = {Vibration, acoustic, temperature, and motor current dataset of rotating machine under varying operating conditions for fault diagnosis},
	volume = {48},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340923001671},
	doi = {10.1016/j.dib.2023.109049},
	abstract = {Rotating machines are often operated under various operating conditions. However, the characteristics of the data varies with their operating conditions. This article presents the time-series dataset, including vibration, acoustic, temperature, and driving current data of rotating machines under varying operating conditions. The dataset was acquired using four ceramic shear ICP based accelerometers, one microphone, two thermocouples, and three current transformer (CT) based on the international organization for standardization (ISO) standard. The conditions of the rotating machine consisted of normal, bearing faults (inner and outer races), shaft misalignment, and rotor unbalance with three different torque load conditions (0 Nm, 2 Nm, and 4 Nm). This article also reports the vibration and driving current dataset of a rolling element bearing under varying speed conditions (680 RPM to 2460 RPM). The established dataset can be used to verify newly developed state-of-the-art methods for fault diagnosis of rotating machines. Mendeley Data. DOI:10.17632/ztmf3m7h5x.6, DOI:10.17632/vxkj334rzv.7, DOI:10.17632/x3vhp8t6hg.7, DOI:10.17632/j8d8pfkvj2.7},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Jung, Wonho and Kim, Seong-Hu and Yun, Sung-Hyun and Bae, Jaewoong and Park, Yong-Hwa},
	month = jun,
	year = {2023},
	keywords = {Ball Bearing, Condition Monitoring, Load Fluctuation, Misalignment, Speed Fluctuation, Unbalance},
	pages = {109049},
}

Rotating machines are often operated under various operating conditions. However, the characteristics of the data varies with their operating conditions. This article presents the time-series dataset, including vibration, acoustic, temperature, and driving current data of rotating machines under varying operating conditions. The dataset was acquired using four ceramic shear ICP based accelerometers, one microphone, two thermocouples, and three current transformer (CT) based on the international organization for standardization (ISO) standard. The conditions of the rotating machine consisted of normal, bearing faults (inner and outer races), shaft misalignment, and rotor unbalance with three different torque load conditions (0 Nm, 2 Nm, and 4 Nm). This article also reports the vibration and driving current dataset of a rolling element bearing under varying speed conditions (680 RPM to 2460 RPM). The established dataset can be used to verify newly developed state-of-the-art methods for fault diagnosis of rotating machines. Mendeley Data. DOI:10.17632/ztmf3m7h5x.6, DOI:10.17632/vxkj334rzv.7, DOI:10.17632/x3vhp8t6hg.7, DOI:10.17632/j8d8pfkvj2.7

@article{ogaili_wind_2023,
	title = {Wind turbine blades fault diagnosis based on vibration dataset analysis},
	volume = {49},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340923005152},
	doi = {10.1016/j.dib.2023.109414},
	abstract = {Globally, wind turbines play a significant role in generating sustainable and clean energy. Ensuring optimal performance and reliability is crucial to minimize failures and reduce operating and maintenance costs. However, due to their conventional design, identifying faults in wind turbines is challenging. This dataset provides vibration data for faulty wind turbine blades, which covers common vibration excitation mechanisms associated with various faults and operating conditions, including wind speed. The introduced faults in the wind turbine blades include surface erosion, cracked blade, mass imbalance, and twist blade fault. This data article serves as a valuable resource for validating condition monitoring methods in industrial wind turbine applications and facilitates a better understanding of vibration signal characteristics associated with different faults.},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Ogaili, Ahmed Ali Farhan and Abdulhady Jaber, Alaa and Hamzah, Mohsin Noori},
	month = aug,
	year = {2023},
	keywords = {Condition monitoring, Fault diagnosis, Vibration signal analysis, Wind turbine blade},
	pages = {109414},
}

Globally, wind turbines play a significant role in generating sustainable and clean energy. Ensuring optimal performance and reliability is crucial to minimize failures and reduce operating and maintenance costs. However, due to their conventional design, identifying faults in wind turbines is challenging. This dataset provides vibration data for faulty wind turbine blades, which covers common vibration excitation mechanisms associated with various faults and operating conditions, including wind speed. The introduced faults in the wind turbine blades include surface erosion, cracked blade, mass imbalance, and twist blade fault. This data article serves as a valuable resource for validating condition monitoring methods in industrial wind turbine applications and facilitates a better understanding of vibration signal characteristics associated with different faults.

@article{de_pater_developing_2023,
	title = {Developing health indicators and {RUL} prognostics for systems with few failure instances and varying operating conditions using a {LSTM} autoencoder},
	volume = {117},
	issn = {0952-1976},
	url = {https://www.sciencedirect.com/science/article/pii/S0952197622005723},
	doi = {10.1016/j.engappai.2022.105582},
	abstract = {Most Remaining Useful Life (RUL) prognostics are obtained using supervised learning models trained with many labelled data samples (i.e., the true RUL is known). In aviation, however, aircraft systems are often preventively replaced before failure. There are thus very few labelled data samples available. We therefore propose a Long Short-Term Memory (LSTM) autoencoder with attention to develop health indicators for an aircraft system instead. This autoencoder is trained with unlabelled data samples (i.e., the true RUL is unknown). Since aircraft fly under various operating conditions (varying altitude, speed, etc.), these conditions are also integrated in the autoencoder. We show that the consideration of the operating conditions leads to robust health indicators and improves significantly the monotonicity, trendability and prognosability of these indicators. These health indicators are further used to predict the RUL of the aircraft system using a similarity-based matching approach. We illustrate our approach for turbofan engines. We show that the consideration of the operating conditions improves the monotonicity of the health indicators by 97\%. Also, our approach leads to accurate RUL estimates with a Root Mean Square Error (RMSE) of 2.67 flights only. Moreover, a 19\% reduction in the RMSE is obtained using our approach in comparison to existing supervised learning models.},
	urldate = {2023-09-28},
	journal = {Engineering Applications of Artificial Intelligence},
	author = {de Pater, Ingeborg and Mitici, Mihaela},
	month = jan,
	year = {2023},
	keywords = {Attention, Autoencoder, Health indicators, Remaining Useful Life prognostics, Unlabelled data samples, Varying operating conditions},
	pages = {105582},
}

Most Remaining Useful Life (RUL) prognostics are obtained using supervised learning models trained with many labelled data samples (i.e., the true RUL is known). In aviation, however, aircraft systems are often preventively replaced before failure. There are thus very few labelled data samples available. We therefore propose a Long Short-Term Memory (LSTM) autoencoder with attention to develop health indicators for an aircraft system instead. This autoencoder is trained with unlabelled data samples (i.e., the true RUL is unknown). Since aircraft fly under various operating conditions (varying altitude, speed, etc.), these conditions are also integrated in the autoencoder. We show that the consideration of the operating conditions leads to robust health indicators and improves significantly the monotonicity, trendability and prognosability of these indicators. These health indicators are further used to predict the RUL of the aircraft system using a similarity-based matching approach. We illustrate our approach for turbofan engines. We show that the consideration of the operating conditions improves the monotonicity of the health indicators by 97%. Also, our approach leads to accurate RUL estimates with a Root Mean Square Error (RMSE) of 2.67 flights only. Moreover, a 19% reduction in the RMSE is obtained using our approach in comparison to existing supervised learning models.

@phdthesis{nguyen_online_2023,
	type = {phdthesis},
	title = {Online machine learning-based predictive maintenance for the railway industry},
	url = {https://theses.hal.science/tel-04164338},
	abstract = {Being an effective long-distance mass transit, the railway will continue to flourish for its limited carbon footprint in the environment. Ensuring the equipment's reliability and passenger safety brings forth the need for efficient maintenance. Apart from the prevalence of corrective and periodic maintenance, predictive maintenance has come into prominence lately. Recent advances in machine learning and the abundance of data drive practitioners to data-driven predictive maintenance. The common practice is to collect data to train a machine learning model, then deploy the model for production and keep it unchanged afterward. We argue that such practice is suboptimal on a data stream. The unboundedness of the stream makes the model prone to incomplete learning. Dynamic changes on the stream introduce novel concepts unseen by the model and decrease its accuracy. The velocity of the stream makes manual labeling infeasible and disables supervised learning algorithms. Therefore, switching from a static, offline learning paradigm to an adaptive, online one is necessary, especially when new generations of connected trains continuously generating sensor data have already been a reality. We investigate the applicability of online machine learning for predictive maintenance on typical complex systems in the railway. First, we develop InterCE as an active learning-based framework that extracts cycles from an unlabeled stream by interacting with a human expert. Then, we implement a long short-term memory autoencoder to transform the extracted cycles into feature vectors that are more compact yet remain representative. Finally, we design CheMoc as a framework that continuously monitors the condition of the systems using online adaptive clustering. Our methods are evaluated on the passenger access systems on two fleets of passenger trains managed by the national railway company SNCF of France.},
	language = {en},
	urldate = {2023-09-15},
	school = {Institut Polytechnique de Paris},
	author = {Nguyen, Minh Huong Le},
	month = jun,
	year = {2023},
}

Being an effective long-distance mass transit, the railway will continue to flourish for its limited carbon footprint in the environment. Ensuring the equipment's reliability and passenger safety brings forth the need for efficient maintenance. Apart from the prevalence of corrective and periodic maintenance, predictive maintenance has come into prominence lately. Recent advances in machine learning and the abundance of data drive practitioners to data-driven predictive maintenance. The common practice is to collect data to train a machine learning model, then deploy the model for production and keep it unchanged afterward. We argue that such practice is suboptimal on a data stream. The unboundedness of the stream makes the model prone to incomplete learning. Dynamic changes on the stream introduce novel concepts unseen by the model and decrease its accuracy. The velocity of the stream makes manual labeling infeasible and disables supervised learning algorithms. Therefore, switching from a static, offline learning paradigm to an adaptive, online one is necessary, especially when new generations of connected trains continuously generating sensor data have already been a reality. We investigate the applicability of online machine learning for predictive maintenance on typical complex systems in the railway. First, we develop InterCE as an active learning-based framework that extracts cycles from an unlabeled stream by interacting with a human expert. Then, we implement a long short-term memory autoencoder to transform the extracted cycles into feature vectors that are more compact yet remain representative. Finally, we design CheMoc as a framework that continuously monitors the condition of the systems using online adaptive clustering. Our methods are evaluated on the passenger access systems on two fleets of passenger trains managed by the national railway company SNCF of France.

@inproceedings{huber_physics-informed_2023,
	title = {Physics-{Informed} {Machine} {Learning} for {Predictive} {Maintenance}: {Applied} {Use}-{Cases}},
	shorttitle = {Physics-{Informed} {Machine} {Learning} for {Predictive} {Maintenance}},
	doi = {10.1109/SDS57534.2023.00016},
	abstract = {The combination of physics and engineering information with data-driven methods like machine learning (ML) and deep learning is gaining attention in various research fields. One of the promising practical applications of such hybrid methods is for supporting maintenance decision making in the form of condition-based and predictive maintenance. In this paper we focus on the potential of physics-informed data augmentation for ML algorithms. We demonstrate possible implementations of the concept using three use cases, differing in their technical systems, their algorithms and their tasks ranging from anomaly detection, through fault diagnostics up to prognostics of the remaining useful life. We elaborate on the benefits and prerequisites of each technique and provide guidelines for future practical implementations in other systems.},
	booktitle = {2023 10th {IEEE} {Swiss} {Conference} on {Data} {Science} ({SDS})},
	author = {Huber, Lilach Goren and Palmé, Thomas and Chao, Manuel Arias},
	month = jun,
	year = {2023},
	note = {ISSN: 2835-3420},
	keywords = {Anomaly Detection, ConditionBased Maintenance, Data science, Decision making, Deep Learning., Deep learning, Distance measurement, Fault Diagnostics, Fault Prognostics, Machine learning algorithms, Prediction algorithms, Predictive Maintenance, Task analysis, physics-informed Machine Learning},
	pages = {66--72},
}

The combination of physics and engineering information with data-driven methods like machine learning (ML) and deep learning is gaining attention in various research fields. One of the promising practical applications of such hybrid methods is for supporting maintenance decision making in the form of condition-based and predictive maintenance. In this paper we focus on the potential of physics-informed data augmentation for ML algorithms. We demonstrate possible implementations of the concept using three use cases, differing in their technical systems, their algorithms and their tasks ranging from anomaly detection, through fault diagnostics up to prognostics of the remaining useful life. We elaborate on the benefits and prerequisites of each technique and provide guidelines for future practical implementations in other systems.

@misc{pashami_explainable_2023,
	title = {Explainable {Predictive} {Maintenance}},
	url = {http://arxiv.org/abs/2306.05120},
	doi = {10.48550/arXiv.2306.05120},
	abstract = {Explainable Artificial Intelligence (XAI) fills the role of a critical interface fostering interactions between sophisticated intelligent systems and diverse individuals, including data scientists, domain experts, end-users, and more. It aids in deciphering the intricate internal mechanisms of ``black box'' Machine Learning (ML), rendering the reasons behind their decisions more understandable. However, current research in XAI primarily focuses on two aspects; ways to facilitate user trust, or to debug and refine the ML model. The majority of it falls short of recognising the diverse types of explanations needed in broader contexts, as different users and varied application areas necessitate solutions tailored to their specific needs. One such domain is Predictive Maintenance (PdM), an exploding area of research under the Industry 4.0 {\textbackslash}\& 5.0 umbrella. This position paper highlights the gap between existing XAI methodologies and the specific requirements for explanations within industrial applications, particularly the Predictive Maintenance field. Despite explainability's crucial role, this subject remains a relatively under-explored area, making this paper a pioneering attempt to bring relevant challenges to the research community's attention. We provide an overview of predictive maintenance tasks and accentuate the need and varying purposes for corresponding explanations. We then list and describe XAI techniques commonly employed in the literature, discussing their suitability for PdM tasks. Finally, to make the ideas and claims more concrete, we demonstrate XAI applied in four specific industrial use cases: commercial vehicles, metro trains, steel plants, and wind farms, spotlighting areas requiring further research.},
	urldate = {2023-08-30},
	publisher = {arXiv},
	author = {Pashami, Sepideh and Nowaczyk, Slawomir and Fan, Yuantao and Jakubowski, Jakub and Paiva, Nuno and Davari, Narjes and Bobek, Szymon and Jamshidi, Samaneh and Sarmadi, Hamid and Alabdallah, Abdallah and Ribeiro, Rita P. and Veloso, Bruno and Sayed-Mouchaweh, Moamar and Rajaoarisoa, Lala and Nalepa, Grzegorz J. and Gama, João},
	month = jun,
	year = {2023},
	note = {arXiv:2306.05120 [cs]},
	keywords = {Computer Science - Artificial Intelligence, I.2.1},
}

Explainable Artificial Intelligence (XAI) fills the role of a critical interface fostering interactions between sophisticated intelligent systems and diverse individuals, including data scientists, domain experts, end-users, and more. It aids in deciphering the intricate internal mechanisms of ``black box'' Machine Learning (ML), rendering the reasons behind their decisions more understandable. However, current research in XAI primarily focuses on two aspects; ways to facilitate user trust, or to debug and refine the ML model. The majority of it falls short of recognising the diverse types of explanations needed in broader contexts, as different users and varied application areas necessitate solutions tailored to their specific needs. One such domain is Predictive Maintenance (PdM), an exploding area of research under the Industry 4.0 \& 5.0 umbrella. This position paper highlights the gap between existing XAI methodologies and the specific requirements for explanations within industrial applications, particularly the Predictive Maintenance field. Despite explainability's crucial role, this subject remains a relatively under-explored area, making this paper a pioneering attempt to bring relevant challenges to the research community's attention. We provide an overview of predictive maintenance tasks and accentuate the need and varying purposes for corresponding explanations. We then list and describe XAI techniques commonly employed in the literature, discussing their suitability for PdM tasks. Finally, to make the ideas and claims more concrete, we demonstrate XAI applied in four specific industrial use cases: commercial vehicles, metro trains, steel plants, and wind farms, spotlighting areas requiring further research.

@article{mitici_prognostics_2023,
	title = {Prognostics for {Lithium}-ion batteries for electric {Vertical} {Take}-off and {Landing} aircraft using data-driven machine learning},
	volume = {12},
	issn = {2666-5468},
	url = {https://www.sciencedirect.com/science/article/pii/S2666546823000058},
	doi = {10.1016/j.egyai.2023.100233},
	abstract = {The health management of batteries is a key enabler for the adoption of Electric Vertical Take-off and Landing vehicles (eVTOLs). Currently, few studies consider the health management of eVTOL batteries. One distinct characteristic of batteries for eVTOLs is that the discharge rates are significantly larger during take-off and landing, compared with the battery discharge rates needed for automotives. Such discharge protocols are expected to impact the long-run health of batteries. This paper proposes a data-driven machine learning framework to estimate the state-of-health and remaining-useful-lifetime of eVTOL batteries under varying flight conditions and taking into account the entire flight profile of the eVTOLs. Three main features are considered for the assessment of the health of the batteries: charge, discharge and temperature. The importance of these features is also quantified. Considering battery charging before flight, a selection of missions for state-of-health and remaining-useful-lifetime prediction is performed. The results show that indeed, discharge-related features have the highest importance when predicting battery state-of-health and remaining-useful-lifetime. Using several machine learning algorithms, it is shown that the battery state-of-health and remaining-useful-life are well estimated using Random Forest regression and Extreme Gradient Boosting, respectively.},
	language = {en},
	urldate = {2023-06-07},
	journal = {Energy and AI},
	author = {Mitici, Mihaela and Hennink, Birgitte and Pavel, Marilena and Dong, Jianning},
	month = apr,
	year = {2023},
	keywords = {Electric Vertical Take-off and Landing vehicles, Lithium-ion battery, Machine learning, Remaining-useful-life, State-of-health},
	pages = {100233},
}

The health management of batteries is a key enabler for the adoption of Electric Vertical Take-off and Landing vehicles (eVTOLs). Currently, few studies consider the health management of eVTOL batteries. One distinct characteristic of batteries for eVTOLs is that the discharge rates are significantly larger during take-off and landing, compared with the battery discharge rates needed for automotives. Such discharge protocols are expected to impact the long-run health of batteries. This paper proposes a data-driven machine learning framework to estimate the state-of-health and remaining-useful-lifetime of eVTOL batteries under varying flight conditions and taking into account the entire flight profile of the eVTOLs. Three main features are considered for the assessment of the health of the batteries: charge, discharge and temperature. The importance of these features is also quantified. Considering battery charging before flight, a selection of missions for state-of-health and remaining-useful-lifetime prediction is performed. The results show that indeed, discharge-related features have the highest importance when predicting battery state-of-health and remaining-useful-lifetime. Using several machine learning algorithms, it is shown that the battery state-of-health and remaining-useful-life are well estimated using Random Forest regression and Extreme Gradient Boosting, respectively.

@article{de_pater_developing_2023-1,
	title = {Developing health indicators and {RUL} prognostics for systems with few failure instances and varying operating conditions using a {LSTM} autoencoder},
	volume = {117},
	issn = {0952-1976},
	url = {https://www.sciencedirect.com/science/article/pii/S0952197622005723},
	doi = {10.1016/j.engappai.2022.105582},
	abstract = {Most Remaining Useful Life (RUL) prognostics are obtained using supervised learning models trained with many labelled data samples (i.e., the true RUL is known). In aviation, however, aircraft systems are often preventively replaced before failure. There are thus very few labelled data samples available. We therefore propose a Long Short-Term Memory (LSTM) autoencoder with attention to develop health indicators for an aircraft system instead. This autoencoder is trained with unlabelled data samples (i.e., the true RUL is unknown). Since aircraft fly under various operating conditions (varying altitude, speed, etc.), these conditions are also integrated in the autoencoder. We show that the consideration of the operating conditions leads to robust health indicators and improves significantly the monotonicity, trendability and prognosability of these indicators. These health indicators are further used to predict the RUL of the aircraft system using a similarity-based matching approach. We illustrate our approach for turbofan engines. We show that the consideration of the operating conditions improves the monotonicity of the health indicators by 97\%. Also, our approach leads to accurate RUL estimates with a Root Mean Square Error (RMSE) of 2.67 flights only. Moreover, a 19\% reduction in the RMSE is obtained using our approach in comparison to existing supervised learning models.},
	language = {en},
	urldate = {2023-06-07},
	journal = {Engineering Applications of Artificial Intelligence},
	author = {de Pater, Ingeborg and Mitici, Mihaela},
	month = jan,
	year = {2023},
	keywords = {Attention, Autoencoder, Health indicators, Remaining Useful Life prognostics, Unlabelled data samples, Varying operating conditions},
	pages = {105582},
}

Most Remaining Useful Life (RUL) prognostics are obtained using supervised learning models trained with many labelled data samples (i.e., the true RUL is known). In aviation, however, aircraft systems are often preventively replaced before failure. There are thus very few labelled data samples available. We therefore propose a Long Short-Term Memory (LSTM) autoencoder with attention to develop health indicators for an aircraft system instead. This autoencoder is trained with unlabelled data samples (i.e., the true RUL is unknown). Since aircraft fly under various operating conditions (varying altitude, speed, etc.), these conditions are also integrated in the autoencoder. We show that the consideration of the operating conditions leads to robust health indicators and improves significantly the monotonicity, trendability and prognosability of these indicators. These health indicators are further used to predict the RUL of the aircraft system using a similarity-based matching approach. We illustrate our approach for turbofan engines. We show that the consideration of the operating conditions improves the monotonicity of the health indicators by 97%. Also, our approach leads to accurate RUL estimates with a Root Mean Square Error (RMSE) of 2.67 flights only. Moreover, a 19% reduction in the RMSE is obtained using our approach in comparison to existing supervised learning models.

@misc{hurtado_continual_2023,
	title = {Continual {Learning} for {Predictive} {Maintenance}: {Overview} and {Challenges}},
	shorttitle = {Continual {Learning} for {Predictive} {Maintenance}},
	url = {http://arxiv.org/abs/2301.12467},
	doi = {10.48550/arXiv.2301.12467},
	abstract = {Machine learning techniques have become one of the main propellers for solving many engineering problems effectively and efficiently. In Predictive Maintenance, for instance, Data-Driven methods have been used to improve predictions of when maintenance is needed on different machines and operative contexts. However, one of the limitations of these methods is that they are trained on a fixed distribution that does not change over time, which seldom happens in real-world applications. When internal or external factors alter the data distribution, the model performance may decrease or even fail unpredictably, resulting in severe consequences for machine maintenance. Continual Learning methods propose ways of adapting prediction models and incorporating new knowledge after deployment. The main objective of these methods is to avoid the plasticity-stability dilemma by updating the parametric model while not forgetting previously learned tasks. In this work, we present the current state of the art in applying Continual Learning to Predictive Maintenance, with an extensive review of both disciplines. We first introduce the two research themes independently, then discuss the current intersection of Continual Learning and Predictive Maintenance. Finally, we discuss the main research directions and conclusions.},
	urldate = {2023-06-03},
	publisher = {arXiv},
	author = {Hurtado, Julio and Salvati, Dario and Semola, Rudy and Bosio, Mattia and Lomonaco, Vincenzo},
	month = jan,
	year = {2023},
	note = {arXiv:2301.12467 [cs]},
	keywords = {Computer Science - Machine Learning},
}

Machine learning techniques have become one of the main propellers for solving many engineering problems effectively and efficiently. In Predictive Maintenance, for instance, Data-Driven methods have been used to improve predictions of when maintenance is needed on different machines and operative contexts. However, one of the limitations of these methods is that they are trained on a fixed distribution that does not change over time, which seldom happens in real-world applications. When internal or external factors alter the data distribution, the model performance may decrease or even fail unpredictably, resulting in severe consequences for machine maintenance. Continual Learning methods propose ways of adapting prediction models and incorporating new knowledge after deployment. The main objective of these methods is to avoid the plasticity-stability dilemma by updating the parametric model while not forgetting previously learned tasks. In this work, we present the current state of the art in applying Continual Learning to Predictive Maintenance, with an extensive review of both disciplines. We first introduce the two research themes independently, then discuss the current intersection of Continual Learning and Predictive Maintenance. Finally, we discuss the main research directions and conclusions.

@article{mercha_machine_2023,
	title = {Machine learning and deep learning for sentiment analysis across languages: {A} survey},
	volume = {531},
	issn = {0925-2312},
	shorttitle = {Machine learning and deep learning for sentiment analysis across languages},
	url = {https://www.sciencedirect.com/science/article/pii/S0925231223001546},
	doi = {10.1016/j.neucom.2023.02.015},
	abstract = {The inception and rapid growth of the Web, social media, and other online forums have resulted in the continuous and rapid generation of opinionated textual data. Several real-world applications have been focusing on determining the sentiments expressed in these data. Owing to the multilinguistic nature of the generated data, there exists an increasing need to perform sentiment analysis on data in diverse languages. This study presents an overview of the methods used to perform sentiment analysis across languages. We primarily focus on multilingual and cross-lingual approaches. This survey covers the early approaches and current advancements that employ machine learning and deep learning models. We categorize these methods and techniques and provide new research directions. Our findings reveal that deep learning techniques have been widely used in both approaches and yield the best results. Additionally, the scarcity of multilingual annotated datasets limits the progress of multilingual and cross-lingual sentiment analyses, and therefore increases the complexity in comparing these techniques and determining the ones with the best performance.},
	language = {en},
	urldate = {2023-03-21},
	journal = {Neurocomputing},
	author = {Mercha, El Mahdi and Benbrahim, Houda},
	month = apr,
	year = {2023},
	keywords = {Cross-lingual sentiment analysis, Deep learning, Machine learning, Multilingual sentiment analysis, Sentiment analysis},
	pages = {195--216},
}

The inception and rapid growth of the Web, social media, and other online forums have resulted in the continuous and rapid generation of opinionated textual data. Several real-world applications have been focusing on determining the sentiments expressed in these data. Owing to the multilinguistic nature of the generated data, there exists an increasing need to perform sentiment analysis on data in diverse languages. This study presents an overview of the methods used to perform sentiment analysis across languages. We primarily focus on multilingual and cross-lingual approaches. This survey covers the early approaches and current advancements that employ machine learning and deep learning models. We categorize these methods and techniques and provide new research directions. Our findings reveal that deep learning techniques have been widely used in both approaches and yield the best results. Additionally, the scarcity of multilingual annotated datasets limits the progress of multilingual and cross-lingual sentiment analyses, and therefore increases the complexity in comparing these techniques and determining the ones with the best performance.

@article{nanath_investigation_2023,
	title = {An investigation of crowdsourcing methods in enhancing the machine learning approach for detecting online recruitment fraud},
	volume = {3},
	issn = {2667-0968},
	url = {https://www.sciencedirect.com/science/article/pii/S2667096823000149},
	doi = {10.1016/j.jjimei.2023.100167},
	abstract = {Misinformation on the web has become a problem of significant impact in an information-driven society. Persistent and large volumes of fake content are being injected, and hence the content (news, articles, jobs, facts) available online is often questionable. This research reviews a range of machine learning algorithms to tackle a specific case of online recruitment fraud (ORF). A model with content features of job posting is tested with five supervised machine learning (ML) algorithms. It then investigates various crowdsourcing techniques that could enhance prediction accuracy and add human insights to machine learning automation. Each crowdsourcing method (explored as human signals online) was tested across the same ML algorithms to test its effectiveness in predicting fake job postings. The testing was conducted by comparing the hybrid models of machine learning and crowdsourced inputs. This study revealed that the best ML algorithm was different in the automated model compared to the hybrid model. Results also indicated that the net promoter type crowdsourced question resulted in the best accuracy in classifying fraudulent and legitimate jobs. The decision tree and generalized linear model demonstrated the highest accuracy among all the tested models.},
	language = {en},
	number = {1},
	urldate = {2023-03-08},
	journal = {International Journal of Information Management Data Insights},
	author = {Nanath, Krishnadas and Olney, Liting},
	month = apr,
	year = {2023},
	keywords = {Crowdsourcing, Fake content, Machine learning, Misinformation, Online recruitment fraud},
	pages = {100167},
}

Misinformation on the web has become a problem of significant impact in an information-driven society. Persistent and large volumes of fake content are being injected, and hence the content (news, articles, jobs, facts) available online is often questionable. This research reviews a range of machine learning algorithms to tackle a specific case of online recruitment fraud (ORF). A model with content features of job posting is tested with five supervised machine learning (ML) algorithms. It then investigates various crowdsourcing techniques that could enhance prediction accuracy and add human insights to machine learning automation. Each crowdsourcing method (explored as human signals online) was tested across the same ML algorithms to test its effectiveness in predicting fake job postings. The testing was conducted by comparing the hybrid models of machine learning and crowdsourced inputs. This study revealed that the best ML algorithm was different in the automated model compared to the hybrid model. Results also indicated that the net promoter type crowdsourced question resulted in the best accuracy in classifying fraudulent and legitimate jobs. The decision tree and generalized linear model demonstrated the highest accuracy among all the tested models.

@article{yao_deep_2023,
	title = {Deep adaptive arbitrary polynomial chaos expansion: {A} mini-data-driven semi-supervised method for uncertainty quantification},
	volume = {229},
	issn = {0951-8320},
	shorttitle = {Deep adaptive arbitrary polynomial chaos expansion},
	url = {https://www.sciencedirect.com/science/article/pii/S095183202200432X},
	doi = {10.1016/j.ress.2022.108813},
	abstract = {All kinds of uncertainties influence the reliability of the engineering system. Thus, uncertainty quantification is significant to the system reliability analysis. Polynomial chaos expansion (PCE) is an effective method for uncertainty quantification while it requires sufficient labeled data to quantify uncertainty accurately. To overcome this problem, this paper proposes the adaptive arbitrary polynomial chaos (aPC) and proves two properties of the adaptive expansion coefficients. Sequentially, a semi-supervised deep adaptive arbitrary polynomial chaos expansion (Deep aPCE) method is proposed based on the adaptive aPC and the deep neural network (DNN). The Deep aPCE method uses two properties of the adaptive aPC to assist in training the DNN by a small amount of labeled data and abundant unlabeled data, significantly reducing the training data cost. On the other hand, the Deep aPCE method adopts the DNN to fine-tune the adaptive expansion coefficients dynamically, improving the accuracy of uncertainty quantification. Besides, the Deep aPCE method can directly construct accurate surrogate models of the high dimensional stochastic systems. Five numerical examples are used to verify the effectiveness of the Deep aPCE method. Finally, the Deep aPCE method is applied to the reliability analysis of an axisymmetric conical aircraft.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Yao, Wen and Zheng, Xiaohu and Zhang, Jun and Wang, Ning and Tang, Guijian},
	month = jan,
	year = {2023},
	keywords = {Arbitrary polynomial chaos expansion, Deep learning, Mini-data, Semi-supervised, Uncertainty quantification},
	pages = {108813},
}

All kinds of uncertainties influence the reliability of the engineering system. Thus, uncertainty quantification is significant to the system reliability analysis. Polynomial chaos expansion (PCE) is an effective method for uncertainty quantification while it requires sufficient labeled data to quantify uncertainty accurately. To overcome this problem, this paper proposes the adaptive arbitrary polynomial chaos (aPC) and proves two properties of the adaptive expansion coefficients. Sequentially, a semi-supervised deep adaptive arbitrary polynomial chaos expansion (Deep aPCE) method is proposed based on the adaptive aPC and the deep neural network (DNN). The Deep aPCE method uses two properties of the adaptive aPC to assist in training the DNN by a small amount of labeled data and abundant unlabeled data, significantly reducing the training data cost. On the other hand, the Deep aPCE method adopts the DNN to fine-tune the adaptive expansion coefficients dynamically, improving the accuracy of uncertainty quantification. Besides, the Deep aPCE method can directly construct accurate surrogate models of the high dimensional stochastic systems. Five numerical examples are used to verify the effectiveness of the Deep aPCE method. Finally, the Deep aPCE method is applied to the reliability analysis of an axisymmetric conical aircraft.

@article{tohme_reliable_2023,
	title = {Reliable neural networks for regression uncertainty estimation},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004306},
	doi = {10.1016/j.ress.2022.108811},
	abstract = {While deep neural networks are highly performant and successful in a wide range of real-world problems, estimating their predictive uncertainty remains a challenging task. To address this challenge, we propose and implement a loss function for regression uncertainty estimation based on the Bayesian Validation Metric (BVM) framework while using ensemble learning. The proposed loss reproduces maximum likelihood estimation in the limiting case. A series of experiments on in-distribution data show that the proposed method is competitive with existing state-of-the-art methods. Experiments on out-of-distribution data show that the proposed method is robust to statistical change and exhibits superior predictive capability.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Tohme, Tony and Vanslette, Kevin and Youcef-Toumi, Kamal},
	month = jan,
	year = {2023},
	keywords = {Neural networks, Predictive uncertainty estimation, Regression, Reliability},
	pages = {108811},
}

While deep neural networks are highly performant and successful in a wide range of real-world problems, estimating their predictive uncertainty remains a challenging task. To address this challenge, we propose and implement a loss function for regression uncertainty estimation based on the Bayesian Validation Metric (BVM) framework while using ensemble learning. The proposed loss reproduces maximum likelihood estimation in the limiting case. A series of experiments on in-distribution data show that the proposed method is competitive with existing state-of-the-art methods. Experiments on out-of-distribution data show that the proposed method is robust to statistical change and exhibits superior predictive capability.

@article{shangguan_train_2023,
	title = {Train wheel degradation generation and prediction based on the time series generation adversarial network},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004355},
	doi = {10.1016/j.ress.2022.108816},
	abstract = {To ensure the safe operation of high-speed railways, it is necessary to assess the reliability of its key components. Among them, as wheels are prone to wear degradation and the wear data acquisition process has the disadvantages of high cost and long cycle. There are few wheels degradation samples, which in turn makes the wheel degradation prediction have large errors. Hence, this paper uses the time series generator adversarial network (TimeGAN) to generate synthetic wheel degradation, in which the original data is segmented through a sliding window to obtain more input sets, and the noise distribution in the generator network is combined with the stationary gamma process (SGP). Then, the wheel degradation at measured distance k is predicted by the Gated Recurrent Unit (GRU) network. To evaluate the effectiveness of the proposed method, different methods in this paper are conducted for the experiment comparison. The experiment result shows that the proposed method has a better effect on the generation of train wheel degradation, and the Kullback-Leibler (KL) divergence and the prediction error are the smallest in the comparison. Hence, the proposed method can provide support for the further reliability analysis of railways and further ensure their operational safety.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Shangguan, Anqi and Xie, Guo and Fei, Rong and Mu, Lingxia and Hei, Xinhong},
	month = jan,
	year = {2023},
	keywords = {Data generation, Degradation analysis, Rail safety, Train wheel},
	pages = {108816},
}

To ensure the safe operation of high-speed railways, it is necessary to assess the reliability of its key components. Among them, as wheels are prone to wear degradation and the wear data acquisition process has the disadvantages of high cost and long cycle. There are few wheels degradation samples, which in turn makes the wheel degradation prediction have large errors. Hence, this paper uses the time series generator adversarial network (TimeGAN) to generate synthetic wheel degradation, in which the original data is segmented through a sliding window to obtain more input sets, and the noise distribution in the generator network is combined with the stationary gamma process (SGP). Then, the wheel degradation at measured distance k is predicted by the Gated Recurrent Unit (GRU) network. To evaluate the effectiveness of the proposed method, different methods in this paper are conducted for the experiment comparison. The experiment result shows that the proposed method has a better effect on the generation of train wheel degradation, and the Kullback-Leibler (KL) divergence and the prediction error are the smallest in the comparison. Hence, the proposed method can provide support for the further reliability analysis of railways and further ensure their operational safety.

@article{ma_multiple_2023,
	title = {Multiple health indicators fusion-based health prognostic for lithium-ion battery using transfer learning and hybrid deep learning method},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004379},
	doi = {10.1016/j.ress.2022.108818},
	abstract = {Accurate state of health (SOH) estimation of lithium-ion battery provides a guarantee for the safe driving of electric vehicles. Most SOH estimation methods based on the machine learning assume that the training and testing data follow the uniform distribution. However, the distribution of the datasets obtained at the different working conditions has discrepancy, which also increases its inherently large computational burden. Therefore, a novel SOH estimation method based on multiple health indicators (HIs) fusion using transfer learning and deep belief network (DBN)-long short-term memory (LSTM) hybrid network is proposed. Transfer learning is used to learn the shared features in the source domain and the target domain. Then, aiming at the insufficiency of shallow network in mining data features, DBN is utilized for SOH estimation. And considering the influence of historical information on future prediction, LSTM cell is used to replace the traditional BP neural network structure. Comparative study is conducted by applying deep and shallow network on the measured data for monitoring SOH of the battery in applications. The experimental results show that the method proposed in this paper is effective, and the performance of knowledge transferring under single domain and cross domain is also verified.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Ma, Yan and Shan, Ce and Gao, Jinwu and Chen, Hong},
	month = jan,
	year = {2023},
	keywords = {Deep belief network, Lithium-ion batteries, Long short term memory, Multiple health indicators, Transfer learning},
	pages = {108818},
}

Accurate state of health (SOH) estimation of lithium-ion battery provides a guarantee for the safe driving of electric vehicles. Most SOH estimation methods based on the machine learning assume that the training and testing data follow the uniform distribution. However, the distribution of the datasets obtained at the different working conditions has discrepancy, which also increases its inherently large computational burden. Therefore, a novel SOH estimation method based on multiple health indicators (HIs) fusion using transfer learning and deep belief network (DBN)-long short-term memory (LSTM) hybrid network is proposed. Transfer learning is used to learn the shared features in the source domain and the target domain. Then, aiming at the insufficiency of shallow network in mining data features, DBN is utilized for SOH estimation. And considering the influence of historical information on future prediction, LSTM cell is used to replace the traditional BP neural network structure. Comparative study is conducted by applying deep and shallow network on the measured data for monitoring SOH of the battery in applications. The experimental results show that the method proposed in this paper is effective, and the performance of knowledge transferring under single domain and cross domain is also verified.

@article{zhou_generic_2023,
	title = {A generic physics-informed neural network-based framework for reliability assessment of multi-state systems},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004537},
	doi = {10.1016/j.ress.2022.108835},
	abstract = {In this paper, we develop a generic physics-informed neural network (PINN)-based framework to assess the reliability of multi-state systems (MSSs). The proposed framework follows a two-step procedure. In the first step, we recast the reliability assessment of MSS as a machine learning problem using the framework of PINN. A feedforward neural network with two individual loss groups is constructed to encode the initial condition and the state transitions governed by ordinary differential equations in MSS, respectively. Next, we tackle the problem of high imbalance in the magnitudes of back-propagated gradients from a multi-task learning perspective and establish a continuous latent function for system reliability assessment. Particularly, we regard each element of the loss function as an individual learning task and project a task’s gradient onto the norm plane of any other task with a conflicting gradient by taking the projecting conflicting gradients (PCGrad) method. We demonstrate the applications of the proposed framework for MSS reliability assessment in a variety of scenarios, including time-independent or dependent state transitions, where system scales increase from small to medium. The computational results indicate that PINN-based framework reveals a promising performance in MSS reliability assessment and incorporation of PCGrad into PINN substantially improves the solution quality and convergence speed of the algorithm.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhou, Taotao and Zhang, Xiaoge and Droguett, Enrique Lopez and Mosleh, Ali},
	month = jan,
	year = {2023},
	keywords = {Gradient projection, Markov process, Multi-state systems, Physics-informed neural network, Reliability assessment},
	pages = {108835},
}

In this paper, we develop a generic physics-informed neural network (PINN)-based framework to assess the reliability of multi-state systems (MSSs). The proposed framework follows a two-step procedure. In the first step, we recast the reliability assessment of MSS as a machine learning problem using the framework of PINN. A feedforward neural network with two individual loss groups is constructed to encode the initial condition and the state transitions governed by ordinary differential equations in MSS, respectively. Next, we tackle the problem of high imbalance in the magnitudes of back-propagated gradients from a multi-task learning perspective and establish a continuous latent function for system reliability assessment. Particularly, we regard each element of the loss function as an individual learning task and project a task’s gradient onto the norm plane of any other task with a conflicting gradient by taking the projecting conflicting gradients (PCGrad) method. We demonstrate the applications of the proposed framework for MSS reliability assessment in a variety of scenarios, including time-independent or dependent state transitions, where system scales increase from small to medium. The computational results indicate that PINN-based framework reveals a promising performance in MSS reliability assessment and incorporation of PCGrad into PINN substantially improves the solution quality and convergence speed of the algorithm.

@article{zhou_availability_2023,
	title = {Availability analysis of shared bikes using abnormal trip data},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004616},
	doi = {10.1016/j.ress.2022.108844},
	abstract = {The users’ cancelling rental data in the bike-sharing system (BSS) is usually regarded as abnormal trip data and is ignored. Abnormal trip data may have implicit information about the availability of shared bikes. So this paper presents an approach based on functional principal components analysis (FPCA) and clustering to advance the shared-bike availability analysis and maintenance strategy optimization using the abnormal trip data. In the proposed approach, the ratio of the cancelling rental number to the total rental number is scored as an index. Their values reflect a smooth variation in availability. The FPCA method is performed to explore the long-term availability variation modes of shared bikes. Then the dominant modes of availability variations are determined using the k-means algorithm. The effectiveness of the proposed approach is illustrated on the real-world trip data of a BSS. The analysis result indicates that the long-term availability level of the referred BSS has decreased from the initial 0.907 to 0.861. In the definite availability variation modes, the availability of one of the variation modes even has decreased to 0.709. Finally, the preventive maintenance model is presented to prevent the deterioration or availability decrease of shared bikes based on the mean functions of availability variation modes.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhou, Yu and Kou, Gang and Guo, Zhen-Zhu and Xiao, Hui},
	month = jan,
	year = {2023},
	keywords = {Abnormal trip data, Availability, Clustering, Functional principal component analysis, Preventive maintenance, Shared bike},
	pages = {108844},
}

The users’ cancelling rental data in the bike-sharing system (BSS) is usually regarded as abnormal trip data and is ignored. Abnormal trip data may have implicit information about the availability of shared bikes. So this paper presents an approach based on functional principal components analysis (FPCA) and clustering to advance the shared-bike availability analysis and maintenance strategy optimization using the abnormal trip data. In the proposed approach, the ratio of the cancelling rental number to the total rental number is scored as an index. Their values reflect a smooth variation in availability. The FPCA method is performed to explore the long-term availability variation modes of shared bikes. Then the dominant modes of availability variations are determined using the k-means algorithm. The effectiveness of the proposed approach is illustrated on the real-world trip data of a BSS. The analysis result indicates that the long-term availability level of the referred BSS has decreased from the initial 0.907 to 0.861. In the definite availability variation modes, the availability of one of the variation modes even has decreased to 0.709. Finally, the preventive maintenance model is presented to prevent the deterioration or availability decrease of shared bikes based on the mean functions of availability variation modes.

@article{yang_data_2023,
	title = {Data {Regeneration} {Based} on {Multiple} {Degradation} {Processes} for {Remaining} {Useful} {Life} {Estimation}},
	volume = {229},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022004847},
	doi = {10.1016/j.ress.2022.108867},
	abstract = {Remaining useful life prediction based on deep learning for critical components demands sufficient and varied degradation samples. However, the field acquisition or laboratory preparation is generally cumbersome or the samples obtained are stereotyped. The paper proposes a data regeneration method based on multiple degradation processes to deal with the dilemma, which consists of three parts: state identification, regeneration rules from run to failure and state databases. In the first part, a global gain index and a local gain index are proposed to identify the different states of components. In the second part, an identical transformation method, a probability distribution of degradation states and data regeneration criteria are proposed to serve regeneration process of samples from run to failure. In the third part, an augmentation framework based on conditional generative adversarial networks is proposed to enrich the samples of the state database, which makes state samples more diverse. The practicability of regenerated samples obtained by the proposed method was verified by two experiments. In each experiment, initial samples, regenerated samples and hybrid samples were established respectively. Experiments with different training samples based on the same network were carried out to verify the effectiveness of the regenerated samples.},
	language = {en},
	urldate = {2022-10-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Yang, Ningning and Wang, Zhijian and Cai, Wenan and Li, Yanfeng},
	month = jan,
	year = {2023},
	keywords = {Data regeneration, Deep learning, Regeneration rules, Remaining useful life, State identification},
	pages = {108867},
}

Remaining useful life prediction based on deep learning for critical components demands sufficient and varied degradation samples. However, the field acquisition or laboratory preparation is generally cumbersome or the samples obtained are stereotyped. The paper proposes a data regeneration method based on multiple degradation processes to deal with the dilemma, which consists of three parts: state identification, regeneration rules from run to failure and state databases. In the first part, a global gain index and a local gain index are proposed to identify the different states of components. In the second part, an identical transformation method, a probability distribution of degradation states and data regeneration criteria are proposed to serve regeneration process of samples from run to failure. In the third part, an augmentation framework based on conditional generative adversarial networks is proposed to enrich the samples of the state database, which makes state samples more diverse. The practicability of regenerated samples obtained by the proposed method was verified by two experiments. In each experiment, initial samples, regenerated samples and hybrid samples were established respectively. Experiments with different training samples based on the same network were carried out to verify the effectiveness of the regenerated samples.

@article{xia_toward_2022,
	title = {Toward cognitive predictive maintenance: {A} survey of graph-based approaches},
	volume = {64},
	issn = {0278-6125},
	shorttitle = {Toward cognitive predictive maintenance},
	url = {https://www.sciencedirect.com/science/article/pii/S0278612522000978},
	doi = {10.1016/j.jmsy.2022.06.002},
	abstract = {Predictive Maintenance (PdM) has continually attracted interest from the manufacturing community due to its significant potential in reducing unexpected machine downtime and related cost. Much attention to existing PdM research has been paid to perceiving the fault, while the identification and estimation processes are affected by many factors. Many existing approaches have not been able to manage the existing knowledge effectively for reasoning the causal relationship of fault. Meanwhile, complete correlation analysis of identified faults and the corresponding root causes is often missing. To address this problem, graph-based approaches (GbA) with cognitive intelligence are proposed, because the GbA are superior in semantic causal inference, heterogeneous association, and visualized explanation. In addition, GbA can achieve promising performance on PdM’s perception tasks by revealing the dependency relationship among parts/components of the equipment. However, despite its advantages, few papers discuss cognitive inference in PdM, let alone GbA. Aiming to fill this gap, this paper concentrates on GbA, and carries out a comprehensive survey organized by the sequential stages in PdM, i.e., anomaly detection, diagnosis, prognosis, and maintenance decision-making. Firstly, GbA and their corresponding graph construction methods are introduced. Secondly, the implementation strategies and instances of GbA in PdM are presented. Finally, challenges and future works toward cognitive PdM are proposed. It is hoped that this work can provide a fundamental basis for researchers and industrial practitioners in adopting GbA-based PdM, and initiate several future research directions to achieve the cognitive PdM.},
	urldate = {2023-10-26},
	journal = {Journal of Manufacturing Systems},
	author = {Xia, Liqiao and Zheng, Pai and Li, Xinyu and Gao, Robert. X. and Wang, Lihui},
	month = jul,
	year = {2022},
	keywords = {Bayesian network, Cognitive computing, Graph neural network, Knowledge graph, Predictive maintenance},
	pages = {107--120},
}

Predictive Maintenance (PdM) has continually attracted interest from the manufacturing community due to its significant potential in reducing unexpected machine downtime and related cost. Much attention to existing PdM research has been paid to perceiving the fault, while the identification and estimation processes are affected by many factors. Many existing approaches have not been able to manage the existing knowledge effectively for reasoning the causal relationship of fault. Meanwhile, complete correlation analysis of identified faults and the corresponding root causes is often missing. To address this problem, graph-based approaches (GbA) with cognitive intelligence are proposed, because the GbA are superior in semantic causal inference, heterogeneous association, and visualized explanation. In addition, GbA can achieve promising performance on PdM’s perception tasks by revealing the dependency relationship among parts/components of the equipment. However, despite its advantages, few papers discuss cognitive inference in PdM, let alone GbA. Aiming to fill this gap, this paper concentrates on GbA, and carries out a comprehensive survey organized by the sequential stages in PdM, i.e., anomaly detection, diagnosis, prognosis, and maintenance decision-making. Firstly, GbA and their corresponding graph construction methods are introduced. Secondly, the implementation strategies and instances of GbA in PdM are presented. Finally, challenges and future works toward cognitive PdM are proposed. It is hoped that this work can provide a fundamental basis for researchers and industrial practitioners in adopting GbA-based PdM, and initiate several future research directions to achieve the cognitive PdM.

@article{veloso_metropt_2022,
	title = {The {MetroPT} dataset for predictive maintenance},
	volume = {9},
	copyright = {2022 The Author(s)},
	issn = {2052-4463},
	url = {https://www.nature.com/articles/s41597-022-01877-3},
	doi = {10.1038/s41597-022-01877-3},
	abstract = {The paper describes the MetroPT data set, an outcome of a Predictive Maintenance project with an urban metro public transportation service in Porto, Portugal. The data was collected in 2022 to develop machine learning methods for online anomaly detection and failure prediction. Several analog sensor signals (pressure, temperature, current consumption), digital signals (control signals, discrete signals), and GPS information (latitude, longitude, and speed) provide a framework that can be easily used and help the development of new machine learning methods. This dataset contains some interesting characteristics and can be a good benchmark for predictive maintenance models.},
	language = {en},
	number = {1},
	urldate = {2023-10-09},
	journal = {Scientific Data},
	author = {Veloso, Bruno and Ribeiro, Rita P. and Gama, João and Pereira, Pedro Mota},
	month = dec,
	year = {2022},
	keywords = {Computer science, Scientific data},
	pages = {764},
}

The paper describes the MetroPT data set, an outcome of a Predictive Maintenance project with an urban metro public transportation service in Porto, Portugal. The data was collected in 2022 to develop machine learning methods for online anomaly detection and failure prediction. Several analog sensor signals (pressure, temperature, current consumption), digital signals (control signals, discrete signals), and GPS information (latitude, longitude, and speed) provide a framework that can be easily used and help the development of new machine learning methods. This dataset contains some interesting characteristics and can be a good benchmark for predictive maintenance models.

@article{dangut_application_2022,
	title = {Application of deep reinforcement learning for extremely rare failure prediction in aircraft maintenance},
	volume = {171},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327022000693},
	doi = {10.1016/j.ymssp.2022.108873},
	abstract = {The use of aircraft operational logs to predict potential failure that may lead to disruption poses many challenges and has yet to be fully explored. Given that aircraft are high-integrity assets, failures are extremely rare, and hence the distribution of relevant log data containing prior indicators will be highly skewed to the normal (healthy) case. This will present a significant challenge in using data-driven techniques because the model will be biased to the heavily weighted no-fault outcomes. This paper presents a novel approach for predicting unscheduled aircraft maintenance action based on deep reinforcement learning techniques using aircraft central maintenance system logs. The algorithm transforms the rare failure prediction problem into a sequential decision-making process that is optimised using a reward system that penalises proposed predictions that result in a false diagnosis and preferentially favours predictions that result in the right diagnosis. The validation data is directly associated with the physical health aspects of the aircraft components. The influence of extremely rare failure prediction on the proposed method is analysed. The effectiveness of the new approach is verified by comparison with previous studies, cost-sensitive and oversampling methods. Performance was evaluated based on G-mean and false-positives rates. The proposed approach shows the superior performance of 20.3\% improvement in G-mean and 97\% reduction in false-positive rate.},
	language = {en},
	urldate = {2022-03-03},
	journal = {Mechanical Systems and Signal Processing},
	author = {Dangut, Maren David and Jennions, Ian K. and King, Steve and Skaf, Zakwan},
	month = may,
	year = {2022},
	keywords = {Aircraft maintenance, Deep reinforcement learning, Extremely rare event, Imbalance classification},
	pages = {108873},
}

The use of aircraft operational logs to predict potential failure that may lead to disruption poses many challenges and has yet to be fully explored. Given that aircraft are high-integrity assets, failures are extremely rare, and hence the distribution of relevant log data containing prior indicators will be highly skewed to the normal (healthy) case. This will present a significant challenge in using data-driven techniques because the model will be biased to the heavily weighted no-fault outcomes. This paper presents a novel approach for predicting unscheduled aircraft maintenance action based on deep reinforcement learning techniques using aircraft central maintenance system logs. The algorithm transforms the rare failure prediction problem into a sequential decision-making process that is optimised using a reward system that penalises proposed predictions that result in a false diagnosis and preferentially favours predictions that result in the right diagnosis. The validation data is directly associated with the physical health aspects of the aircraft components. The influence of extremely rare failure prediction on the proposed method is analysed. The effectiveness of the new approach is verified by comparison with previous studies, cost-sensitive and oversampling methods. Performance was evaluated based on G-mean and false-positives rates. The proposed approach shows the superior performance of 20.3% improvement in G-mean and 97% reduction in false-positive rate.

@inproceedings{ye_continual_2022,
	address = {Cham},
	series = {Lecture {Notes} in {Computer} {Science}},
	title = {Continual {Variational} {Autoencoder} {Learning} via {Online} {Cooperative} {Memorization}},
	isbn = {978-3-031-20050-2},
	doi = {10.1007/978-3-031-20050-2_31},
	abstract = {Due to their inference, data representation and reconstruction properties, Variational Autoencoders (VAE) have been successfully used in continual learning classification tasks. However, their ability to generate images with specifications corresponding to the classes and databases learned during Continual Learning (CL) is not well understood and catastrophic forgetting remains a significant challenge. In this paper, we firstly analyze the forgetting behaviour of VAEs by developing a new theoretical framework that formulates CL as a dynamic optimal transport problem. This framework proves approximate bounds to the data likelihood without requiring the task information and explains how the prior knowledge is lost during the training process. We then propose a novel memory buffering approach, namely the Online Cooperative Memorization (OCM) framework, which consists of a Short-Term Memory (STM) that continually stores recent samples to provide future information for the model, and a Long-Term Memory (LTM) aiming to preserve a wide diversity of samples. The proposed OCM transfers certain samples from STM to LTM according to the information diversity selection criterion without requiring any supervised signals. The OCM framework is then combined with a dynamic VAE expansion mixture network for further enhancing its performance.},
	language = {en},
	booktitle = {Computer {Vision} – {ECCV} 2022},
	publisher = {Springer Nature Switzerland},
	author = {Ye, Fei and Bors, Adrian G.},
	editor = {Avidan, Shai and Brostow, Gabriel and Cissé, Moustapha and Farinella, Giovanni Maria and Hassner, Tal},
	year = {2022},
	keywords = {Continual learning, Lifelong generative modelling, VAE},
	pages = {531--549},
}

Due to their inference, data representation and reconstruction properties, Variational Autoencoders (VAE) have been successfully used in continual learning classification tasks. However, their ability to generate images with specifications corresponding to the classes and databases learned during Continual Learning (CL) is not well understood and catastrophic forgetting remains a significant challenge. In this paper, we firstly analyze the forgetting behaviour of VAEs by developing a new theoretical framework that formulates CL as a dynamic optimal transport problem. This framework proves approximate bounds to the data likelihood without requiring the task information and explains how the prior knowledge is lost during the training process. We then propose a novel memory buffering approach, namely the Online Cooperative Memorization (OCM) framework, which consists of a Short-Term Memory (STM) that continually stores recent samples to provide future information for the model, and a Long-Term Memory (LTM) aiming to preserve a wide diversity of samples. The proposed OCM transfers certain samples from STM to LTM according to the information diversity selection criterion without requiring any supervised signals. The OCM framework is then combined with a dynamic VAE expansion mixture network for further enhancing its performance.

@inproceedings{davari_fault_2022,
	address = {Cham},
	series = {Lecture {Notes} in {Computer} {Science}},
	title = {A {Fault} {Detection} {Framework} {Based} on {LSTM} {Autoencoder}: {A} {Case} {Study} for {Volvo} {Bus} {Data} {Set}},
	isbn = {978-3-031-01333-1},
	shorttitle = {A {Fault} {Detection} {Framework} {Based} on {LSTM} {Autoencoder}},
	doi = {10.1007/978-3-031-01333-1_4},
	abstract = {This study applies a data-driven anomaly detection framework based on a Long Short-Term Memory (LSTM) autoencoder network for several subsystems of a public transport bus. The proposed framework efficiently detects abnormal data, significantly reducing the false alarm rate compared to available alternatives. Using historical repair records, we demonstrate how detection of abnormal sequences in the signals can be used for predicting equipment failures. The deviations from normal operation patterns are detected by analysing the data collected from several on-board sensors (e.g., wet tank air pressure, engine speed, engine load) installed on the bus. The performance of LSTM autoencoder (LSTM-AE) is compared against the multi-layer autoencoder (mlAE) network in the same anomaly detection framework. The experimental results show that the performance indicators of the LSTM-AE network, in terms of F1 Score, Recall, and Precision, are better than those of the mlAE network.},
	language = {en},
	booktitle = {Advances in {Intelligent} {Data} {Analysis} {XX}},
	publisher = {Springer International Publishing},
	author = {Davari, Narjes and Pashami, Sepideh and Veloso, Bruno and Nowaczyk, Sławomir and Fan, Yuantao and Pereira, Pedro Mota and Ribeiro, Rita P. and Gama, João},
	editor = {Bouadi, Tassadit and Fromont, Elisa and Hüllermeier, Eyke},
	year = {2022},
	keywords = {Autoencoder, Fault detection, LSTM, Outliers, Time series},
	pages = {39--52},
}

This study applies a data-driven anomaly detection framework based on a Long Short-Term Memory (LSTM) autoencoder network for several subsystems of a public transport bus. The proposed framework efficiently detects abnormal data, significantly reducing the false alarm rate compared to available alternatives. Using historical repair records, we demonstrate how detection of abnormal sequences in the signals can be used for predicting equipment failures. The deviations from normal operation patterns are detected by analysing the data collected from several on-board sensors (e.g., wet tank air pressure, engine speed, engine load) installed on the bus. The performance of LSTM autoencoder (LSTM-AE) is compared against the multi-layer autoencoder (mlAE) network in the same anomaly detection framework. The experimental results show that the performance indicators of the LSTM-AE network, in terms of F1 Score, Recall, and Precision, are better than those of the mlAE network.

@article{calikus_wisdom_2022,
	title = {Wisdom of the contexts: active ensemble learning for contextual anomaly detection},
	volume = {36},
	issn = {1573-756X},
	shorttitle = {Wisdom of the contexts},
	url = {https://doi.org/10.1007/s10618-022-00868-7},
	doi = {10.1007/s10618-022-00868-7},
	abstract = {In contextual anomaly detection, an object is only considered anomalous within a specific context. Most existing methods use a single context based on a set of user-specified contextual features. However, identifying the right context can be very challenging in practice, especially in datasets with a large number of attributes. Furthermore, in real-world systems, there might be multiple anomalies that occur in different contexts and, therefore, require a combination of several “useful” contexts to unveil them. In this work, we propose a novel approach, called wisdom of the contexts (WisCon), to effectively detect complex contextual anomalies in situations where the true contextual and behavioral attributes are unknown. Our method constructs an ensemble of multiple contexts, with varying importance scores, based on the assumption that not all useful contexts are equally so. We estimate the importance of each context using an active learning approach with a novel query strategy. Experiments show that WisCon significantly outperforms existing baselines in different categories (i.e., active learning methods, unsupervised contextual and non-contextual anomaly detectors) on 18 datasets. Furthermore, the results support our initial hypothesis that there is no single perfect context that successfully uncovers all kinds of contextual anomalies, and leveraging the “wisdom” of multiple contexts is necessary.},
	language = {en},
	number = {6},
	urldate = {2023-05-21},
	journal = {Data Mining and Knowledge Discovery},
	author = {Calikus, Ece and Nowaczyk, Sławomir and Bouguelia, Mohamed-Rafik and Dikmen, Onur},
	month = nov,
	year = {2022},
	keywords = {Active learning, Anomaly detection, Contextual anomaly detection, Ensemble learning},
	pages = {2410--2458},
}

In contextual anomaly detection, an object is only considered anomalous within a specific context. Most existing methods use a single context based on a set of user-specified contextual features. However, identifying the right context can be very challenging in practice, especially in datasets with a large number of attributes. Furthermore, in real-world systems, there might be multiple anomalies that occur in different contexts and, therefore, require a combination of several “useful” contexts to unveil them. In this work, we propose a novel approach, called wisdom of the contexts (WisCon), to effectively detect complex contextual anomalies in situations where the true contextual and behavioral attributes are unknown. Our method constructs an ensemble of multiple contexts, with varying importance scores, based on the assumption that not all useful contexts are equally so. We estimate the importance of each context using an active learning approach with a novel query strategy. Experiments show that WisCon significantly outperforms existing baselines in different categories (i.e., active learning methods, unsupervised contextual and non-contextual anomaly detectors) on 18 datasets. Furthermore, the results support our initial hypothesis that there is no single perfect context that successfully uncovers all kinds of contextual anomalies, and leveraging the “wisdom” of multiple contexts is necessary.

@article{gomes_survey_2022,
	title = {A {Survey} on {Semi}-supervised {Learning} for {Delayed} {Partially} {Labelled} {Data} {Streams}},
	volume = {55},
	issn = {0360-0300},
	url = {https://dl.acm.org/doi/10.1145/3523055},
	doi = {10.1145/3523055},
	abstract = {Unlabelled data appear in many domains and are particularly relevant to streaming applications, where even though data is abundant, labelled data is rare. To address the learning problems associated with such data, one can ignore the unlabelled data and focus only on the labelled data (supervised learning); use the labelled data and attempt to leverage the unlabelled data (semi-supervised learning); or assume some labels will be available on request (active learning). The first approach is the simplest, yet the amount of labelled data available will limit the predictive performance. The second relies on finding and exploiting the underlying characteristics of the data distribution. The third depends on an external agent to provide the required labels in a timely fashion. This survey pays special attention to methods that leverage unlabelled data in a semi-supervised setting. We also discuss the delayed labelling issue, which impacts both fully supervised and semi-supervised methods. We propose a unified problem setting, discuss the learning guarantees and existing methods, and explain the differences between related problem settings. Finally, we review the current benchmarking practices and propose adaptations to enhance them.},
	number = {4},
	urldate = {2023-03-31},
	journal = {ACM Computing Surveys},
	author = {Gomes, Heitor Murilo and Grzenda, Maciej and Mello, Rodrigo and Read, Jesse and Le Nguyen, Minh Huong and Bifet, Albert},
	month = nov,
	year = {2022},
	keywords = {Semi-supervised learning, concept drift, data streams, delayed labeling, verification latency},
	pages = {75:1--75:42},
}

Unlabelled data appear in many domains and are particularly relevant to streaming applications, where even though data is abundant, labelled data is rare. To address the learning problems associated with such data, one can ignore the unlabelled data and focus only on the labelled data (supervised learning); use the labelled data and attempt to leverage the unlabelled data (semi-supervised learning); or assume some labels will be available on request (active learning). The first approach is the simplest, yet the amount of labelled data available will limit the predictive performance. The second relies on finding and exploiting the underlying characteristics of the data distribution. The third depends on an external agent to provide the required labels in a timely fashion. This survey pays special attention to methods that leverage unlabelled data in a semi-supervised setting. We also discuss the delayed labelling issue, which impacts both fully supervised and semi-supervised methods. We propose a unified problem setting, discuss the learning guarantees and existing methods, and explain the differences between related problem settings. Finally, we review the current benchmarking practices and propose adaptations to enhance them.

@inproceedings{montiel_online_2022,
	address = {New York, NY, USA},
	series = {{KDD} '22},
	title = {Online {Clustering}: {Algorithms}, {Evaluation}, {Metrics}, {Applications} and {Benchmarking}},
	isbn = {978-1-4503-9385-0},
	shorttitle = {Online {Clustering}},
	url = {https://doi.org/10.1145/3534678.3542600},
	doi = {10.1145/3534678.3542600},
	abstract = {Online clustering algorithms play a critical role in data science, especially with the advantages regarding time, memory usage and complexity, while maintaining a high performance compared to traditional clustering methods. This tutorial serves, first, as a survey on online machine learning and, in particular, data stream clustering methods. During this tutorial, state-of-the-art algorithms and the associated core research threads will be presented by identifying different categories based on distance, density grids and hidden statistical models. Clustering validity indices, an important part of the clustering process which are usually neglected or replaced with classification metrics, resulting in misleading interpretation of final results, will also be deeply investigated. Then, this introduction will be put into the context with River, a go-to Python library merged between Creme and scikit-multiflow. It is also the first open-source project to include an online clustering module that can facilitate reproducibility and allow direct further improvements. From this, we propose methods of clustering configuration, applications and settings for benchmarking, using real-world problems and datasets.},
	urldate = {2023-03-31},
	booktitle = {Proceedings of the 28th {ACM} {SIGKDD} {Conference} on {Knowledge} {Discovery} and {Data} {Mining}},
	publisher = {Association for Computing Machinery},
	author = {Montiel, Jacob and Ngo, Hoang-Anh and Le-Nguyen, Minh-Huong and Bifet, Albert},
	month = aug,
	year = {2022},
	keywords = {benchmarking, data streams, decision support, online clustering, stream clustering, stream learning},
	pages = {4808--4809},
}

Online clustering algorithms play a critical role in data science, especially with the advantages regarding time, memory usage and complexity, while maintaining a high performance compared to traditional clustering methods. This tutorial serves, first, as a survey on online machine learning and, in particular, data stream clustering methods. During this tutorial, state-of-the-art algorithms and the associated core research threads will be presented by identifying different categories based on distance, density grids and hidden statistical models. Clustering validity indices, an important part of the clustering process which are usually neglected or replaced with classification metrics, resulting in misleading interpretation of final results, will also be deeply investigated. Then, this introduction will be put into the context with River, a go-to Python library merged between Creme and scikit-multiflow. It is also the first open-source project to include an online clustering module that can facilitate reproducibility and allow direct further improvements. From this, we propose methods of clustering configuration, applications and settings for benchmarking, using real-world problems and datasets.

@article{elhaik_principal_2022,
	title = {Principal {Component} {Analyses} ({PCA})-based findings in population genetic studies are highly biased and must be reevaluated},
	volume = {12},
	copyright = {2022 The Author(s)},
	issn = {2045-2322},
	url = {https://www.nature.com/articles/s41598-022-14395-4},
	doi = {10.1038/s41598-022-14395-4},
	abstract = {Principal Component Analysis (PCA) is a multivariate analysis that reduces the complexity of datasets while preserving data covariance. The outcome can be visualized on colorful scatterplots, ideally with only a minimal loss of information. PCA applications, implemented in well-cited packages like EIGENSOFT and PLINK, are extensively used as the foremost analyses in population genetics and related fields (e.g., animal and plant or medical genetics). PCA outcomes are used to shape study design, identify, and characterize individuals and populations, and draw historical and ethnobiological conclusions on origins, evolution, dispersion, and relatedness. The replicability crisis in science has prompted us to evaluate whether PCA results are reliable, robust, and replicable. We analyzed twelve common test cases using an intuitive color-based model alongside human population data. We demonstrate that PCA results can be artifacts of the data and can be easily manipulated to generate desired outcomes. PCA adjustment also yielded unfavorable outcomes in association studies. PCA results may not be reliable, robust, or replicable as the field assumes. Our findings raise concerns about the validity of results reported in the population genetics literature and related fields that place a disproportionate reliance upon PCA outcomes and the insights derived from them. We conclude that PCA may have a biasing role in genetic investigations and that 32,000-216,000 genetic studies should be reevaluated. An alternative mixed-admixture population genetic model is discussed.},
	language = {en},
	number = {1},
	urldate = {2023-03-08},
	journal = {Scientific Reports},
	author = {Elhaik, Eran},
	month = aug,
	year = {2022},
	note = {Number: 1
Publisher: Nature Publishing Group},
	keywords = {Computational models, Population genetics},
	pages = {14683},
}

Principal Component Analysis (PCA) is a multivariate analysis that reduces the complexity of datasets while preserving data covariance. The outcome can be visualized on colorful scatterplots, ideally with only a minimal loss of information. PCA applications, implemented in well-cited packages like EIGENSOFT and PLINK, are extensively used as the foremost analyses in population genetics and related fields (e.g., animal and plant or medical genetics). PCA outcomes are used to shape study design, identify, and characterize individuals and populations, and draw historical and ethnobiological conclusions on origins, evolution, dispersion, and relatedness. The replicability crisis in science has prompted us to evaluate whether PCA results are reliable, robust, and replicable. We analyzed twelve common test cases using an intuitive color-based model alongside human population data. We demonstrate that PCA results can be artifacts of the data and can be easily manipulated to generate desired outcomes. PCA adjustment also yielded unfavorable outcomes in association studies. PCA results may not be reliable, robust, or replicable as the field assumes. Our findings raise concerns about the validity of results reported in the population genetics literature and related fields that place a disproportionate reliance upon PCA outcomes and the insights derived from them. We conclude that PCA may have a biasing role in genetic investigations and that 32,000-216,000 genetic studies should be reevaluated. An alternative mixed-admixture population genetic model is discussed.

@article{emmert-streib_taxonomy_2022,
	title = {Taxonomy of machine learning paradigms: {A} data-centric perspective},
	volume = {12},
	issn = {1942-4795},
	shorttitle = {Taxonomy of machine learning paradigms},
	url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/widm.1470},
	doi = {10.1002/widm.1470},
	abstract = {Machine learning is a field composed of various pillars. Traditionally, supervised learning (SL), unsupervised learning (UL), and reinforcement learning (RL) are the dominating learning paradigms that inspired the field since the 1950s. Based on these, thousands of different methods have been developed during the last seven decades used in nearly all application domains. However, recently, other learning paradigms are gaining momentum which complement and extend the above learning paradigms significantly. These are multi-label learning (MLL), semi-supervised learning (SSL), one-class classification (OCC), positive-unlabeled learning (PUL), transfer learning (TL), multi-task learning (MTL), and one-shot learning (OSL). The purpose of this article is a systematic discussion of these modern learning paradigms and their connection to the traditional ones. We discuss each of the learning paradigms formally by defining key constituents and paying particular attention to the data requirements for allowing an easy connection to applications. That means, we assume a data-driven perspective. This perspective will also allow a systematic identification of relations between the individual learning paradigms in the form of a learning-paradigm graph (LP-graph). Overall, the LP-graph establishes a taxonomy among 10 different learning paradigms. This article is categorized under: Technologies {\textgreater} Machine Learning Application Areas {\textgreater} Science and Technology Fundamental Concepts of Data and Knowledge {\textgreater} Key Design Issues in Data Mining},
	language = {en},
	number = {5},
	urldate = {2023-03-08},
	journal = {WIREs Data Mining and Knowledge Discovery},
	author = {Emmert-Streib, Frank and Dehmer, Matthias},
	year = {2022},
	note = {\_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/widm.1470},
	keywords = {artificial intelligence, machine learning, multi-label learning, multi-task learning, transfer learning},
	pages = {e1470},
}

Machine learning is a field composed of various pillars. Traditionally, supervised learning (SL), unsupervised learning (UL), and reinforcement learning (RL) are the dominating learning paradigms that inspired the field since the 1950s. Based on these, thousands of different methods have been developed during the last seven decades used in nearly all application domains. However, recently, other learning paradigms are gaining momentum which complement and extend the above learning paradigms significantly. These are multi-label learning (MLL), semi-supervised learning (SSL), one-class classification (OCC), positive-unlabeled learning (PUL), transfer learning (TL), multi-task learning (MTL), and one-shot learning (OSL). The purpose of this article is a systematic discussion of these modern learning paradigms and their connection to the traditional ones. We discuss each of the learning paradigms formally by defining key constituents and paying particular attention to the data requirements for allowing an easy connection to applications. That means, we assume a data-driven perspective. This perspective will also allow a systematic identification of relations between the individual learning paradigms in the form of a learning-paradigm graph (LP-graph). Overall, the LP-graph establishes a taxonomy among 10 different learning paradigms. This article is categorized under: Technologies \textgreater Machine Learning Application Areas \textgreater Science and Technology Fundamental Concepts of Data and Knowledge \textgreater Key Design Issues in Data Mining

@inproceedings{le-nguyen_real-time_2022,
	address = {Lisbon, Portugal},
	title = {Real-time learning for real-time data: online machine learning for predictive maintenance of railway systems},
	booktitle = {Transport {Research} {Arena} ({TRA})},
	author = {Le-Nguyen, Minh-Huong and Turgis, Fabien and Fayemi, Pierre-Emmanuel and Bifet, Albert},
	month = nov,
	year = {2022},
}

@inproceedings{turgis_health_2022,
	address = {Birmingham, United Kingdom},
	title = {Health state characterization using clustering algorithms for railway maintenance},
	abstract = {To perform maintenance of a large rolling fleet, with operational constraints due to mass transit, a mixed maintenance solution combining real-time data analysis and condition-based maintenance has been integrated into the SNCF maintenance process in 2017. Based on a prognostic expert system, this solution relies on constant signaling thresholds defined using technical knowledge and physical models to assess the health state of a system. As the health of a system differs from one train to another and independently evolves in time, constant signaling thresholds do not always take into account maintenance load, maintenance infrastructure availabilities and effect of aging during the lifetime of the system. To overtake these limitations and enhance the current maintenance solution, an upgrade of the existing expert system has been developed and funded by SNCF Materiel, using dynamic signaling thresholds, based on the study of health indicators distribution across the whole fleet. This article describes the concept of this new hybrid system, mixing expert system and machine learning tools. It shows how dynamic thresholds can be computed and how clustering algorithms can be used to identify and characterize potential failure modes},
	author = {Turgis, Fabien and Audier, Pierre and Nemoz, Valentin and Marion, Rémy},
	year = {2022},
}

To perform maintenance of a large rolling fleet, with operational constraints due to mass transit, a mixed maintenance solution combining real-time data analysis and condition-based maintenance has been integrated into the SNCF maintenance process in 2017. Based on a prognostic expert system, this solution relies on constant signaling thresholds defined using technical knowledge and physical models to assess the health state of a system. As the health of a system differs from one train to another and independently evolves in time, constant signaling thresholds do not always take into account maintenance load, maintenance infrastructure availabilities and effect of aging during the lifetime of the system. To overtake these limitations and enhance the current maintenance solution, an upgrade of the existing expert system has been developed and funded by SNCF Materiel, using dynamic signaling thresholds, based on the study of health indicators distribution across the whole fleet. This article describes the concept of this new hybrid system, mixing expert system and machine learning tools. It shows how dynamic thresholds can be computed and how clustering algorithms can be used to identify and characterize potential failure modes

@inproceedings{le-nguyen_continuous_2022,
	title = {Continuous {Health} {Monitoring} of {Machinery} using {Online} {Clustering} on {Unlabeled} {Data} {Streams}},
	doi = {10.1109/BigData55660.2022.10021002},
	abstract = {Maintenance is an important support function to ensure the reliability, safety, and availability in the railway. Lately, machine learning has become a major player and allows practitioners to build intricate learning models for machinery maintenance. Commonly, a model is trained on static data and is retrained on new data that exhibit novelties unknown to the model. On the contrary, online machine learning is a learning paradigm that adapts the models to new data, thus enabling adaptive, lifelong learning. Our goal is to leverage online learning on unlabeled data streams to enhance railway machinery maintenance. We propose Continuous Health Monitoring using Online Clustering (CheMoc) as an unsupervised method that learns the health profiles of the systems incrementally, assesses their working condition continuously via an adaptive health score, and works efficiently on streaming data. We evaluate CheMoc on a real-world data set from a national railway company. The results show that CheMoc discovered relevant health clusters, as confirmed by a domain expert, and processed the data of an entire year under two hours using only 600 MB of memory.},
	booktitle = {2022 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Le-Nguyen, Minh-Huong and Turgis, Fabien and Fayemi, Pierre-Emmanuel and Bifet, Albert},
	month = dec,
	year = {2022},
	keywords = {Adaptation models, Companies, Employee welfare, Machine learning, Maintenance engineering, Memory management, Rail transportation, maintenance, online clustering, railway},
	pages = {1866--1873},
}

Maintenance is an important support function to ensure the reliability, safety, and availability in the railway. Lately, machine learning has become a major player and allows practitioners to build intricate learning models for machinery maintenance. Commonly, a model is trained on static data and is retrained on new data that exhibit novelties unknown to the model. On the contrary, online machine learning is a learning paradigm that adapts the models to new data, thus enabling adaptive, lifelong learning. Our goal is to leverage online learning on unlabeled data streams to enhance railway machinery maintenance. We propose Continuous Health Monitoring using Online Clustering (CheMoc) as an unsupervised method that learns the health profiles of the systems incrementally, assesses their working condition continuously via an adaptive health score, and works efficiently on streaming data. We evaluate CheMoc on a real-world data set from a national railway company. The results show that CheMoc discovered relevant health clusters, as confirmed by a domain expert, and processed the data of an entire year under two hours using only 600 MB of memory.

@inproceedings{libera_right_2022,
	address = {Cham},
	series = {Lecture {Notes} in {Computer} {Science}},
	title = {‘{Right} to {Be} {Forgotten}’: {Analyzing} the {Impact} of {Forgetting} {Data} {Using} {K}-{NN} {Algorithm} in {Data} {Stream} {Learning}},
	isbn = {978-3-031-15086-9},
	shorttitle = {‘{Right} to {Be} {Forgotten}’},
	doi = {10.1007/978-3-031-15086-9_34},
	abstract = {New international regulations concerning personal management data guarantee the ‘Right to Be Forgotten’. One might request to have their data erased from third-party tools and services. This requirement is especially challenging when considering the behavior of machine learning estimators that will need to forget portions of their knowledge. In this paper, we investigate the impact of these learning and forgetting policies in data stream learning. In data stream mining, the sheer volume of instances typically makes it unfeasible to store the data or retraining the learning models from scratch. Hence, more efficient solutions are needed to deal with the dynamic nature of online machine learning. We modify an incremental k-NN classifier to enable it to erase its past data and we also investigate the impact of data forgetting in the obtained predictive performance. Our proposal is compared against the original k-NN algorithm using seven non-stationary stream datasets. Our results show that the forgetting-enabled algorithm can achieve similar prediction patterns compared to the vanilla one, although it yields lower predictive performance at the beginning of the learning process. Such a scenario is a typical cold-start behavior often observed in data stream mining applications, and not necessarily related to the employed forgetting mechanisms.},
	language = {en},
	booktitle = {Electronic {Government}},
	publisher = {Springer International Publishing},
	author = {Libera, Caio and Miranda, Leandro and Bernardini, Flávia and Mastelini, Saulo and Viterbo, José},
	editor = {Janssen, Marijn and Csáki, Csaba and Lindgren, Ida and Loukis, Euripidis and Melin, Ulf and Viale Pereira, Gabriela and Rodríguez Bolívar, Manuel Pedro and Tambouris, Efthimios},
	year = {2022},
	keywords = {Data stream, K-NN, Lazy learning, Right to be forgotten, Stream learning},
	pages = {530--542},
}

New international regulations concerning personal management data guarantee the ‘Right to Be Forgotten’. One might request to have their data erased from third-party tools and services. This requirement is especially challenging when considering the behavior of machine learning estimators that will need to forget portions of their knowledge. In this paper, we investigate the impact of these learning and forgetting policies in data stream learning. In data stream mining, the sheer volume of instances typically makes it unfeasible to store the data or retraining the learning models from scratch. Hence, more efficient solutions are needed to deal with the dynamic nature of online machine learning. We modify an incremental k-NN classifier to enable it to erase its past data and we also investigate the impact of data forgetting in the obtained predictive performance. Our proposal is compared against the original k-NN algorithm using seven non-stationary stream datasets. Our results show that the forgetting-enabled algorithm can achieve similar prediction patterns compared to the vanilla one, although it yields lower predictive performance at the beginning of the learning process. Such a scenario is a typical cold-start behavior often observed in data stream mining applications, and not necessarily related to the employed forgetting mechanisms.

@misc{nozawa_empirical_2022,
	title = {Empirical {Evaluation} and {Theoretical} {Analysis} for {Representation} {Learning}: {A} {Survey}},
	shorttitle = {Empirical {Evaluation} and {Theoretical} {Analysis} for {Representation} {Learning}},
	url = {http://arxiv.org/abs/2204.08226},
	doi = {10.48550/arXiv.2204.08226},
	abstract = {Representation learning enables us to automatically extract generic feature representations from a dataset to solve another machine learning task. Recently, extracted feature representations by a representation learning algorithm and a simple predictor have exhibited state-of-the-art performance on several machine learning tasks. Despite its remarkable progress, there exist various ways to evaluate representation learning algorithms depending on the application because of the flexibility of representation learning. To understand the current representation learning, we review evaluation methods of representation learning algorithms and theoretical analyses. On the basis of our evaluation survey, we also discuss the future direction of representation learning. Note that this survey is the extended version of Nozawa and Sato (2022).},
	urldate = {2022-07-29},
	publisher = {arXiv},
	author = {Nozawa, Kento and Sato, Issei},
	month = apr,
	year = {2022},
	note = {arXiv:2204.08226 [cs]},
	keywords = {Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning},
}

Representation learning enables us to automatically extract generic feature representations from a dataset to solve another machine learning task. Recently, extracted feature representations by a representation learning algorithm and a simple predictor have exhibited state-of-the-art performance on several machine learning tasks. Despite its remarkable progress, there exist various ways to evaluate representation learning algorithms depending on the application because of the flexibility of representation learning. To understand the current representation learning, we review evaluation methods of representation learning algorithms and theoretical analyses. On the basis of our evaluation survey, we also discuss the future direction of representation learning. Note that this survey is the extended version of Nozawa and Sato (2022).

@article{sui_dynamic_2022,
	title = {Dynamic {Sparse} {Subspace} {Clustering} for {Evolving} {High}-{Dimensional} {Data} {Streams}},
	volume = {52},
	issn = {2168-2275},
	doi = {10.1109/TCYB.2020.3023973},
	abstract = {In an era of ubiquitous large-scale evolving data streams, data stream clustering (DSC) has received lots of attention because the scale of the data streams far exceeds the ability of expert human analysts. It has been observed that high-dimensional data are usually distributed in a union of low-dimensional subspaces. In this article, we propose a novel sparse representation-based DSC algorithm, called evolutionary dynamic sparse subspace clustering (EDSSC). It can cope with the time-varying nature of subspaces underlying the evolving data streams, such as subspace emergence, disappearance, and recurrence. The proposed EDSSC consists of two phases: 1) static learning and 2) online clustering. During the first phase, a data structure for storing the statistic summary of data streams, called EDSSC summary, is proposed which can better address the dilemma between the two conflicting goals: 1) saving more points for accuracy of subspace clustering (SC) and 2) discarding more points for the efficiency of DSC. By further proposing an algorithm to estimate the subspace number, the proposed EDSSC does not need to know the number of subspaces. In the second phase, a more suitable index, called the average sparsity concentration index (ASCI), is proposed, which dramatically promotes the clustering accuracy compared to the conventionally utilized SCI index. In addition, the subspace evolution detection model based on the Page-Hinkley test is proposed where the appearing, disappearing, and recurring subspaces can be detected and adapted. Extinct experiments on real-world data streams show that the EDSSC outperforms the state-of-the-art online SC approaches.},
	number = {6},
	journal = {IEEE Transactions on Cybernetics},
	author = {Sui, Jinping and Liu, Zhen and Liu, Li and Jung, Alexander and Li, Xiang},
	month = jun,
	year = {2022},
	note = {Conference Name: IEEE Transactions on Cybernetics},
	keywords = {Adaptation models, Clustering algorithms, Data models, Data stream clustering (DSC), Data structures, Heuristic algorithms, Indexes, Task analysis, high-dimensional data stream, sparse representation, subspace clustering (SC)},
	pages = {4173--4186},
}

In an era of ubiquitous large-scale evolving data streams, data stream clustering (DSC) has received lots of attention because the scale of the data streams far exceeds the ability of expert human analysts. It has been observed that high-dimensional data are usually distributed in a union of low-dimensional subspaces. In this article, we propose a novel sparse representation-based DSC algorithm, called evolutionary dynamic sparse subspace clustering (EDSSC). It can cope with the time-varying nature of subspaces underlying the evolving data streams, such as subspace emergence, disappearance, and recurrence. The proposed EDSSC consists of two phases: 1) static learning and 2) online clustering. During the first phase, a data structure for storing the statistic summary of data streams, called EDSSC summary, is proposed which can better address the dilemma between the two conflicting goals: 1) saving more points for accuracy of subspace clustering (SC) and 2) discarding more points for the efficiency of DSC. By further proposing an algorithm to estimate the subspace number, the proposed EDSSC does not need to know the number of subspaces. In the second phase, a more suitable index, called the average sparsity concentration index (ASCI), is proposed, which dramatically promotes the clustering accuracy compared to the conventionally utilized SCI index. In addition, the subspace evolution detection model based on the Page-Hinkley test is proposed where the appearing, disappearing, and recurring subspaces can be detected and adapted. Extinct experiments on real-world data streams show that the EDSSC outperforms the state-of-the-art online SC approaches.

@article{yu_meta-add_2022,
	title = {Meta-{ADD}: {A} meta-learning based pre-trained model for concept drift active detection},
	volume = {608},
	issn = {0020-0255},
	shorttitle = {Meta-{ADD}},
	url = {https://www.sciencedirect.com/science/article/pii/S0020025522007125},
	doi = {10.1016/j.ins.2022.07.022},
	abstract = {Concept drift is a phenomenon that commonly happened in data streams and need to be detected, because it means the statistical properties of a target variable, which the model is trying to predict, change over time in an unseen way. Most current detection methods are based on a hypothesis test framework. As a result, in these detection methods, a hypothesis test is need to be set, and more importantly, cannot obtain the type of drift. The setting of a hypothesis test requires an understanding of data streams, and cannot obtain the type of concept drift results in the loss of drift information. Hence, in this paper, to get rid of the setting of hypothesis test, and obtain the type of concept drift, we propose Active Drift Detection based on Meta learning (Meta-ADD), a novel framework that learns to classify concept drift by offline pre-training a model on data stream with known drifts, then online fine-tuning model to improve detection accuracy. Specifically, in the pre-trained phase, we extract meta-features based on the error rates of various concept drift, after which a pre-trained model called meta-detector is developed via a prototypical neural network by representing various concept drift classes as corresponding prototypes. In the detection phase, the meta-detector is fine-tuned to adapt to the real data stream via a simple stream-based active learning. Hence, Meta-ADD does not need a hypothesis test to detect concept drifts and identify their types automatically, which can directly support drift understand. The experiment results verify the effectiveness of Meta-ADD.},
	language = {en},
	urldate = {2022-07-12},
	journal = {Information Sciences},
	author = {Yu, Hang and Zhang, Qingyong and Liu, Tianyu and Lu, Jie and Wen, Yimin and Zhang, Guangquan},
	month = aug,
	year = {2022},
	keywords = {Concept drift, Drift detection, Pre-trained model, Prototypical neural networks},
	pages = {996--1009},
}

Concept drift is a phenomenon that commonly happened in data streams and need to be detected, because it means the statistical properties of a target variable, which the model is trying to predict, change over time in an unseen way. Most current detection methods are based on a hypothesis test framework. As a result, in these detection methods, a hypothesis test is need to be set, and more importantly, cannot obtain the type of drift. The setting of a hypothesis test requires an understanding of data streams, and cannot obtain the type of concept drift results in the loss of drift information. Hence, in this paper, to get rid of the setting of hypothesis test, and obtain the type of concept drift, we propose Active Drift Detection based on Meta learning (Meta-ADD), a novel framework that learns to classify concept drift by offline pre-training a model on data stream with known drifts, then online fine-tuning model to improve detection accuracy. Specifically, in the pre-trained phase, we extract meta-features based on the error rates of various concept drift, after which a pre-trained model called meta-detector is developed via a prototypical neural network by representing various concept drift classes as corresponding prototypes. In the detection phase, the meta-detector is fine-tuned to adapt to the real data stream via a simple stream-based active learning. Hence, Meta-ADD does not need a hypothesis test to detect concept drifts and identify their types automatically, which can directly support drift understand. The experiment results verify the effectiveness of Meta-ADD.

@article{zhang_online_2022,
	title = {Online {Learning} of {Wearable} {Sensing} for {Human} {Activity} {Recognition}},
	issn = {2327-4662},
	doi = {10.1109/JIOT.2022.3188785},
	abstract = {The paper presents a novel semi supervised learning method for wearable sensors to recognize human activities. The proposed method is termed as tri-VFDT (Very Fast Decision Tree). The proposed method is a more efficient version of the Hoeffding tree and three VFDT are generated from the original labeled example set and refined using unlabeled examples. Based on the heuristic growth characteristics of VFDT, a tri training framework is proposed which uses unlabeled data to update the model without labeled data. This significantly reduces the computational time and storage of the data processing. In addition, the proposed method is embedded into wearable devices for online learning, while the test data flow is regarded as the unlabeled data to update model. The experiment collects data stream of 16 minutes with motion state switching frequently while the wearable devices recognize motions in real time. An experimental comparison has also been undertaken for performance evaluation between the wearable and computation using desktop computer. The obtained results show that only minor difference in terms of the f1-score rendered by the proposed method online or offline. This is a prominent characteristic for wearable computing within internet of thing (IOT).},
	journal = {IEEE Internet of Things Journal},
	author = {Zhang, Yiwei and Gao, Bin and Yang, Daili and Woo, Wai Lok and Wen, Houlai},
	year = {2022},
	note = {Conference Name: IEEE Internet of Things Journal},
	keywords = {Classification algorithms, Computational modeling, Feature extraction, Histograms, Predictive models, Training, Wearable computers, online learning, real-time activity recognition, semi-supervised learning, wearable device},
	pages = {1--1},
}

The paper presents a novel semi supervised learning method for wearable sensors to recognize human activities. The proposed method is termed as tri-VFDT (Very Fast Decision Tree). The proposed method is a more efficient version of the Hoeffding tree and three VFDT are generated from the original labeled example set and refined using unlabeled examples. Based on the heuristic growth characteristics of VFDT, a tri training framework is proposed which uses unlabeled data to update the model without labeled data. This significantly reduces the computational time and storage of the data processing. In addition, the proposed method is embedded into wearable devices for online learning, while the test data flow is regarded as the unlabeled data to update model. The experiment collects data stream of 16 minutes with motion state switching frequently while the wearable devices recognize motions in real time. An experimental comparison has also been undertaken for performance evaluation between the wearable and computation using desktop computer. The obtained results show that only minor difference in terms of the f1-score rendered by the proposed method online or offline. This is a prominent characteristic for wearable computing within internet of thing (IOT).

@article{gonzalez-muniz_health_2022,
	title = {Health indicator for machine condition monitoring built in the latent space of a deep autoencoder},
	volume = {224},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022001417},
	doi = {10.1016/j.ress.2022.108482},
	abstract = {The construction of effective health indicators plays a key role in the engineering systems field: they reflect the degradation degree of the system under study, thus providing vital information for critical tasks ranging from anomaly detection to remaining useful life estimation, with benefits such as reduced maintenance costs, improved productivity or increased machine availability. The reconstruction error of deep autoencoders has been widely used in the literature for this purpose, but this approach does not fully exploit the hierarchical nature of deep models. Instead, we propose to take advantage of the disentangled representations of data that are available in the latent space of autoencoders, by using the latent reconstruction error as machine health indicator. We have tested our proposal on three different datasets, considering two types of autoencoders (deep autoencoder and variational autoencoder), and comparing its performance with that of state-of-the-art approaches in terms of well-known quality metrics. The results of the research demonstrate the capability of our health indicator to outperform conventional approaches, in the three datasets, and regardless of the type of autoencoder used to generate the residuals. In addition, we provide some intuition on the suitability of latent spaces for the monitoring of machinery condition.},
	language = {en},
	urldate = {2022-05-14},
	journal = {Reliability Engineering \& System Safety},
	author = {González-Muñiz, Ana and Díaz, Ignacio and Cuadrado, Abel A. and García-Pérez, Diego},
	month = aug,
	year = {2022},
	keywords = {Anomaly detection, Deep autoencoder, Engineering systems, Health indicator, Latent space},
	pages = {108482},
}

The construction of effective health indicators plays a key role in the engineering systems field: they reflect the degradation degree of the system under study, thus providing vital information for critical tasks ranging from anomaly detection to remaining useful life estimation, with benefits such as reduced maintenance costs, improved productivity or increased machine availability. The reconstruction error of deep autoencoders has been widely used in the literature for this purpose, but this approach does not fully exploit the hierarchical nature of deep models. Instead, we propose to take advantage of the disentangled representations of data that are available in the latent space of autoencoders, by using the latent reconstruction error as machine health indicator. We have tested our proposal on three different datasets, considering two types of autoencoders (deep autoencoder and variational autoencoder), and comparing its performance with that of state-of-the-art approaches in terms of well-known quality metrics. The results of the research demonstrate the capability of our health indicator to outperform conventional approaches, in the three datasets, and regardless of the type of autoencoder used to generate the residuals. In addition, we provide some intuition on the suitability of latent spaces for the monitoring of machinery condition.

@article{reinartz_pytep_2022,
	title = {{pyTEP}: {A} {Python} package for interactive simulations of the {Tennessee} {Eastman} process},
	volume = {18},
	issn = {2352-7110},
	shorttitle = {{pyTEP}},
	url = {https://www.sciencedirect.com/science/article/pii/S2352711022000449},
	doi = {10.1016/j.softx.2022.101053},
	abstract = {pyTEP is an open-source simulation API for the Tennessee Eastman process in Python. It facilitates the setup of complex simulation scenarios and provides the option of interactive simulation. The Tennessee Eastman process has been the go-to benchmark for statistical process monitoring and machine learning based fault-detection approaches for continuous chemical processes in recent years, but its potential outside these domains remains largely untapped. Existing simulators are tailored towards simulations of stationary operating conditions in the presence of faults, but further extensions for more complex simulation scenarios are time-consuming, which may discourage researchers from adopting the process. Through pyTEPs API, users can configure simulations, change operating conditions and store simulation data without being exposed to the underlying mechanics of the simulator. In addition to the newly introduced features, pyTEP promises more versatility and more straightforward usage than existing TEP simulators.},
	language = {en},
	urldate = {2022-05-04},
	journal = {SoftwareX},
	author = {Reinartz, Christopher and Enevoldsen, Thomas T.},
	month = jun,
	year = {2022},
	keywords = {Chemical process simulation, Python, Simulation framework, Tennessee Eastman process},
	pages = {101053},
}

pyTEP is an open-source simulation API for the Tennessee Eastman process in Python. It facilitates the setup of complex simulation scenarios and provides the option of interactive simulation. The Tennessee Eastman process has been the go-to benchmark for statistical process monitoring and machine learning based fault-detection approaches for continuous chemical processes in recent years, but its potential outside these domains remains largely untapped. Existing simulators are tailored towards simulations of stationary operating conditions in the presence of faults, but further extensions for more complex simulation scenarios are time-consuming, which may discourage researchers from adopting the process. Through pyTEPs API, users can configure simulations, change operating conditions and store simulation data without being exposed to the underlying mechanics of the simulator. In addition to the newly introduced features, pyTEP promises more versatility and more straightforward usage than existing TEP simulators.

@article{haug_standardized_2022,
	title = {Standardized {Evaluation} of {Machine} {Learning} {Methods} for {Evolving} {Data} {Streams}},
	url = {http://arxiv.org/abs/2204.13625},
	abstract = {Due to the unspecified and dynamic nature of data streams, online machine learning requires powerful and flexible solutions. However, evaluating online machine learning methods under realistic conditions is difficult. Existing work therefore often draws on different heuristics and simulations that do not necessarily produce meaningful and reliable results. Indeed, in the absence of common evaluation standards, it often remains unclear how online learning methods will perform in practice or in comparison to similar work. In this paper, we propose a comprehensive set of properties for high-quality machine learning in evolving data streams. In particular, we discuss sensible performance measures and evaluation strategies for online predictive modelling, online feature selection and concept drift detection. As one of the first works, we also look at the interpretability of online learning methods. The proposed evaluation standards are provided in a new Python framework called float. Float is completely modular and allows the simultaneous integration of common libraries, such as scikit-multiflow or river, with custom code. Float is open-sourced and can be accessed at https://github.com/haugjo/float. In this sense, we hope that our work will contribute to more standardized, reliable and realistic testing and comparison of online machine learning methods.},
	urldate = {2022-05-03},
	journal = {arXiv:2204.13625 [cs, stat]},
	author = {Haug, Johannes and Tramountani, Effi and Kasneci, Gjergji},
	month = apr,
	year = {2022},
	note = {arXiv: 2204.13625},
	keywords = {Computer Science - Machine Learning, Statistics - Machine Learning},
}

Due to the unspecified and dynamic nature of data streams, online machine learning requires powerful and flexible solutions. However, evaluating online machine learning methods under realistic conditions is difficult. Existing work therefore often draws on different heuristics and simulations that do not necessarily produce meaningful and reliable results. Indeed, in the absence of common evaluation standards, it often remains unclear how online learning methods will perform in practice or in comparison to similar work. In this paper, we propose a comprehensive set of properties for high-quality machine learning in evolving data streams. In particular, we discuss sensible performance measures and evaluation strategies for online predictive modelling, online feature selection and concept drift detection. As one of the first works, we also look at the interpretability of online learning methods. The proposed evaluation standards are provided in a new Python framework called float. Float is completely modular and allows the simultaneous integration of common libraries, such as scikit-multiflow or river, with custom code. Float is open-sourced and can be accessed at https://github.com/haugjo/float. In this sense, we hope that our work will contribute to more standardized, reliable and realistic testing and comparison of online machine learning methods.

@article{guo_online_2022,
	title = {Online {Anomaly} {Detection} of {Industrial} {IoT} {Based} on {Hybrid} {Machine} {Learning} {Architecture}},
	volume = {2022},
	issn = {1687-5265},
	url = {https://www.hindawi.com/journals/cin/2022/8568917/},
	doi = {10.1155/2022/8568917},
	abstract = {Industrial IoT (IIoT) in Industry 4.0 integrates everything at the level of information technology with the level of technology of operation and aims to improve Business to Business (B2B) services (from production to public services). It includes Machine to Machine (M2M) interaction either for process control (e.g., factory processes, fleet tracking) or as part of self-organizing cyber-physical distributed control systems without human intervention. A critical factor in completing the abovementioned actions is the development of intelligent software systems in the context of automatic control of the business environment, with the ability to analyze in real-time the existing equipment through the available interfaces (hardware-in-the-loop). In this spirit, this paper presents an advanced intelligent approach to real-time monitoring of the operation of industrial equipment. A hybrid novel methodology that combines memory neural networks is used, and Bayesian methods that examine a variety of characteristic quantities of vibration signals that are exported in the field of time, with the aim of real-time detection of abnormalities in active IIoT equipment are also used.},
	language = {en},
	urldate = {2022-05-03},
	journal = {Computational Intelligence and Neuroscience},
	author = {Guo, Jia and Shen, Yue},
	month = apr,
	year = {2022},
	note = {Publisher: Hindawi},
	pages = {e8568917},
}

Industrial IoT (IIoT) in Industry 4.0 integrates everything at the level of information technology with the level of technology of operation and aims to improve Business to Business (B2B) services (from production to public services). It includes Machine to Machine (M2M) interaction either for process control (e.g., factory processes, fleet tracking) or as part of self-organizing cyber-physical distributed control systems without human intervention. A critical factor in completing the abovementioned actions is the development of intelligent software systems in the context of automatic control of the business environment, with the ability to analyze in real-time the existing equipment through the available interfaces (hardware-in-the-loop). In this spirit, this paper presents an advanced intelligent approach to real-time monitoring of the operation of industrial equipment. A hybrid novel methodology that combines memory neural networks is used, and Bayesian methods that examine a variety of characteristic quantities of vibration signals that are exported in the field of time, with the aim of real-time detection of abnormalities in active IIoT equipment are also used.

@article{andersen_easy_2022,
	title = {An easy to use {GUI} for simulating big data using {Tennessee} {Eastman} process},
	volume = {38},
	issn = {1099-1638},
	url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/qre.2975},
	doi = {10.1002/qre.2975},
	abstract = {Data-driven process monitoring and control techniques and their application to industrial chemical processes are gaining popularity due to the current focus on Industry 4.0, digitalization and the Internet of Things. However, for the development of such techniques, there are significant barriers that must be overcome in obtaining sufficiently large and reliable datasets. As a result, the use of real plant and process data in developing and testing data-driven process monitoring and control tools can be difficult without investing significant efforts in acquiring, treating, and interpreting the data. Therefore, researchers need a tool that effortlessly generates large amounts of realistic and reliable process data without the requirement for additional data treatment or interpretation. In this work, we propose a data generation platform based on the Tennessee Eastman Process simulation benchmark. A graphical user interface (GUI) developed in MATLAB Simulink is presented that enables users to generate massive amounts of data for testing applicability of big data concepts in the realm of process control for continuous time-dependent processes. An R-Shiny app that interacts with the data generation tool is also presented for illustration purposes. The app can visualize the results generated by the Tennessee Eastman Process and can carry out a standard fault detection and diagnosis studies based on PCA. The data generator GUI is available free of charge for research purposes at https://github.com/dtuprodana/TEP.},
	language = {en},
	number = {1},
	urldate = {2022-05-02},
	journal = {Quality and Reliability Engineering International},
	author = {Andersen, Emil B. and Udugama, Isuru A. and Gernaey, Krist V. and Khan, Abdul R. and Bayer, Christoph and Kulahci, Murat},
	year = {2022},
	note = {\_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/qre.2975},
	keywords = {chemical process, digitalization, industry 4.0, process monitoring and control, process simulator, process surveillance},
	pages = {264--282},
}

Data-driven process monitoring and control techniques and their application to industrial chemical processes are gaining popularity due to the current focus on Industry 4.0, digitalization and the Internet of Things. However, for the development of such techniques, there are significant barriers that must be overcome in obtaining sufficiently large and reliable datasets. As a result, the use of real plant and process data in developing and testing data-driven process monitoring and control tools can be difficult without investing significant efforts in acquiring, treating, and interpreting the data. Therefore, researchers need a tool that effortlessly generates large amounts of realistic and reliable process data without the requirement for additional data treatment or interpretation. In this work, we propose a data generation platform based on the Tennessee Eastman Process simulation benchmark. A graphical user interface (GUI) developed in MATLAB Simulink is presented that enables users to generate massive amounts of data for testing applicability of big data concepts in the realm of process control for continuous time-dependent processes. An R-Shiny app that interacts with the data generation tool is also presented for illustration purposes. The app can visualize the results generated by the Tennessee Eastman Process and can carry out a standard fault detection and diagnosis studies based on PCA. The data generator GUI is available free of charge for research purposes at https://github.com/dtuprodana/TEP.

@article{watson_sequential_2022,
	title = {Sequential detection of a temporary change in multivariate time series},
	volume = {127},
	issn = {1051-2004},
	url = {https://www.sciencedirect.com/science/article/pii/S1051200422001622},
	doi = {10.1016/j.dsp.2022.103545},
	abstract = {In this work, we aim to provide a new and efficient recursive detection method for temporarily monitored signals. Motivated by the case of the propagation of an event over a field of sensors, we assumed that the change in the statistical properties in the monitored signals can only be temporary. Unfortunately, to our best knowledge, existing recursive and simple detection techniques such as the ones based on the cumulative sum (CUSUM) do not consider the temporary aspect of the change in a multivariate time series. In this paper, we propose a novel simple and efficient sequential detection algorithm, named Temporary-Event-CUSUM (TE-CUSUM). By combining with a new adaptive way to aggregate local CUSUM variables from each data stream, we empirically show that the TE-CUSUM has a very good detection rate in the case of an event passing through a field of sensors in a very noisy environment.},
	language = {en},
	urldate = {2022-04-09},
	journal = {Digital Signal Processing},
	author = {Watson, Victor and Septier, François and Armand, Patrick and Duchenne, Christophe},
	month = jul,
	year = {2022},
	keywords = {CUSUM, Multivariate time series, Sequential detection, Temporary event},
	pages = {103545},
}

In this work, we aim to provide a new and efficient recursive detection method for temporarily monitored signals. Motivated by the case of the propagation of an event over a field of sensors, we assumed that the change in the statistical properties in the monitored signals can only be temporary. Unfortunately, to our best knowledge, existing recursive and simple detection techniques such as the ones based on the cumulative sum (CUSUM) do not consider the temporary aspect of the change in a multivariate time series. In this paper, we propose a novel simple and efficient sequential detection algorithm, named Temporary-Event-CUSUM (TE-CUSUM). By combining with a new adaptive way to aggregate local CUSUM variables from each data stream, we empirically show that the TE-CUSUM has a very good detection rate in the case of an event passing through a field of sensors in a very noisy environment.

@article{malik_human_2022,
	title = {Human action interpretation using convolutional neural network: a survey},
	volume = {33},
	issn = {1432-1769},
	shorttitle = {Human action interpretation using convolutional neural network},
	url = {https://doi.org/10.1007/s00138-022-01291-0},
	doi = {10.1007/s00138-022-01291-0},
	abstract = {Human action interpretation (HAI) is one of the trending domains in the era of computer vision. It can further be divided into human action recognition (HAR) and human action detection (HAD). The HAR analyzes frames and provides label(s) to overall video, whereas the HAD localizes actor first, in each frame, and then estimates the action score for the detected region. The effectiveness of a HAI model is highly dependent on the representation of spatiotemporal features and the model’s architectural design. For the effective representation of these features, various studies have been carried out. Moreover, to better learn these features and to get the action score on the basis of these features, different designs of deep architectures have also been proposed. Among various deep architectures, convolutional neural network (CNN) is relatively more explored for HAI due to its lesser computational cost. To provide overview of these efforts, various surveys have been published to date; however, none of these surveys is focusing the features’ representation and design of proposed architectures in detail. Secondly, none of these studies is focusing the pose assisted HAI techniques. This study provides a more detailed survey on existing CNN-based HAI techniques by incorporating the frame level as well as pose level spatiotemporal features-based techniques. Besides these, it offers comparative study on different publicly available datasets used to evaluate HAI models based on various spatiotemporal features’ representations. Furthermore, it also discusses the limitations and challenges of the HAI and concludes that human action interpretation from visual data is still very far from the actual interpretation of human action in realistic videos which are continuous in nature and may contain multiple human beings performing multiple actions sequentially or in parallel.},
	language = {en},
	number = {3},
	urldate = {2022-03-25},
	journal = {Machine Vision and Applications},
	author = {Malik, Zainab and Shapiai, Mohd Ibrahim Bin},
	month = mar,
	year = {2022},
	pages = {37},
}

Human action interpretation (HAI) is one of the trending domains in the era of computer vision. It can further be divided into human action recognition (HAR) and human action detection (HAD). The HAR analyzes frames and provides label(s) to overall video, whereas the HAD localizes actor first, in each frame, and then estimates the action score for the detected region. The effectiveness of a HAI model is highly dependent on the representation of spatiotemporal features and the model’s architectural design. For the effective representation of these features, various studies have been carried out. Moreover, to better learn these features and to get the action score on the basis of these features, different designs of deep architectures have also been proposed. Among various deep architectures, convolutional neural network (CNN) is relatively more explored for HAI due to its lesser computational cost. To provide overview of these efforts, various surveys have been published to date; however, none of these surveys is focusing the features’ representation and design of proposed architectures in detail. Secondly, none of these studies is focusing the pose assisted HAI techniques. This study provides a more detailed survey on existing CNN-based HAI techniques by incorporating the frame level as well as pose level spatiotemporal features-based techniques. Besides these, it offers comparative study on different publicly available datasets used to evaluate HAI models based on various spatiotemporal features’ representations. Furthermore, it also discusses the limitations and challenges of the HAI and concludes that human action interpretation from visual data is still very far from the actual interpretation of human action in realistic videos which are continuous in nature and may contain multiple human beings performing multiple actions sequentially or in parallel.

@article{han_streamdfp_2022,
	title = {{StreamDFP}: {A} {General} {Stream} {Mining} {Framework} for {Adaptive} {Disk} {Failure} {Prediction}},
	issn = {1557-9956},
	shorttitle = {{StreamDFP}},
	doi = {10.1109/TC.2022.3160365},
	abstract = {We explore machine learning for accurately predicting imminent disk failures and hence providing proactive fault tolerance for modern large-scale storage systems. Current disk failure prediction approaches are mostly offline and assume that the disk logs required for training learning models are available a priori. However, disk logs are often continuously generated as an evolving data stream, in which the statistical patterns vary over time (also known as concept drift). Such a challenge motivates the need of online techniques that perform training and prediction on the incoming stream of disk logs in real time, while being adaptive to concept drift. We first measure and demonstrate the existence of concept drift on various disk models from Backblaze and Alibaba Cloud. Motivated by our study, we design STREAMDFP, a general stream mining framework for disk failure prediction with concept-drift adaption based on three key techniques, namely online labeling, concept-drift-aware training, and general prediction, with a primary objective of supporting various machine learning algorithms. We extend STREAMDFP to support online transfer learning for minority disk models with concept-drift adaptation. Our evaluation shows that STREAMDFP improves the prediction accuracy significantly compared to without concept-drift adaptation under various settings, and achieves reasonably high stream processing performance.},
	journal = {IEEE Transactions on Computers},
	author = {Han, Shujie and Lee, Patrick P. C. and Shen, Zhirong and He, Cheng and Liu, Yi and Huang, Tao},
	year = {2022},
	note = {Conference Name: IEEE Transactions on Computers},
	keywords = {Adaptation models, Machine learning algorithms, Prediction algorithms, Predictive models, Production, Random forests, Training, and online transfer learning, concept drift, disk failure prediction, stream mining},
	pages = {1--1},
}

We explore machine learning for accurately predicting imminent disk failures and hence providing proactive fault tolerance for modern large-scale storage systems. Current disk failure prediction approaches are mostly offline and assume that the disk logs required for training learning models are available a priori. However, disk logs are often continuously generated as an evolving data stream, in which the statistical patterns vary over time (also known as concept drift). Such a challenge motivates the need of online techniques that perform training and prediction on the incoming stream of disk logs in real time, while being adaptive to concept drift. We first measure and demonstrate the existence of concept drift on various disk models from Backblaze and Alibaba Cloud. Motivated by our study, we design STREAMDFP, a general stream mining framework for disk failure prediction with concept-drift adaption based on three key techniques, namely online labeling, concept-drift-aware training, and general prediction, with a primary objective of supporting various machine learning algorithms. We extend STREAMDFP to support online transfer learning for minority disk models with concept-drift adaptation. Our evaluation shows that STREAMDFP improves the prediction accuracy significantly compared to without concept-drift adaptation under various settings, and achieves reasonably high stream processing performance.

@article{zhang_real-time_2022,
	title = {Real-time prediction of rate of penetration by combining attention-based gated recurrent unit network and fully connected neural networks},
	issn = {0920-4105},
	url = {https://www.sciencedirect.com/science/article/pii/S0920410522002820},
	doi = {10.1016/j.petrol.2022.110396},
	abstract = {Data-driven models are widely used to predict rate of penetration. However, there are still challenges on real-time predictions considering influences of formation properties and bit wear. In this paper, a novel data-driven model is proposed to tackle this problem by combining an attention-based Gated Recurrent Unit network and fully connected neural networks. At first, input features of the model are elaborately selected by physical drilling laws and statistical analyzes. Then, four sub-networks are employed to construct the whole model structure, where formation properties are assessed using well-logging data and bit wear is evaluated by introducing an attention-based Gated Recurrent Unit network. Next, the model is dynamically updated with data streams by implementing the sliding window method to realize real-time predictions. Finally, the model performance is thoroughly analyzed based on ten field drilling datasets after optimizing model hyperparameters using the orthogonal experiment method. Results indicate that the model is accurate and robust to give predictions after training with the first several data streams. Compared with the conventional data-driven models, the proposed model shows great superiority due to the sub-network structure, the Gated Recurrent Unit network, and the attention mechanism. The model proposed herein opens opportunities for real-time prediction of rate of penetration in the field with high accuracy and robustness.},
	language = {en},
	urldate = {2022-03-19},
	journal = {Journal of Petroleum Science and Engineering},
	author = {Zhang, Chengkai and Song, Xianzhi and Su, Yinao and Li, Gensheng},
	month = mar,
	year = {2022},
	keywords = {Attention mechanism, Data-driven model, Gated recurrent unit (GRU), Neural network, Rate of penetration, Real-time prediction},
	pages = {110396},
}

Data-driven models are widely used to predict rate of penetration. However, there are still challenges on real-time predictions considering influences of formation properties and bit wear. In this paper, a novel data-driven model is proposed to tackle this problem by combining an attention-based Gated Recurrent Unit network and fully connected neural networks. At first, input features of the model are elaborately selected by physical drilling laws and statistical analyzes. Then, four sub-networks are employed to construct the whole model structure, where formation properties are assessed using well-logging data and bit wear is evaluated by introducing an attention-based Gated Recurrent Unit network. Next, the model is dynamically updated with data streams by implementing the sliding window method to realize real-time predictions. Finally, the model performance is thoroughly analyzed based on ten field drilling datasets after optimizing model hyperparameters using the orthogonal experiment method. Results indicate that the model is accurate and robust to give predictions after training with the first several data streams. Compared with the conventional data-driven models, the proposed model shows great superiority due to the sub-network structure, the Gated Recurrent Unit network, and the attention mechanism. The model proposed herein opens opportunities for real-time prediction of rate of penetration in the field with high accuracy and robustness.

@article{spanninger_review_2022,
	title = {A review of train delay prediction approaches},
	volume = {22},
	issn = {2210-9706},
	url = {https://www.sciencedirect.com/science/article/pii/S2210970622000166},
	doi = {10.1016/j.jrtpm.2022.100312},
	abstract = {Railway operations are vulnerable to delays. Accurate predictions of train arrival and departure delays improve the passenger service quality and are essential for real-time railway traffic management to minimise their further spreading. This review provides a synoptic overview and discussion covering the breadth of diverse approaches to predict train delays. We first categorise research contributions based on their underlying modelling paradigm (data-driven and event-driven) and their mathematical model. We then distinguish between very short to long-term predictions and classify different input data sources that have been considered in the literature. We further discuss advantages and disadvantages of producing deterministic versus stochastic predictions, the applicability of different approaches during disruptions and their interpretability. By comparing the results of the included contributions, we can indicate that the prediction error generally increases when broadening the prediction horizon. We find that data-driven approaches might have the edge on event-driven approaches in terms of prediction accuracy, whereas event-driven approaches that explicitly model the dynamics and dependencies of railway traffic have their strength in providing interpretable predictions, and are more robust concerning disruption scenarios. The growing availability of railway operations data is expected to increase the appeal of big-data and machine learning methods.},
	language = {en},
	urldate = {2022-03-19},
	journal = {Journal of Rail Transport Planning \& Management},
	author = {Spanninger, Thomas and Trivella, Alessio and Büchel, Beda and Corman, Francesco},
	month = jun,
	year = {2022},
	keywords = {Forecasting, Prediction, Railways, Train delay},
	pages = {100312},
}

Railway operations are vulnerable to delays. Accurate predictions of train arrival and departure delays improve the passenger service quality and are essential for real-time railway traffic management to minimise their further spreading. This review provides a synoptic overview and discussion covering the breadth of diverse approaches to predict train delays. We first categorise research contributions based on their underlying modelling paradigm (data-driven and event-driven) and their mathematical model. We then distinguish between very short to long-term predictions and classify different input data sources that have been considered in the literature. We further discuss advantages and disadvantages of producing deterministic versus stochastic predictions, the applicability of different approaches during disruptions and their interpretability. By comparing the results of the included contributions, we can indicate that the prediction error generally increases when broadening the prediction horizon. We find that data-driven approaches might have the edge on event-driven approaches in terms of prediction accuracy, whereas event-driven approaches that explicitly model the dynamics and dependencies of railway traffic have their strength in providing interpretable predictions, and are more robust concerning disruption scenarios. The growing availability of railway operations data is expected to increase the appeal of big-data and machine learning methods.

@article{balakrishna_d-acsm_2022,
	title = {D-{ACSM}: a technique for dynamically assigning and adjusting cluster patterns for {IoT} data analysis},
	issn = {1573-0484},
	shorttitle = {D-{ACSM}},
	url = {https://doi.org/10.1007/s11227-022-04427-1},
	doi = {10.1007/s11227-022-04427-1},
	abstract = {With rapid advancements in wireless communications and sensor technologies, the Internet of Things (IoT) has advanced dramatically in past years. In IoT, the data created by a large number of sensors are extremely intricate, diverse, and enormous, and it is unprocessed. These may have underlying patterns that are not visible that must be discovered to do large-scale data analysis. Several clustering algorithms have been developed and proved effective in data analysis in recent decades; however, they are intentionally designed for dealing with static data and infeasible for processing huge data in IoT environments. As a result, this research proposes a Density-based Adaptive Cluster Split and Merge (D-ACSM) technique for dynamically assigning and changing cluster patterns for IoT data processing to solve this challenge. For successful cluster analysis, the local density and minimum distance between dynamic data objects were first measured. In addition, the D-ACSM technique used Cluster Splitting and Merging (CSM) to alter cluster patterns between surrounding dynamic data objects. In addition, the suggested D-ACSM technique’s results were evaluated using four IoT benchmarked datasets that varied in the number of arriving data objects. Finally, the proposed D-ACSM technique improves the results of the performance metrics by 4\%, 5\%, 3\%, and 6\% on the BWS-AS dataset, CRAWDAD dataset, Minute\_Weather dataset, and LinkedSensorData dataset, respectively, when compared to the AC-ICSM, IMMFC, and IAPNA techniques used for cluster analysis in all data chunks.},
	language = {en},
	urldate = {2022-03-17},
	journal = {The Journal of Supercomputing},
	author = {Balakrishna, Sivadi},
	month = mar,
	year = {2022},
}

With rapid advancements in wireless communications and sensor technologies, the Internet of Things (IoT) has advanced dramatically in past years. In IoT, the data created by a large number of sensors are extremely intricate, diverse, and enormous, and it is unprocessed. These may have underlying patterns that are not visible that must be discovered to do large-scale data analysis. Several clustering algorithms have been developed and proved effective in data analysis in recent decades; however, they are intentionally designed for dealing with static data and infeasible for processing huge data in IoT environments. As a result, this research proposes a Density-based Adaptive Cluster Split and Merge (D-ACSM) technique for dynamically assigning and changing cluster patterns for IoT data processing to solve this challenge. For successful cluster analysis, the local density and minimum distance between dynamic data objects were first measured. In addition, the D-ACSM technique used Cluster Splitting and Merging (CSM) to alter cluster patterns between surrounding dynamic data objects. In addition, the suggested D-ACSM technique’s results were evaluated using four IoT benchmarked datasets that varied in the number of arriving data objects. Finally, the proposed D-ACSM technique improves the results of the performance metrics by 4%, 5%, 3%, and 6% on the BWS-AS dataset, CRAWDAD dataset, Minute_Weather dataset, and LinkedSensorData dataset, respectively, when compared to the AC-ICSM, IMMFC, and IAPNA techniques used for cluster analysis in all data chunks.

@article{liu_aircraft_2022,
	title = {Aircraft engine remaining useful life estimation via a double attention-based data-driven architecture},
	volume = {221},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022000102},
	doi = {10.1016/j.ress.2022.108330},
	abstract = {Remaining useful life (RUL) estimation has been intensively studied, given its important role in prognostics and health management (PHM) of industry. Recently, data-driven structures such as convolutional neural networks (CNNs), have achieved outstanding RUL prediction performance. However, conventional CNNs do not include an adequate mechanism for adaptively weighing input features. In this paper, we propose a double attention-based data-driven framework for aircraft engine RUL prognostics. Specifically, a channel attention-based CNN was utilized to apply greater weights to more significant features. Next, a Transformer was used to focus attention on these features at critical time steps. We validated the effectiveness of the proposed framework on benchmark datasets for aircraft engine RUL estimation. The experimental results indicate that the proposed double attention-based architecture outperformed the existing state-of-the-art (SOTA) algorithms. The double attention-based RUL prediction method can detect the risk of equipment failure and reduce loss.},
	language = {en},
	urldate = {2022-03-15},
	journal = {Reliability Engineering \& System Safety},
	author = {Liu, Lu and Song, Xiao and Zhou, Zhetao},
	month = may,
	year = {2022},
	keywords = {Aircraft engine, Double attention, Remaining useful life estimation, Transformer network},
	pages = {108330},
}

Remaining useful life (RUL) estimation has been intensively studied, given its important role in prognostics and health management (PHM) of industry. Recently, data-driven structures such as convolutional neural networks (CNNs), have achieved outstanding RUL prediction performance. However, conventional CNNs do not include an adequate mechanism for adaptively weighing input features. In this paper, we propose a double attention-based data-driven framework for aircraft engine RUL prognostics. Specifically, a channel attention-based CNN was utilized to apply greater weights to more significant features. Next, a Transformer was used to focus attention on these features at critical time steps. We validated the effectiveness of the proposed framework on benchmark datasets for aircraft engine RUL estimation. The experimental results indicate that the proposed double attention-based architecture outperformed the existing state-of-the-art (SOTA) algorithms. The double attention-based RUL prediction method can detect the risk of equipment failure and reduce loss.

@article{de_pater_alarm-based_2022,
	title = {Alarm-based predictive maintenance scheduling for aircraft engines with imperfect {Remaining} {Useful} {Life} prognostics},
	volume = {221},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832022000175},
	doi = {10.1016/j.ress.2022.108341},
	abstract = {The increasing availability of condition monitoring data for aircraft components has incentivized the development of Remaining Useful Life (RUL) prognostics in the past years. However, only few studies consider the integration of such prognostics into maintenance planning. In this paper we propose a dynamic, predictive maintenance scheduling framework for a fleet of aircraft taking into account imperfect RUL prognostics. These prognostics are periodically updated. Based on the evolution of the prognostics over time, alarms are triggered. The scheduling of maintenance tasks is initiated only after these alarms are triggered. Alarms ensure that maintenance tasks are not rescheduled multiple times. A maintenance task is scheduled using a safety factor, to account for potential errors in the RUL prognostics and thus avoid component failures. We illustrate our approach for a fleet of 20 aircraft, each equipped with 2 turbofan engines. A Convolution Neural Network is proposed to obtain RUL prognostics. An integer linear program is used to schedule aircraft for maintenance. With our alarm-based maintenance framework, the costs with engine failures account for only 7.4\% of the total maintenance costs. In general, we provide a roadmap to integrate imperfect RUL prognostics into the maintenance planning of a fleet of vehicles.},
	language = {en},
	urldate = {2022-03-15},
	journal = {Reliability Engineering \& System Safety},
	author = {de Pater, Ingeborg and Reijns, Arthur and Mitici, Mihaela},
	month = may,
	year = {2022},
	keywords = {Aircraft maintenance, Fleet of aircraft, Predictive maintenance planning, RUL prognostics, Turbofan engines},
	pages = {108341},
}

The increasing availability of condition monitoring data for aircraft components has incentivized the development of Remaining Useful Life (RUL) prognostics in the past years. However, only few studies consider the integration of such prognostics into maintenance planning. In this paper we propose a dynamic, predictive maintenance scheduling framework for a fleet of aircraft taking into account imperfect RUL prognostics. These prognostics are periodically updated. Based on the evolution of the prognostics over time, alarms are triggered. The scheduling of maintenance tasks is initiated only after these alarms are triggered. Alarms ensure that maintenance tasks are not rescheduled multiple times. A maintenance task is scheduled using a safety factor, to account for potential errors in the RUL prognostics and thus avoid component failures. We illustrate our approach for a fleet of 20 aircraft, each equipped with 2 turbofan engines. A Convolution Neural Network is proposed to obtain RUL prognostics. An integer linear program is used to schedule aircraft for maintenance. With our alarm-based maintenance framework, the costs with engine failures account for only 7.4% of the total maintenance costs. In general, we provide a roadmap to integrate imperfect RUL prognostics into the maintenance planning of a fleet of vehicles.

@article{lara-benitez_data_2022,
	title = {Data streams classification using deep learning under different speeds and drifts},
	issn = {1367-0751},
	url = {https://doi.org/10.1093/jigpal/jzac033},
	doi = {10.1093/jigpal/jzac033},
	abstract = {Processing data streams arriving at high speed requires the development of models that can provide fast and accurate predictions. Although deep neural networks are the state-of-the-art for many machine learning tasks, their performance in real-time data streaming scenarios is a research area that has not yet been fully addressed. Nevertheless, much effort has been put into the adaption of complex deep learning (DL) models to streaming tasks by reducing the processing time. The design of the asynchronous dual-pipeline DL framework allows making predictions of incoming instances and updating the model simultaneously, using two separate layers. The aim of this work is to assess the performance of different types of DL architectures for data streaming classification using this framework. We evaluate models such as multi-layer perceptrons, recurrent, convolutional and temporal convolutional neural networks over several time series datasets that are simulated as streams at different speeds. In addition, we evaluate how the different architectures react to concept drifts typically found in evolving data streams. The obtained results indicate that convolutional architectures achieve a higher performance in terms of accuracy and efficiency, but are also the most sensitive to concept drifts.},
	urldate = {2022-03-14},
	journal = {Logic Journal of the IGPL},
	author = {Lara-Benítez, Pedro and Carranza-García, Manuel and Gutiérrez-Avilés, David and Riquelme, José C},
	month = feb,
	year = {2022},
	pages = {jzac033},
}

Processing data streams arriving at high speed requires the development of models that can provide fast and accurate predictions. Although deep neural networks are the state-of-the-art for many machine learning tasks, their performance in real-time data streaming scenarios is a research area that has not yet been fully addressed. Nevertheless, much effort has been put into the adaption of complex deep learning (DL) models to streaming tasks by reducing the processing time. The design of the asynchronous dual-pipeline DL framework allows making predictions of incoming instances and updating the model simultaneously, using two separate layers. The aim of this work is to assess the performance of different types of DL architectures for data streaming classification using this framework. We evaluate models such as multi-layer perceptrons, recurrent, convolutional and temporal convolutional neural networks over several time series datasets that are simulated as streams at different speeds. In addition, we evaluate how the different architectures react to concept drifts typically found in evolving data streams. The obtained results indicate that convolutional architectures achieve a higher performance in terms of accuracy and efficiency, but are also the most sensitive to concept drifts.

@article{chen_physics-informed_2022,
	title = {Physics-{Informed} {LSTM} hyperparameters selection for gearbox fault detection},
	volume = {171},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327022000942},
	doi = {10.1016/j.ymssp.2022.108907},
	abstract = {A situation often encountered in the condition monitoring (CM) and health management of gearboxes is that a large volume of CM data (e.g., vibration signal) collected from a healthy state is available but CM data from a faulty state unavailable. Fault detection under such a situation is usually tackled by modeling the baseline CM data and then detect the fault by examining any deviation of the baseline model versus newly monitored data. Given that the CM data is mostly time series, the long-short term memory (LSTM) neural network can be employed for baseline CM data modeling. The LSTM is free from the choice of the number of lagged input time series and can also store both long-term and short-term time series dependency information. However, we found that an LSTM with its hyperparameters selected whilst minimizing validation mean squared error (VAMSE) does not differentiate the faulty and healthy states well. There is still room for detectability improvement. In this paper, we propose a physics-informed hyperparameters selection strategy for the LSTM identification and subsequently the fault detection of gearboxes. The key idea of the proposed strategy is to select hyperparameters based on maximizing the discrepancy between healthy and physics-informed faulty states, as opposed to minimizing VAMSE. Case studies have been conducted to detect the gear tooth crack and tooth wear using laboratory test rigs. Results have shown that the proposed physics-informed hyperparameters selection strategy returns an LSTM that can better detect these faults than the LSTM returned from minimizing VAMSE.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Mechanical Systems and Signal Processing},
	author = {Chen, Yuejian and Rao, Meng and Feng, Ke and Zuo, Ming J.},
	month = may,
	year = {2022},
	keywords = {Fault Detection, Gearbox, Long-Short Term Memory, Physics-Informed Hyperparameters Selection},
	pages = {108907},
}

A situation often encountered in the condition monitoring (CM) and health management of gearboxes is that a large volume of CM data (e.g., vibration signal) collected from a healthy state is available but CM data from a faulty state unavailable. Fault detection under such a situation is usually tackled by modeling the baseline CM data and then detect the fault by examining any deviation of the baseline model versus newly monitored data. Given that the CM data is mostly time series, the long-short term memory (LSTM) neural network can be employed for baseline CM data modeling. The LSTM is free from the choice of the number of lagged input time series and can also store both long-term and short-term time series dependency information. However, we found that an LSTM with its hyperparameters selected whilst minimizing validation mean squared error (VAMSE) does not differentiate the faulty and healthy states well. There is still room for detectability improvement. In this paper, we propose a physics-informed hyperparameters selection strategy for the LSTM identification and subsequently the fault detection of gearboxes. The key idea of the proposed strategy is to select hyperparameters based on maximizing the discrepancy between healthy and physics-informed faulty states, as opposed to minimizing VAMSE. Case studies have been conducted to detect the gear tooth crack and tooth wear using laboratory test rigs. Results have shown that the proposed physics-informed hyperparameters selection strategy returns an LSTM that can better detect these faults than the LSTM returned from minimizing VAMSE.

@article{wu_condition-based_2022,
	series = {The 29th {CIRP} {Conference} on {Life} {Cycle} {Engineering}, {April} 4 – 6, 2022, {Leuven}, {Belgium}.},
	title = {Condition-{Based} {Monitoring} and {Novel} {Fault} {Detection} {Based} on {Incremental} {Learning} {Applied} to {Rotary} {Systems}},
	volume = {105},
	issn = {2212-8271},
	url = {https://www.sciencedirect.com/science/article/pii/S2212827122001329},
	doi = {10.1016/j.procir.2022.02.131},
	abstract = {Thanks to the development of new technologies such as sensor networks and advanced computational power, the research field of condition-based monitoring (CBM) has drawn increasing attention in manufacturing. With the aim of enhancing equipment reliability, leading to a reduction in maintenance costs, one of the most crucial challenges dealing with CBM is the detection or prediction of unseen/uncharacterized event during manufacturing system operation. Therefore, the identification of novel fault conditions and learning of new patterns are believed to be an important and mandatory thrust in CBM research. In this work, a convolutional neural networks and autoencoder (CNN-AE) based incremental learning method is presented. It applies CNN-AE to identify various types and severities of faults under scenarios of previously unseen fault conditions. In this method, the spectrograms generated from raw sensory signals are acquired and labelled along with working condition information via inspection. A model composed of two sub-models for novel fault identification is then constructed and trained using the time-frequency spectrograms of available conditions of operation from a rotary system. One sub-model based on CNN is for known fault classification and the other sub-model based on AE is for novelty detection, where the two sub-models share an architecture for improving efficiency and accuracy. Finally, incremental learning is performed to retrain the model with the data identified as a novel fault condition. The performance of this method is validated via an experimental case study conducted on a fault machinery simulator.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia CIRP},
	author = {Wu, Haiyue and Huang, Aihua and Sutherland, John W.},
	month = jan,
	year = {2022},
	keywords = {condition-based monitoring, incremental learning, novel fault detection, rotary system},
	pages = {788--793},
}

Thanks to the development of new technologies such as sensor networks and advanced computational power, the research field of condition-based monitoring (CBM) has drawn increasing attention in manufacturing. With the aim of enhancing equipment reliability, leading to a reduction in maintenance costs, one of the most crucial challenges dealing with CBM is the detection or prediction of unseen/uncharacterized event during manufacturing system operation. Therefore, the identification of novel fault conditions and learning of new patterns are believed to be an important and mandatory thrust in CBM research. In this work, a convolutional neural networks and autoencoder (CNN-AE) based incremental learning method is presented. It applies CNN-AE to identify various types and severities of faults under scenarios of previously unseen fault conditions. In this method, the spectrograms generated from raw sensory signals are acquired and labelled along with working condition information via inspection. A model composed of two sub-models for novel fault identification is then constructed and trained using the time-frequency spectrograms of available conditions of operation from a rotary system. One sub-model based on CNN is for known fault classification and the other sub-model based on AE is for novelty detection, where the two sub-models share an architecture for improving efficiency and accuracy. Finally, incremental learning is performed to retrain the model with the data identified as a novel fault condition. The performance of this method is validated via an experimental case study conducted on a fault machinery simulator.

@article{kharitonov_comparative_2022,
	series = {3rd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing}},
	title = {Comparative analysis of machine learning models for anomaly detection in manufacturing},
	volume = {200},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050922003398},
	doi = {10.1016/j.procs.2022.01.330},
	abstract = {The introduction of various technologies in the context of Industry 4.0 allowed collecting monitoring data for various fields in manufacturing. Shop-floor and production data can be used for further analysis to extract knowledge. In this paper, an extensive evaluation of ten Machine Learning (ML) models for anomaly detection in manufacturing is conducted. The evaluation is conducted on multiple distinct ML algorithms, including conventional ML and a representative of Deep Neural Network (DNN) based algorithms. The ML models are trained on real production schedules to detect anomalous behavior in the overall system efficiency as well as violations in the delivery date of jobs. Multiple combinations of relevant features are tested during the training of the models. In essence, the objective of the ML models is to detect anomalous unknown breakdowns in the machines that lead to disruption in the overall performance of the system. The evaluation of the ML models is conducted on independent datasets with artificially injected machine breakdowns to establish the ground truth. These schedules with machine breakdowns are obtained through a simulation model. The results point to the high performance of a conventional ML algorithm, KNN. The performance demonstrated by a DNN-based AutoEncoder, conventional Local Outlier Factor-based Feature Bagging, and a recent Copula-Based Outlier Detection (COPOD) algorithms suggest limited applicability, which strongly depends on the shape of the data.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia Computer Science},
	author = {Kharitonov, Andrey and Nahhas, Abdulrahman and Pohl, Matthias and Turowski, Klaus},
	month = jan,
	year = {2022},
	keywords = {Anomaly detection, Industry 4.0, Machine learning, Predictive Maintenance},
	pages = {1288--1297},
}

The introduction of various technologies in the context of Industry 4.0 allowed collecting monitoring data for various fields in manufacturing. Shop-floor and production data can be used for further analysis to extract knowledge. In this paper, an extensive evaluation of ten Machine Learning (ML) models for anomaly detection in manufacturing is conducted. The evaluation is conducted on multiple distinct ML algorithms, including conventional ML and a representative of Deep Neural Network (DNN) based algorithms. The ML models are trained on real production schedules to detect anomalous behavior in the overall system efficiency as well as violations in the delivery date of jobs. Multiple combinations of relevant features are tested during the training of the models. In essence, the objective of the ML models is to detect anomalous unknown breakdowns in the machines that lead to disruption in the overall performance of the system. The evaluation of the ML models is conducted on independent datasets with artificially injected machine breakdowns to establish the ground truth. These schedules with machine breakdowns are obtained through a simulation model. The results point to the high performance of a conventional ML algorithm, KNN. The performance demonstrated by a DNN-based AutoEncoder, conventional Local Outlier Factor-based Feature Bagging, and a recent Copula-Based Outlier Detection (COPOD) algorithms suggest limited applicability, which strongly depends on the shape of the data.

@article{terziyan_explainable_2022,
	series = {3rd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing}},
	title = {Explainable {AI} for {Industry} 4.0: {Semantic} {Representation} of {Deep} {Learning} {Models}},
	volume = {200},
	issn = {1877-0509},
	shorttitle = {Explainable {AI} for {Industry} 4.0},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050922002290},
	doi = {10.1016/j.procs.2022.01.220},
	abstract = {Artificial Intelligence is an important asset of Industry 4.0. Current discoveries within machine learning and particularly in deep learning enable qualitative change within the industrial processes, applications, systems and products. However, there is an important challenge related to explainability of (and, therefore, trust to) the decisions made by the deep learning models (aka black-boxes) and their poor capacity for being integrated with each other. Explainable artificial intelligence is needed instead but without loss of effectiveness of the deep learning models. In this paper we present the transformation technique between black-box models and explainable (as well as interoperable) classifiers on the basis of semantic rules via automatic recreation of the training datasets and retraining the decision trees (explainable models) in between. Our transformation technique results to explainable rule-based classifiers with good performance and efficient training process due to embedded incremental ignorance discovery and adversarial samples (“corner cases”) generation algorithms. We have also shown the use-case scenario for such explainable and interoperable classifiers, which is collaborative condition monitoring, diagnostics and predictive maintenance of distributed (and isolated) smart industrial assets while preserving data and knowledge privacy of the users.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia Computer Science},
	author = {Terziyan, Vagan and Vitko, Oleksandra},
	month = jan,
	year = {2022},
	keywords = {Explainable Artificial Intelligence, Industry 4.0, deep learning, predictive maintenance, semantic web},
	pages = {216--226},
}

Artificial Intelligence is an important asset of Industry 4.0. Current discoveries within machine learning and particularly in deep learning enable qualitative change within the industrial processes, applications, systems and products. However, there is an important challenge related to explainability of (and, therefore, trust to) the decisions made by the deep learning models (aka black-boxes) and their poor capacity for being integrated with each other. Explainable artificial intelligence is needed instead but without loss of effectiveness of the deep learning models. In this paper we present the transformation technique between black-box models and explainable (as well as interoperable) classifiers on the basis of semantic rules via automatic recreation of the training datasets and retraining the decision trees (explainable models) in between. Our transformation technique results to explainable rule-based classifiers with good performance and efficient training process due to embedded incremental ignorance discovery and adversarial samples (“corner cases”) generation algorithms. We have also shown the use-case scenario for such explainable and interoperable classifiers, which is collaborative condition monitoring, diagnostics and predictive maintenance of distributed (and isolated) smart industrial assets while preserving data and knowledge privacy of the users.

@article{wandji_characterization_2022,
	series = {The 29th {CIRP} {Conference} on {Life} {Cycle} {Engineering}, {April} 4 – 6, 2022, {Leuven}, {Belgium}.},
	title = {Characterization of the state of health of a complex system at the end of use},
	volume = {105},
	issn = {2212-8271},
	url = {https://www.sciencedirect.com/science/article/pii/S2212827122000099},
	doi = {10.1016/j.procir.2022.02.009},
	abstract = {The state of health (SoH) of an end-of-life product is one of the levers for optimizing the circular economy (CE) process in order to allow the product life-extension. Many approaches have been developed in the literature to estimate the SoH of a complex system (CS). In this study, we asked ourselves the following two questions: First, how to optimize the circular lifecycle scenarios of the components of a product at its end of life? And second, how to estimate the SoH of a product at the end of its life? To answer these questions, we proceeded as follows. First of all, the state of health of a product needs to be considered as an important parameter as well as performance or reliability. To estimate the SoH, it is necessary to identify the product parameters to be observed. The problem here is to choose the most relevant parameters among all those available for a CS. To do this, we have proposed a conditional-based maintenance approach (CBM) which consists in establishing the fault tree of a product. It consists of functionally breaking down a product into its various components and identifying the main failures for each of them. Then, these failures are used to identify the parameters to be monitored. Second, based on the most relevant parameters, the health indicators needed to estimate the SoH of the product are obtained. Then, the Prognostic and Health Management approach (PHM) is proposed in order to estimate the SoH. In the objective of providing a general solution that could be used for estimating the health status of any product, we have proposed a generic framework for the PHM approach. It serves as a guide in choosing the right approach according to the situation. Then, we proposed a decision-making strategy to optimize the process of orienting components in circular loops. This strategy is based only on the technical-functional indicator, which is the SoH of the components. Finally, we showed an example of the implementation of the proposed method for the case of the electrical scooter motor.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia CIRP},
	author = {Wandji, Christian and Rejeb, Helmi Ben and Zwolinski, Peggy},
	month = jan,
	year = {2022},
	keywords = {Health Management, Health indicator, Prognostic, Sate of health, circular economy, condition-based maintenance, lifecycle engineering},
	pages = {49--54},
}

The state of health (SoH) of an end-of-life product is one of the levers for optimizing the circular economy (CE) process in order to allow the product life-extension. Many approaches have been developed in the literature to estimate the SoH of a complex system (CS). In this study, we asked ourselves the following two questions: First, how to optimize the circular lifecycle scenarios of the components of a product at its end of life? And second, how to estimate the SoH of a product at the end of its life? To answer these questions, we proceeded as follows. First of all, the state of health of a product needs to be considered as an important parameter as well as performance or reliability. To estimate the SoH, it is necessary to identify the product parameters to be observed. The problem here is to choose the most relevant parameters among all those available for a CS. To do this, we have proposed a conditional-based maintenance approach (CBM) which consists in establishing the fault tree of a product. It consists of functionally breaking down a product into its various components and identifying the main failures for each of them. Then, these failures are used to identify the parameters to be monitored. Second, based on the most relevant parameters, the health indicators needed to estimate the SoH of the product are obtained. Then, the Prognostic and Health Management approach (PHM) is proposed in order to estimate the SoH. In the objective of providing a general solution that could be used for estimating the health status of any product, we have proposed a generic framework for the PHM approach. It serves as a guide in choosing the right approach according to the situation. Then, we proposed a decision-making strategy to optimize the process of orienting components in circular loops. This strategy is based only on the technical-functional indicator, which is the SoH of the components. Finally, we showed an example of the implementation of the proposed method for the case of the electrical scooter motor.

@article{coelho_predictive_2022,
	series = {3rd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing}},
	title = {Predictive maintenance on sensorized stamping presses by time series segmentation, anomaly detection, and classification algorithms},
	volume = {200},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050922003271},
	doi = {10.1016/j.procs.2022.01.318},
	abstract = {Sheet metal forming tools, like stamping presses, play an ubiquitous role in the manufacture of several products. With increasing requirements of quality and efficiency, ensuring maximum uptime of these tools is fundamental to marketplace competitiveness. Using anomaly detection and predictive maintenance techniques, it is possible to develop lower risk and more intelligent approaches to maintenance scheduling, however, industrial implementations of these methods remain scarce due to the difficulties of obtaining acceptable results in real-world scenarios, making applications of such techniques in stamping processes seldom found. In this work, we propose a combination of two distinct approaches: (a) time segmentation together with feature dimension reduction and anomaly detection; and (b) machine learning classification algorithms, for effective downtime prediction. The approach (a)+(b) allows for an improvement rate up to 22.971\% of the macro F1-score, when compared to sole approach (b). A ROC AUC index of 96\% is attained by using Randomized Decision Trees, being the best classifier of twelve tested. An use case with a decentralized predictive maintenance architecture for the downtime forecasting of a stamping press, which is a critical machine in the manufacturing facilities of Bosch Thermo Technology, is discussed.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia Computer Science},
	author = {Coelho, Daniel and Costa, Diogo and Rocha, Eugénio M. and Almeida, Duarte and Santos, José P.},
	month = jan,
	year = {2022},
	keywords = {Anomaly Detection, Machine Learning, Predictive Maintenance, Time Segmentation},
	pages = {1184--1193},
}

Sheet metal forming tools, like stamping presses, play an ubiquitous role in the manufacture of several products. With increasing requirements of quality and efficiency, ensuring maximum uptime of these tools is fundamental to marketplace competitiveness. Using anomaly detection and predictive maintenance techniques, it is possible to develop lower risk and more intelligent approaches to maintenance scheduling, however, industrial implementations of these methods remain scarce due to the difficulties of obtaining acceptable results in real-world scenarios, making applications of such techniques in stamping processes seldom found. In this work, we propose a combination of two distinct approaches: (a) time segmentation together with feature dimension reduction and anomaly detection; and (b) machine learning classification algorithms, for effective downtime prediction. The approach (a)+(b) allows for an improvement rate up to 22.971% of the macro F1-score, when compared to sole approach (b). A ROC AUC index of 96% is attained by using Randomized Decision Trees, being the best classifier of twelve tested. An use case with a decentralized predictive maintenance architecture for the downtime forecasting of a stamping press, which is a critical machine in the manufacturing facilities of Bosch Thermo Technology, is discussed.

@article{nikfar_two-phase_2022,
	series = {3rd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing}},
	title = {A {Two}-{Phase} {Machine} {Learning} {Approach} for {Predictive} {Maintenance} of {Low} {Voltage} {Industrial} {Motors}},
	volume = {200},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050922002198},
	doi = {10.1016/j.procs.2022.01.210},
	abstract = {Predictive maintenance and sound operating industrial equipment are essential for nearly any production plant. The absence of a systematic maintenance program and data-driven mindset in making manufacturing decisions may result in serious safety risks, unexpected equipment damages, and financial strain. Condition monitoring and predictive maintenance management systems are commonly used in tandem with the Internet of Things, linking sensors on machines and transmitting the data through a wireless network to a data-logging center that will allow further analysis and support decision making. The system described in this paper measures vibrations using sensors attached to low voltage motors and then utilizes a two-phase machine learning approach for predictive maintenance. In the first phase, we conducted an analysis to look for any abnormal behavior, and in the second phase, we attempted to determine the type of specific faults that may occur. The proposed predictive maintenance system aims to reduce the fault detection time and assist with diagnosing the type of fault occurring. We utilized and tested three machine learning algorithms to detect abnormal motor behavior: support vector machine, backpropagation neural network, and random forest. For predicting the type of specific motor faults that may occur, we used a support vector machine. This two-phase machine learning approach demonstrated promising results in detecting abnormal behavior in low voltage motors. Therefore, integrating this machine learning component as a part of a predictive maintenance system can result in high confidence about the motor condition, reduce maintenance cost, and enhance the safety of the operators and the machines.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Procedia Computer Science},
	author = {Nikfar, Mohsen and Bitencourt, Julia and Mykoniatis, Konstantinos},
	month = jan,
	year = {2022},
	keywords = {Predictive maintenance, condition monitoring, low voltage industrial motors, machine learning},
	pages = {111--120},
}

Predictive maintenance and sound operating industrial equipment are essential for nearly any production plant. The absence of a systematic maintenance program and data-driven mindset in making manufacturing decisions may result in serious safety risks, unexpected equipment damages, and financial strain. Condition monitoring and predictive maintenance management systems are commonly used in tandem with the Internet of Things, linking sensors on machines and transmitting the data through a wireless network to a data-logging center that will allow further analysis and support decision making. The system described in this paper measures vibrations using sensors attached to low voltage motors and then utilizes a two-phase machine learning approach for predictive maintenance. In the first phase, we conducted an analysis to look for any abnormal behavior, and in the second phase, we attempted to determine the type of specific faults that may occur. The proposed predictive maintenance system aims to reduce the fault detection time and assist with diagnosing the type of fault occurring. We utilized and tested three machine learning algorithms to detect abnormal motor behavior: support vector machine, backpropagation neural network, and random forest. For predicting the type of specific motor faults that may occur, we used a support vector machine. This two-phase machine learning approach demonstrated promising results in detecting abnormal behavior in low voltage motors. Therefore, integrating this machine learning component as a part of a predictive maintenance system can result in high confidence about the motor condition, reduce maintenance cost, and enhance the safety of the operators and the machines.

@article{ong_predictive_2022,
	title = {Predictive {Maintenance} {Model} for {IIoT}-based {Manufacturing}: {A} {Transferable} {Deep} {Reinforcement} {Learning} {Approach}},
	issn = {2327-4662},
	shorttitle = {Predictive {Maintenance} {Model} for {IIoT}-based {Manufacturing}},
	doi = {10.1109/JIOT.2022.3151862},
	abstract = {The Industrial Internet of Things (IIoT) is crucial for accurately assessing the state of complex equipment in order to perform predictive maintenance (PdM) successfully. However, existing IIoT-based PdM frameworks do not consider the influence of various practical yet complex system factors, such as the real-time production states, machine health, and maintenance manpower resources. For this reason, we propose a generic PdM optimization framework to assist maintenance teams in prioritizing and resolving maintenance task conflicts under real-world manufacturing conditions. Specifically, the PdM framework aims to jointly optimize edge-based machine network uptime and allocation of manpower resources in a stochastic IIoT-enabled manufacturing environment using the model-free Deep Reinforcement Learning (DRL) methods. Since DRL requires a significant amount of training data, we propose and demonstrate the use of Transfer Learning (TL) method to assist DRL in learning more efficiently by incorporating expert demonstrations, termed TL with demonstrations (TLD). TLD reduces training wall-time by 58\% compared to baseline methods, and we conduct numerous experiments to illustrate the performance, robustness, and scalability of TLD. Finally, we discuss the general benefits and limitations of the proposed TL method, which are not well addressed in the existing literature but could be beneficial to both researchers and industry practitioners.},
	journal = {IEEE Internet of Things Journal},
	author = {Ong, Kevin Shen Hoong and Wang, Wenbo and Hieu, Nguyen Quang and Niyato, Dusit and Friedrichs, Thomas},
	year = {2022},
	note = {Conference Name: IEEE Internet of Things Journal},
	keywords = {Industrial Internet of Things, Industrial Internet of Things (IIoT), Maintenance engineering, Manufacturing, Production, Resource management, Task analysis, Transfer learning, decision support, deep reinforcement learning., predictive maintenance, resource management, transfer learning},
	pages = {1--1},
}

The Industrial Internet of Things (IIoT) is crucial for accurately assessing the state of complex equipment in order to perform predictive maintenance (PdM) successfully. However, existing IIoT-based PdM frameworks do not consider the influence of various practical yet complex system factors, such as the real-time production states, machine health, and maintenance manpower resources. For this reason, we propose a generic PdM optimization framework to assist maintenance teams in prioritizing and resolving maintenance task conflicts under real-world manufacturing conditions. Specifically, the PdM framework aims to jointly optimize edge-based machine network uptime and allocation of manpower resources in a stochastic IIoT-enabled manufacturing environment using the model-free Deep Reinforcement Learning (DRL) methods. Since DRL requires a significant amount of training data, we propose and demonstrate the use of Transfer Learning (TL) method to assist DRL in learning more efficiently by incorporating expert demonstrations, termed TL with demonstrations (TLD). TLD reduces training wall-time by 58% compared to baseline methods, and we conduct numerous experiments to illustrate the performance, robustness, and scalability of TLD. Finally, we discuss the general benefits and limitations of the proposed TL method, which are not well addressed in the existing literature but could be beneficial to both researchers and industry practitioners.

@article{yao_tool_2022,
	title = {Tool remaining useful life prediction using deep transfer reinforcement learning based on long short-term memory networks},
	volume = {118},
	issn = {1433-3015},
	url = {https://doi.org/10.1007/s00170-021-07950-2},
	doi = {10.1007/s00170-021-07950-2},
	abstract = {Tool wear and faults will affect the quality of machined workpiece and damage the continuity of manufacturing. The accurate prediction of remaining useful life (RUL) is significant to guarantee the processing quality and improve the productivity of automatic system. At present, the most commonly used methods for tool RUL prediction are trained by history fault data. However, when researching on new types of tools or processing high value parts, fault datasets are difficult to acquire, which leads to RUL prediction a challenge under limited fault data. To overcome the shortcomings of above prediction methods, a deep transfer reinforcement learning (DTRL) network based on long short-term memory (LSTM) network is presented in this paper. Local features are extracted from consecutive sensor data to track the tool states, and the trained network size can be dynamically adjusted by controlling time sequence length. Then in DTRL network, LSTM network is employed to construct the value function approximation for smoothly processing temporal information and mining long-term dependencies. On this basis, a novel strategy of Q-function update and transfer is presented to transfer the deep reinforcement learning (DRL) network trained by historical fault data to a new tool for RUL prediction. Finally, tool wear experiments are performed to validate effectiveness of the DTRL model. The prediction results demonstrate that the proposed method has high accuracy and generalization for similar tools and cutting conditions.},
	language = {en},
	number = {3},
	urldate = {2022-03-03},
	journal = {The International Journal of Advanced Manufacturing Technology},
	author = {Yao, Jiachen and Lu, Baochun and Zhang, Junli},
	month = jan,
	year = {2022},
	pages = {1077--1086},
}

Tool wear and faults will affect the quality of machined workpiece and damage the continuity of manufacturing. The accurate prediction of remaining useful life (RUL) is significant to guarantee the processing quality and improve the productivity of automatic system. At present, the most commonly used methods for tool RUL prediction are trained by history fault data. However, when researching on new types of tools or processing high value parts, fault datasets are difficult to acquire, which leads to RUL prediction a challenge under limited fault data. To overcome the shortcomings of above prediction methods, a deep transfer reinforcement learning (DTRL) network based on long short-term memory (LSTM) network is presented in this paper. Local features are extracted from consecutive sensor data to track the tool states, and the trained network size can be dynamically adjusted by controlling time sequence length. Then in DTRL network, LSTM network is employed to construct the value function approximation for smoothly processing temporal information and mining long-term dependencies. On this basis, a novel strategy of Q-function update and transfer is presented to transfer the deep reinforcement learning (DRL) network trained by historical fault data to a new tool for RUL prediction. Finally, tool wear experiments are performed to validate effectiveness of the DTRL model. The prediction results demonstrate that the proposed method has high accuracy and generalization for similar tools and cutting conditions.

@article{zhang_health_2022,
	title = {Health status assessment and prediction for pumped storage units using a novel health degradation index},
	volume = {171},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327022000978},
	doi = {10.1016/j.ymssp.2022.108910},
	abstract = {To improve the safety and stability of pumped storage units (PSUs), we propose a novel health degradation index (HDI) to achieve real-time health status assessment for PSUs. Based on the HDI, the health degradation trend is predicted by the combination of variational mode decomposition (VMD) and gated recurrent unit (GRU). Firstly, the complex fitting relationship between the operating parameters of the PSU and its shaft vibration is established based on health monitoring data by multi-head self-attentive neural network (MSNN), which is regarded as a health benchmark model (HBM). MSNN reveals the high-dimensional mutually coupled relationship between different factors influencing the vibration. Secondly, to well describe the uncertainty information inherent in the vibration, Gaussian cloud model (GCM) is used to describe the vibration from both quantitative and qualitative perspectives. Then, the HDI is defined by the Kullback-Leibler divergence between the observed GCM and the health GCM predicted by HBM. Finally, VMD is used to decompose the complex HDI series into some simplistic components, while GRU is used to predict separately on the components. The final results are obtained by combining the component prediction results. The proposed method is applied in a PSU in China. The experimental results as well as several comparative studies demonstrate its outstanding performance.},
	language = {en},
	urldate = {2022-02-15},
	journal = {Mechanical Systems and Signal Processing},
	author = {Zhang, Xiaoyuan and Jiang, Yajun and Li, Chaoshun and Zhang, Jinhao},
	month = may,
	year = {2022},
	keywords = {Gated recurrent units, Gaussian cloud model, Health status assessment and prediction, Multi-head self-attentive mechanism, Pumped storage units, Variational mode decomposition},
	pages = {108910},
}

To improve the safety and stability of pumped storage units (PSUs), we propose a novel health degradation index (HDI) to achieve real-time health status assessment for PSUs. Based on the HDI, the health degradation trend is predicted by the combination of variational mode decomposition (VMD) and gated recurrent unit (GRU). Firstly, the complex fitting relationship between the operating parameters of the PSU and its shaft vibration is established based on health monitoring data by multi-head self-attentive neural network (MSNN), which is regarded as a health benchmark model (HBM). MSNN reveals the high-dimensional mutually coupled relationship between different factors influencing the vibration. Secondly, to well describe the uncertainty information inherent in the vibration, Gaussian cloud model (GCM) is used to describe the vibration from both quantitative and qualitative perspectives. Then, the HDI is defined by the Kullback-Leibler divergence between the observed GCM and the health GCM predicted by HBM. Finally, VMD is used to decompose the complex HDI series into some simplistic components, while GRU is used to predict separately on the components. The final results are obtained by combining the component prediction results. The proposed method is applied in a PSU in China. The experimental results as well as several comparative studies demonstrate its outstanding performance.

@article{velasco-gallego_real-time_2022,
	title = {A real-time data-driven framework for the identification of steady states of marine machinery},
	volume = {121},
	issn = {0141-1187},
	url = {https://www.sciencedirect.com/science/article/pii/S0141118722000086},
	doi = {10.1016/j.apor.2022.103052},
	abstract = {While maritime transportation is the primary means of long-haul transportation of goods to and from the EU, it continues to present a significant number of casualties and fatalities owing to damage to ship equipment; damage attributed to machinery failures during daily ship operations. Therefore, the implementation of state-of-the-art inspection and maintenance activities are of paramount importance to adequately ensure the proper functioning of systems. Accordingly, Internet of Ships paradigm has emerged to guarantee the interconnectivity of maritime objects. Such technology is still in its infancy, and thus several challenges need to be addressed. An example of which is data preparation, critical to ensure data quality while avoiding biased results in further analysis to enhance transportation operations. As part of developing a real-time intelligent system to assist with instant decision-making strategies that enhance ship and systems availability, operability, and profitability, a data-driven framework for the identification of steady states of marine machinery based on image generation and connected component analysis is proposed. The identification of such states is of preeminent importance, as non-operational states may adversely alter the results outlined. A case study of three diesel generators of a tanker ship is introduced to validate the developed framework. Results of this study demonstrated the outperformance of the proposed model in relation to the widely implemented clustering models k-means and GMMs with EM algorithm. As such, the proposed framework can adequately identify steady states appropriately to guarantee the detection of such states in real-time, whilst ensuring computational efficiency and model effectiveness.},
	language = {en},
	urldate = {2022-02-15},
	journal = {Applied Ocean Research},
	author = {Velasco-Gallego, Christian and Lazakis, Iraklis},
	month = apr,
	year = {2022},
	keywords = {Connected component analysis, Marine machinery, Maritime transportation, Markov chains, Smart maintenance, Steady states identification},
	pages = {103052},
}

While maritime transportation is the primary means of long-haul transportation of goods to and from the EU, it continues to present a significant number of casualties and fatalities owing to damage to ship equipment; damage attributed to machinery failures during daily ship operations. Therefore, the implementation of state-of-the-art inspection and maintenance activities are of paramount importance to adequately ensure the proper functioning of systems. Accordingly, Internet of Ships paradigm has emerged to guarantee the interconnectivity of maritime objects. Such technology is still in its infancy, and thus several challenges need to be addressed. An example of which is data preparation, critical to ensure data quality while avoiding biased results in further analysis to enhance transportation operations. As part of developing a real-time intelligent system to assist with instant decision-making strategies that enhance ship and systems availability, operability, and profitability, a data-driven framework for the identification of steady states of marine machinery based on image generation and connected component analysis is proposed. The identification of such states is of preeminent importance, as non-operational states may adversely alter the results outlined. A case study of three diesel generators of a tanker ship is introduced to validate the developed framework. Results of this study demonstrated the outperformance of the proposed model in relation to the widely implemented clustering models k-means and GMMs with EM algorithm. As such, the proposed framework can adequately identify steady states appropriately to guarantee the detection of such states in real-time, whilst ensuring computational efficiency and model effectiveness.

@article{zhao_bayesian_2022,
	title = {A {Bayesian} approach to comparing human reliability analysis methods using human performance data},
	volume = {219},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021006918},
	doi = {10.1016/j.ress.2021.108213},
	abstract = {Various methods for human reliability analysis have been developed, but a rigorous approach to quantitatively comparing these methods is still lacking. This research proposed a Bayesian approach with an attempt to address this problem. The Bayesian approach is based on ensemble modeling, which outputs the weighted average of the human error probability predictions by the human reliability analysis methods to be compared. Before incorporating any human performance data, the weights in the ensemble model represent the prior probabilities of or one’s prior beliefs in the human reliability analysis methods. Using human performance data, the weights can be updated based on Bayes’ rule to reflect one’s updated beliefs in the human reliability analysis methods. The ensemble model with updated weights itself can be further used as a posterior predictive model for human reliability. The proposed approach is demonstrated using the human performance data collected from the international human reliability analysis empirical study. The results show that the posterior beliefs vary with the data set used in the analysis. Future research using a larger human performance data set is expected to reach more conclusive comparisons.},
	language = {en},
	urldate = {2021-12-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhao, Yunfei},
	month = mar,
	year = {2022},
	keywords = {Analyst-to-analyst variability, Bayesian analysis, Ensemble modeling, Human error probability, Human performance data, Human reliability analysis, Quantitative comparison},
	pages = {108213},
}

Various methods for human reliability analysis have been developed, but a rigorous approach to quantitatively comparing these methods is still lacking. This research proposed a Bayesian approach with an attempt to address this problem. The Bayesian approach is based on ensemble modeling, which outputs the weighted average of the human error probability predictions by the human reliability analysis methods to be compared. Before incorporating any human performance data, the weights in the ensemble model represent the prior probabilities of or one’s prior beliefs in the human reliability analysis methods. Using human performance data, the weights can be updated based on Bayes’ rule to reflect one’s updated beliefs in the human reliability analysis methods. The ensemble model with updated weights itself can be further used as a posterior predictive model for human reliability. The proposed approach is demonstrated using the human performance data collected from the international human reliability analysis empirical study. The results show that the posterior beliefs vary with the data set used in the analysis. Future research using a larger human performance data set is expected to reach more conclusive comparisons.

@article{yin_quantitative_2022,
	title = {Quantitative analysis for resilience-based urban rail systems: {A} hybrid knowledge-based and data-driven approach},
	volume = {219},
	issn = {0951-8320},
	shorttitle = {Quantitative analysis for resilience-based urban rail systems},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021006670},
	doi = {10.1016/j.ress.2021.108183},
	abstract = {The rapid expansions of urban rail networks are faced with the growing number of disruptions caused by the complex rail signaling systems, incorrect driving behaviors, and extreme weather. Since urban rail systems are inherently complex and many of these disruptions are usually uncertain and inevitable, the rail managers have gradually paid more attention to the ability to withstand and quickly recover. Nevertheless, only a small number of recent developments have tried to address the ability of an urban rail system to recover from disruptions while considering the inherent structures. In this work, we propose a hybrid knowledge-based and data-driven approach for quantitative analysis of resilience. The aim is to model the causal relationships to quantify the importance of different perturbations to the overall resilience criteria. A set of key features related to the risk assessment and system resilience are summarized according to the historical data in Beijing Metro. Then, we develop a training procedure based on the structure of BN and historical data. Finally, we embed this hybrid approach into software that is applied to Beijing Metro. The results demonstrate the quantitative relationships between system resilience and different types of events.},
	language = {en},
	urldate = {2021-12-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Yin, Jiateng and Ren, Xianliang and Liu, Ronghui and Tang, Tao and Su, Shuai},
	month = mar,
	year = {2022},
	keywords = {Bayesian network, Quantitative, Resilience, Transportation, Urban rail systems},
	pages = {108183},
}

The rapid expansions of urban rail networks are faced with the growing number of disruptions caused by the complex rail signaling systems, incorrect driving behaviors, and extreme weather. Since urban rail systems are inherently complex and many of these disruptions are usually uncertain and inevitable, the rail managers have gradually paid more attention to the ability to withstand and quickly recover. Nevertheless, only a small number of recent developments have tried to address the ability of an urban rail system to recover from disruptions while considering the inherent structures. In this work, we propose a hybrid knowledge-based and data-driven approach for quantitative analysis of resilience. The aim is to model the causal relationships to quantify the importance of different perturbations to the overall resilience criteria. A set of key features related to the risk assessment and system resilience are summarized according to the historical data in Beijing Metro. Then, we develop a training procedure based on the structure of BN and historical data. Finally, we embed this hybrid approach into software that is applied to Beijing Metro. The results demonstrate the quantitative relationships between system resilience and different types of events.

@article{tornyeviadzi_systematic_2022,
	title = {A systematic framework for dynamic nodal vulnerability assessment of water distribution networks based on multilayer networks},
	volume = {219},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021006955},
	doi = {10.1016/j.ress.2021.108217},
	abstract = {Nodal demands vary throughout the day, as such any vulnerability analysis based on static networks, which considers daily average demands cannot realistically represent the criticality of nodes in the network. This study presents a systematic framework, which couples multilayer networks, structural reducibility and a Demand Adjusted Vulnerability Measure for dynamic nodal vulnerability assessment of water distribution networks (WDNs) under extended period simulation. Within this framework, we present the novel idea of characterizing the dynamics of WDNs with multi-slice networks, which captures the state of the network within a predefined temporal window taking into consideration the directional flow in pipes and the operational status of pumps, valves etc. Using a benchmark WDN, Net 3, as a case study we have demonstrated the importance of demand variations and operational status of various components, no matter how minuscule their operational time, on nodal vulnerability assessment in WDNs. The results indicated that the framework evaluates the criticality of all types of nodes, even intermediary nodes with zero base demand, within any temporal window much more realistically than conventional vulnerability analysis methods based on single (static) networks. Structural reducibility unearthed correlations between the operational status of source nodes and pumps on the general dynamics of the distribution system. The multilayer framework opens a new frontier in vulnerability analysis of WDNs and could serve as a tool for stakeholders in accessing node criticality, impact of various failure scenarios and optimal scheduling of maintenance routines.},
	language = {en},
	urldate = {2021-12-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Tornyeviadzi, Hoese Michel and Owusu-Ansah, Emmanuel and Mohammed, Hadi and Seidu, Razak},
	month = mar,
	year = {2022},
	keywords = {Dynamic vulnerability assessment model, Multilayer networks, Node vulnerability, Structural reducibility, Water distribution networks},
	pages = {108217},
}

Nodal demands vary throughout the day, as such any vulnerability analysis based on static networks, which considers daily average demands cannot realistically represent the criticality of nodes in the network. This study presents a systematic framework, which couples multilayer networks, structural reducibility and a Demand Adjusted Vulnerability Measure for dynamic nodal vulnerability assessment of water distribution networks (WDNs) under extended period simulation. Within this framework, we present the novel idea of characterizing the dynamics of WDNs with multi-slice networks, which captures the state of the network within a predefined temporal window taking into consideration the directional flow in pipes and the operational status of pumps, valves etc. Using a benchmark WDN, Net 3, as a case study we have demonstrated the importance of demand variations and operational status of various components, no matter how minuscule their operational time, on nodal vulnerability assessment in WDNs. The results indicated that the framework evaluates the criticality of all types of nodes, even intermediary nodes with zero base demand, within any temporal window much more realistically than conventional vulnerability analysis methods based on single (static) networks. Structural reducibility unearthed correlations between the operational status of source nodes and pumps on the general dynamics of the distribution system. The multilayer framework opens a new frontier in vulnerability analysis of WDNs and could serve as a tool for stakeholders in accessing node criticality, impact of various failure scenarios and optimal scheduling of maintenance routines.

@article{saraygord_afshari_machine_2022,
	title = {Machine learning-based methods in structural reliability analysis: {A} review},
	volume = {219},
	issn = {0951-8320},
	shorttitle = {Machine learning-based methods in structural reliability analysis},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021007018},
	doi = {10.1016/j.ress.2021.108223},
	abstract = {Structural Reliability analysis (SRA) is one of the prominent fields in civil and mechanical engineering. However, an accurate SRA in most cases deals with complex and costly numerical problems. Machine learning-based (ML) techniques have been introduced to the SRA problems to deal with this huge computational cost and increase accuracy. This paper presents a review of the development and use of ML models in SRA. The review includes the most common types of ML methods used in SRA. More specifically, the application of artificial neural networks (ANN), support vector machines (SVM), Bayesian methods and Kriging estimation with active learning perspective in SRA are explained, and a state-of-the-art review of the prominent literature in these fields is presented. Aiming towards a fast and accurate SRA, the ML techniques adopted for the approximation of the limit state function with Monte Carlo simulation (MCS), first/second-order reliability methods (FORM/SORM) or MCS with importance sampling well as the methods for efficiently computing the probabilities of rare events in complex structural systems. In this regard, the focus of the current manuscript is on the different models’ structures and diverse applications of each ML method in different aspects of SRA. Moreover, imperative considerations on the management of samples in the Monte Carlo simulation for SRA purposes and the treatment of the SRA problem as pattern recognition or classification task are provided. This review helps the researchers in civil and mechanical engineering, especially those who are focused on reliability and structural analysis or dealing with product assurance problems.},
	language = {en},
	urldate = {2021-12-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Saraygord Afshari, Sajad and Enayatollahi, Fatemeh and Xu, Xiangyang and Liang, Xihui},
	month = mar,
	year = {2022},
	keywords = {Artificial neural networks, Bayesian analysis, Kriging estimation, Monte carlo simulation, Response surface method, Structural reliability, Support vector machines, Surrogate modeling},
	pages = {108223},
}

Structural Reliability analysis (SRA) is one of the prominent fields in civil and mechanical engineering. However, an accurate SRA in most cases deals with complex and costly numerical problems. Machine learning-based (ML) techniques have been introduced to the SRA problems to deal with this huge computational cost and increase accuracy. This paper presents a review of the development and use of ML models in SRA. The review includes the most common types of ML methods used in SRA. More specifically, the application of artificial neural networks (ANN), support vector machines (SVM), Bayesian methods and Kriging estimation with active learning perspective in SRA are explained, and a state-of-the-art review of the prominent literature in these fields is presented. Aiming towards a fast and accurate SRA, the ML techniques adopted for the approximation of the limit state function with Monte Carlo simulation (MCS), first/second-order reliability methods (FORM/SORM) or MCS with importance sampling well as the methods for efficiently computing the probabilities of rare events in complex structural systems. In this regard, the focus of the current manuscript is on the different models’ structures and diverse applications of each ML method in different aspects of SRA. Moreover, imperative considerations on the management of samples in the Monte Carlo simulation for SRA purposes and the treatment of the SRA problem as pattern recognition or classification task are provided. This review helps the researchers in civil and mechanical engineering, especially those who are focused on reliability and structural analysis or dealing with product assurance problems.

@article{hesabi_deep_2022,
	title = {A deep learning predictive model for selective maintenance optimization},
	volume = {219},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S095183202100675X},
	doi = {10.1016/j.ress.2021.108191},
	abstract = {This paper develops a predictive selective maintenance framework using deep learning and mathematical programming. We consider a multi-component system executing consecutive production missions with scheduled intermission maintenance breaks. During the intermission breaks, several maintenance actions can improve each component's remaining useful life at a given cost. An optimization model is developed to identify a subset of maintenance actions to perform on the components. The objective is to minimize the total cost under intermission break time limitation. The total cost is composed of maintenance and failure costs; it depends on the success probabilities of the subsequent missions. To estimate these probabilities, the optimization model interacts with a long short-term memory network. The resulting predictive selective maintenance framework is validated using a benchmarking data set provided by NASA for a Modular Aero-Propulsion System Simulation of a Commercial Turbofan Engine. Its performance is highlighted when compared with the model-based approach. The results illustrate the advantages of the predictive selective maintenance framework to predict the health condition of each component with accuracy and deal with the selective maintenance of series systems.},
	language = {en},
	urldate = {2021-12-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Hesabi, Hadis and Nourelfath, Mustapha and Hajji, Adnène},
	month = mar,
	year = {2022},
	keywords = {Deep learning, Optimization, Predictive models, Remaining useful life, Selective maintenance},
	pages = {108191},
}

This paper develops a predictive selective maintenance framework using deep learning and mathematical programming. We consider a multi-component system executing consecutive production missions with scheduled intermission maintenance breaks. During the intermission breaks, several maintenance actions can improve each component's remaining useful life at a given cost. An optimization model is developed to identify a subset of maintenance actions to perform on the components. The objective is to minimize the total cost under intermission break time limitation. The total cost is composed of maintenance and failure costs; it depends on the success probabilities of the subsequent missions. To estimate these probabilities, the optimization model interacts with a long short-term memory network. The resulting predictive selective maintenance framework is validated using a benchmarking data set provided by NASA for a Modular Aero-Propulsion System Simulation of a Commercial Turbofan Engine. Its performance is highlighted when compared with the model-based approach. The results illustrate the advantages of the predictive selective maintenance framework to predict the health condition of each component with accuracy and deal with the selective maintenance of series systems.

@article{he_semi-supervised_2022,
	title = {A semi-supervised {GAN} method for {RUL} prediction using failure and suspension histories},
	volume = {168},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327021009833},
	doi = {10.1016/j.ymssp.2021.108657},
	abstract = {Deep learning methods have shown great potential to provide reliable remaining useful life (RUL) predictions in Prognostics and Health Management applications. However, deep learning models, particularly supervised learning methods, are strongly dependent on a large number of failure histories. In practice, engineering assets are generally replaced by new ones before failure during planned maintenance, resulting in a small number of failure histories and often times more than twice as many suspension histories. In this paper, a semi-supervised generative adversarial network (GAN) regression model is developed to consider both failure and suspension histories for RUL predictions. The proposed GAN model utilizes conditional multi-task objective functions to capture useful information from suspension histories to improve prediction accuracy, instead of simply treating them as unlabeled data. The method will not directly predict the failure times of suspension histories, but match the statistical information between similar failure and suspension histories to the greatest extent for model training. As a result, the failure information of suspension histories will not only rely on the failure histories but also on the generated data, thereby improving the model generalization, especially when the amount of data is limited. In addition, a robustness evaluation method is proposed to assess the uncertainty of the prognostic model caused by the scarce failure data. The accuracy and credibility of the proposed approach are validated by using two case studies.},
	language = {en},
	urldate = {2021-12-09},
	journal = {Mechanical Systems and Signal Processing},
	author = {He, Rui and Tian, Zhigang and Zuo, Ming J.},
	month = apr,
	year = {2022},
	keywords = {Generative adversarial network, Prediction, Remaining useful life, Semi-supervised learning, Suspension history},
	pages = {108657},
}

Deep learning methods have shown great potential to provide reliable remaining useful life (RUL) predictions in Prognostics and Health Management applications. However, deep learning models, particularly supervised learning methods, are strongly dependent on a large number of failure histories. In practice, engineering assets are generally replaced by new ones before failure during planned maintenance, resulting in a small number of failure histories and often times more than twice as many suspension histories. In this paper, a semi-supervised generative adversarial network (GAN) regression model is developed to consider both failure and suspension histories for RUL predictions. The proposed GAN model utilizes conditional multi-task objective functions to capture useful information from suspension histories to improve prediction accuracy, instead of simply treating them as unlabeled data. The method will not directly predict the failure times of suspension histories, but match the statistical information between similar failure and suspension histories to the greatest extent for model training. As a result, the failure information of suspension histories will not only rely on the failure histories but also on the generated data, thereby improving the model generalization, especially when the amount of data is limited. In addition, a robustness evaluation method is proposed to assess the uncertainty of the prognostic model caused by the scarce failure data. The accuracy and credibility of the proposed approach are validated by using two case studies.

@article{li_emerging_2022,
	title = {The emerging graph neural networks for intelligent fault diagnostics and prognostics: {A} guideline and a benchmark study},
	volume = {168},
	issn = {0888-3270},
	shorttitle = {The emerging graph neural networks for intelligent fault diagnostics and prognostics},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327021009791},
	doi = {10.1016/j.ymssp.2021.108653},
	abstract = {Deep learning (DL)-based methods have advanced the field of Prognostics and Health Management (PHM) in recent years, because of their powerful feature representation ability. The data in PHM are typically regular data represented in the Euclidean space. Nevertheless, there are an increasing number of applications that consider the relationships and interdependencies of data and represent the data in the form of graphs. Such kind of irregular data in non-Euclidean space pose a huge challenge to the existing DL-based methods, making some important operations (e.g., convolutions) easily applied to Euclidean space but difficult to model graph data in non-Euclidean space. Recently, graph neural networks (GNNs), as the emerging neural networks, have been utilized to model and analyze the graph data. However, there still lacks a guideline on leveraging GNNs for realizing intelligent fault diagnostics and prognostics. To fill this research gap, a practical guideline is proposed in this paper, and a novel intelligent fault diagnostics and prognostics framework based on GNN is established to illustrate how the proposed guideline works. In this framework, three types of graph construction methods are provided, and seven kinds of graph convolutional networks (GCNs) with four different graph pooling methods are investigated. To afford benchmark results for helping further study, a comprehensive evaluation of these models is performed on eight datasets, including six fault diagnosis datasets and two prognosis datasets. Finally, four issues related to the performance of GCNs are discussed and potential research directions are provided. The code library is available at: https://github.com/HazeDT/PHMGNNBenchmark.},
	language = {en},
	urldate = {2021-12-09},
	journal = {Mechanical Systems and Signal Processing},
	author = {Li, Tianfu and Zhou, Zheng and Li, Sinan and Sun, Chuang and Yan, Ruqiang and Chen, Xuefeng},
	month = apr,
	year = {2022},
	keywords = {Benchmark results, Graph neural networks, Intelligent fault diagnostics and prognostics, Practical guideline, Prognostics and health management},
	pages = {108653},
}

Deep learning (DL)-based methods have advanced the field of Prognostics and Health Management (PHM) in recent years, because of their powerful feature representation ability. The data in PHM are typically regular data represented in the Euclidean space. Nevertheless, there are an increasing number of applications that consider the relationships and interdependencies of data and represent the data in the form of graphs. Such kind of irregular data in non-Euclidean space pose a huge challenge to the existing DL-based methods, making some important operations (e.g., convolutions) easily applied to Euclidean space but difficult to model graph data in non-Euclidean space. Recently, graph neural networks (GNNs), as the emerging neural networks, have been utilized to model and analyze the graph data. However, there still lacks a guideline on leveraging GNNs for realizing intelligent fault diagnostics and prognostics. To fill this research gap, a practical guideline is proposed in this paper, and a novel intelligent fault diagnostics and prognostics framework based on GNN is established to illustrate how the proposed guideline works. In this framework, three types of graph construction methods are provided, and seven kinds of graph convolutional networks (GCNs) with four different graph pooling methods are investigated. To afford benchmark results for helping further study, a comprehensive evaluation of these models is performed on eight datasets, including six fault diagnosis datasets and two prognosis datasets. Finally, four issues related to the performance of GCNs are discussed and potential research directions are provided. The code library is available at: https://github.com/HazeDT/PHMGNNBenchmark.

@article{ding_self-supervised_2022,
	title = {Self-supervised pretraining via contrast learning for intelligent incipient fault detection of bearings},
	volume = {218},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021006207},
	doi = {10.1016/j.ress.2021.108126},
	abstract = {Data-driven approaches for prognostic and health management (PHM) increasingly rely on massive historical data, yet annotations are expensive and time-consuming. Learning approaches that utilize semi-labeled or unlabeled data are becoming increasingly popular. In this paper, a self-supervised pre-training via contrast learning (SSPCL) is introduced to learn discriminative representations from unlabeled bearing datasets. Specifically, the SSPCL employs momentum contrast learning (MCL) to investigate the local representation in terms of instance-level discrimination contrast. Further, we propose a specific architecture for SSPCL deployment on bearing vibration signals by presenting several data augmentations for 1D sequences. On this basis, we put forward an incipient fault detection method based on SSPCL for run-to-failure cycle of rolling bearings. This approach transfers the SSPCL pre-trained model to a specific semi-supervised downstream task, effectively utilizing all unlabeled data and relying on only a little priori knowledge. A case study on FEMTO-ST datasets shows that the fine-tuned model is competent for incipient fault detection, outperforming other state-of-the-art methods. Furthermore, a supplemental case on a self-built fault datasets further demonstrate the great potential and superiority of our proposed SSPCL method in PHM.},
	language = {en},
	urldate = {2021-11-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Ding, Yifei and Zhuang, Jichao and Ding, Peng and Jia, Minping},
	month = feb,
	year = {2022},
	keywords = {Data augmentation, Fault diagnosis, Incipient fault detection, Prognostic and health management, Self-supervised pretraining, Unsupervised learning},
	pages = {108126},
}

Data-driven approaches for prognostic and health management (PHM) increasingly rely on massive historical data, yet annotations are expensive and time-consuming. Learning approaches that utilize semi-labeled or unlabeled data are becoming increasingly popular. In this paper, a self-supervised pre-training via contrast learning (SSPCL) is introduced to learn discriminative representations from unlabeled bearing datasets. Specifically, the SSPCL employs momentum contrast learning (MCL) to investigate the local representation in terms of instance-level discrimination contrast. Further, we propose a specific architecture for SSPCL deployment on bearing vibration signals by presenting several data augmentations for 1D sequences. On this basis, we put forward an incipient fault detection method based on SSPCL for run-to-failure cycle of rolling bearings. This approach transfers the SSPCL pre-trained model to a specific semi-supervised downstream task, effectively utilizing all unlabeled data and relying on only a little priori knowledge. A case study on FEMTO-ST datasets shows that the fine-tuned model is competent for incipient fault detection, outperforming other state-of-the-art methods. Furthermore, a supplemental case on a self-built fault datasets further demonstrate the great potential and superiority of our proposed SSPCL method in PHM.

@article{dindar_hierarchical_2022,
	title = {A hierarchical {Bayesian}-based model for hazard analysis of climate effect on failures of railway turnout components},
	volume = {218},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021006220},
	doi = {10.1016/j.ress.2021.108130},
	abstract = {There has been a considerable increase in derailment investigations, in particular at railway turnouts (RTs), as the majority of derailments lead to lengthy disruptions to the appropriate rail operation and catastrophic consequences, being potentially severely hazardous to human safety and health, as well as rail equipment. This paper investigates the impact of climates with different features across the US on the derailments to light up a scientific way for understanding importance of climatic impact. To achieve this, official derailment reports over the last five years are examined in detail. By means of geographic segmentation associated with spatial analysis, different exposure levels of various regions have been identified and implemented into a Bayesian hierarchical model using samples by the M–H algorithm. As a result, the paper reaches interesting scientific findings of climate behaviour on turnout-related component failures resulting in derailments. The findings show extreme climate patterns impact considerably the component failures of rail turnouts. Therefore, it is indicated that turnout-related failure estimates on a large-scale region with extreme cold and hot zones could be investigated when the suggested methodology of this paper is considered.},
	language = {en},
	urldate = {2021-11-29},
	journal = {Reliability Engineering \& System Safety},
	author = {Dindar, Serdar and Kaewunruen, Sakdirat and An, Min},
	month = feb,
	year = {2022},
	keywords = {Bayesian network, Climate effect, Derailment, Railway operation},
	pages = {108130},
}

There has been a considerable increase in derailment investigations, in particular at railway turnouts (RTs), as the majority of derailments lead to lengthy disruptions to the appropriate rail operation and catastrophic consequences, being potentially severely hazardous to human safety and health, as well as rail equipment. This paper investigates the impact of climates with different features across the US on the derailments to light up a scientific way for understanding importance of climatic impact. To achieve this, official derailment reports over the last five years are examined in detail. By means of geographic segmentation associated with spatial analysis, different exposure levels of various regions have been identified and implemented into a Bayesian hierarchical model using samples by the M–H algorithm. As a result, the paper reaches interesting scientific findings of climate behaviour on turnout-related component failures resulting in derailments. The findings show extreme climate patterns impact considerably the component failures of rail turnouts. Therefore, it is indicated that turnout-related failure estimates on a large-scale region with extreme cold and hot zones could be investigated when the suggested methodology of this paper is considered.

@article{zhou_maintenance_2022,
	title = {Maintenance optimisation of multicomponent systems using hierarchical coordinated reinforcement learning},
	volume = {217},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021005767},
	doi = {10.1016/j.ress.2021.108078},
	abstract = {The Markov decision process (MDP) is a widely used method to optimise the maintenance of multicomponent systems, which can provide a system-level maintenance action at each decision point to address various dependences among components. However, MDP suffers from the “curse of dimensionality” and can only process small-scale systems. This paper develops a hierarchical coordinated reinforcement learning (HCRL) algorithm to optimise the maintenance of large-scale multicomponent systems. Both parameters of agents and the coordination relationship among agents are designed based on system characteristics. Furthermore, the hierarchical structure of agents is established according to the structural importance measures of components. The effectiveness of the proposed HCRL algorithm is validated using two maintenance optimisation problems, one on a natural gas plant system and the other using a 12-component series system under dependant competing risks. Results show that the proposed HCRL outperforms methods in two recently published papers and other benchmark approaches including the new emerging deep reinforcement learning.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhou, Yifan and Li, Bangcheng and Lin, Tian Ran},
	month = jan,
	year = {2022},
	keywords = {Condition based maintenance, Coordinated reinforcement learning, Deep reinforcement learning, Hierarchical multiagent reinforcement learning},
	pages = {108078},
}

The Markov decision process (MDP) is a widely used method to optimise the maintenance of multicomponent systems, which can provide a system-level maintenance action at each decision point to address various dependences among components. However, MDP suffers from the “curse of dimensionality” and can only process small-scale systems. This paper develops a hierarchical coordinated reinforcement learning (HCRL) algorithm to optimise the maintenance of large-scale multicomponent systems. Both parameters of agents and the coordination relationship among agents are designed based on system characteristics. Furthermore, the hierarchical structure of agents is established according to the structural importance measures of components. The effectiveness of the proposed HCRL algorithm is validated using two maintenance optimisation problems, one on a natural gas plant system and the other using a 12-component series system under dependant competing risks. Results show that the proposed HCRL outperforms methods in two recently published papers and other benchmark approaches including the new emerging deep reinforcement learning.

@article{dinh_multi-level_2022,
	title = {Multi-level opportunistic predictive maintenance for multi-component systems with economic dependence and assembly/disassembly impacts},
	volume = {217},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021005573},
	doi = {10.1016/j.ress.2021.108055},
	abstract = {For maintenance optimization of multi-component systems, opportunistic maintenance has been addressed in many studies since it allows considering the advantages of dependences between components in maintenance decision-making process. In the literature, economic dependence, which implies that joint maintenance of several components can reduce the maintenance cost, has been widely studied in the framework of opportunistic maintenance. There are however very few existing studies considering the advantages of structural dependence, whereby maintenance of a component requires disassembly of other components, in maintenance optimization. To face this issue, the objective of this paper is to propose a multi-level opportunistic predictive maintenance approach considering both economic and structural dependence. In that way, the economic and structural dependences between components are firstly formulated. A degradation model considering disassembly impacts is then developed. For opportunistic maintenance decision-making, two opportunistic thresholds are introduced. When corrective/preventive maintenance occurs, the first opportunistic threshold (eRo) is defined to select non-disassembled components for opportunistic maintenance. This first opportunistic decision allows considering the economic dependence between components. In addition, the maintenance of the selected components may require disassembly of other components which could be also good candidates to be opportunistically maintained. So, the second opportunistic threshold, sRo(sRo≥eRo), is then developed to select one or several disassembled components to be opportunistically maintained. To evaluate the performance of the proposed opportunistic maintenance approach, a cost model is developed. Particle swarm optimization algorithm is then applied to find the optimal decision variables. Finally, the proposed opportunistic maintenance approach is illustrated through a conveyor system to show its feasibility and added value in maintenance optimization framework.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Reliability Engineering \& System Safety},
	author = {Dinh, Duc-Hanh and Do, Phuc and Iung, Benoit},
	month = jan,
	year = {2022},
	keywords = {Economic dependence, Maintenance optimization, Multi-component system, Opportunistic maintenance, Structural dependence},
	pages = {108055},
}

For maintenance optimization of multi-component systems, opportunistic maintenance has been addressed in many studies since it allows considering the advantages of dependences between components in maintenance decision-making process. In the literature, economic dependence, which implies that joint maintenance of several components can reduce the maintenance cost, has been widely studied in the framework of opportunistic maintenance. There are however very few existing studies considering the advantages of structural dependence, whereby maintenance of a component requires disassembly of other components, in maintenance optimization. To face this issue, the objective of this paper is to propose a multi-level opportunistic predictive maintenance approach considering both economic and structural dependence. In that way, the economic and structural dependences between components are firstly formulated. A degradation model considering disassembly impacts is then developed. For opportunistic maintenance decision-making, two opportunistic thresholds are introduced. When corrective/preventive maintenance occurs, the first opportunistic threshold (eRo) is defined to select non-disassembled components for opportunistic maintenance. This first opportunistic decision allows considering the economic dependence between components. In addition, the maintenance of the selected components may require disassembly of other components which could be also good candidates to be opportunistically maintained. So, the second opportunistic threshold, sRo(sRo≥eRo), is then developed to select one or several disassembled components to be opportunistically maintained. To evaluate the performance of the proposed opportunistic maintenance approach, a cost model is developed. Particle swarm optimization algorithm is then applied to find the optimal decision variables. Finally, the proposed opportunistic maintenance approach is illustrated through a conveyor system to show its feasibility and added value in maintenance optimization framework.

@article{jia_reliability_2022,
	title = {Reliability analysis for complex system with multi-source data integration and multi-level data transmission},
	volume = {217},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021005536},
	doi = {10.1016/j.ress.2021.108050},
	abstract = {The Bayesian theory is appealing for reliability analysis of complex system using the data incorporation. However, there are still gaps among existing methods, such as that the common method is applicable to two-level system, multi-source data for components are not considered, k-out-of-n and standby structures are not studied with series and parallel simultaneously, the reliability is usually estimated as discrete variable, etc. For these problems, a Bayesian-based method is proposed on system under multiple levels. First, multi-source data for each target are integrated in lower level by Bayesian theory to derive the posterior distribution for reliability. Next, they are transmitted to higher level through the deterministic function concerning reliability depending on system structure. Further, the transmitted data from lower level are transformed to induced prior distribution for parameters in distribution of lifetime and integrated with native data in higher level to obtain the posterior for reliability. Finally, the reliability and remaining useful lifetime with respect to times in system-level are presented after the successive information propagation. An illustrative example is given to show the application of proposed method. Together with the sensitivity analysis, it proves that this method is feasible, practical and robust.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Reliability Engineering \& System Safety},
	author = {Jia, Xiang and Guo, Bo},
	month = jan,
	year = {2022},
	keywords = {Data integration, Data transmission, Multi-level data, Multi-source data, Reliability analysis, System},
	pages = {108050},
}

The Bayesian theory is appealing for reliability analysis of complex system using the data incorporation. However, there are still gaps among existing methods, such as that the common method is applicable to two-level system, multi-source data for components are not considered, k-out-of-n and standby structures are not studied with series and parallel simultaneously, the reliability is usually estimated as discrete variable, etc. For these problems, a Bayesian-based method is proposed on system under multiple levels. First, multi-source data for each target are integrated in lower level by Bayesian theory to derive the posterior distribution for reliability. Next, they are transmitted to higher level through the deterministic function concerning reliability depending on system structure. Further, the transmitted data from lower level are transformed to induced prior distribution for parameters in distribution of lifetime and integrated with native data in higher level to obtain the posterior for reliability. Finally, the reliability and remaining useful lifetime with respect to times in system-level are presented after the successive information propagation. An illustrative example is given to show the application of proposed method. Together with the sensitivity analysis, it proves that this method is feasible, practical and robust.

@article{arias_chao_fusing_2022,
	title = {Fusing physics-based and deep learning models for prognostics},
	volume = {217},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004725},
	doi = {10.1016/j.ress.2021.107961},
	abstract = {Physics-based and data-driven models for remaining useful lifetime (RUL) prediction typically suffer from two major challenges that limit their applicability to complex real-world domains: (1) the incompleteness of physics-based models and (2) the limited representativeness of the training dataset for data-driven models. Combining the advantages of these two approaches while overcoming some of their limitations, we propose a novel hybrid framework for fusing the information from physics-based performance models with deep learning algorithms for prognostics of complex safety-critical systems. In the proposed framework, we use physics-based performance models to infer unobservable model parameters related to a system’s components health by solving a calibration problem. These parameters are subsequently combined with sensor readings and used as input to a deep neural network, thereby generating a data-driven prognostics model with physics-augmented features. The performance of the hybrid framework is evaluated on an extensive case study comprising run-to-failure degradation trajectories from a fleet of nine turbofan engines under real flight conditions. The experimental results show that the hybrid framework outperforms purely data-driven approaches by extending the prediction horizon by nearly 127\%. Furthermore, it requires less training data and is less sensitive to the limited representativeness of the dataset as compared to purely data-driven approaches. Furthermore, we demonstrated the feasibility of the proposed framework on the original CMAPSS dataset, thereby confirming its superior performance.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Reliability Engineering \& System Safety},
	author = {Arias Chao, Manuel and Kulkarni, Chetan and Goebel, Kai and Fink, Olga},
	month = jan,
	year = {2022},
	keywords = {CMAPSS, Deep learning, Hybrid model, Prognostics},
	pages = {107961},
}

Physics-based and data-driven models for remaining useful lifetime (RUL) prediction typically suffer from two major challenges that limit their applicability to complex real-world domains: (1) the incompleteness of physics-based models and (2) the limited representativeness of the training dataset for data-driven models. Combining the advantages of these two approaches while overcoming some of their limitations, we propose a novel hybrid framework for fusing the information from physics-based performance models with deep learning algorithms for prognostics of complex safety-critical systems. In the proposed framework, we use physics-based performance models to infer unobservable model parameters related to a system’s components health by solving a calibration problem. These parameters are subsequently combined with sensor readings and used as input to a deep neural network, thereby generating a data-driven prognostics model with physics-augmented features. The performance of the hybrid framework is evaluated on an extensive case study comprising run-to-failure degradation trajectories from a fleet of nine turbofan engines under real flight conditions. The experimental results show that the hybrid framework outperforms purely data-driven approaches by extending the prediction horizon by nearly 127%. Furthermore, it requires less training data and is less sensitive to the limited representativeness of the dataset as compared to purely data-driven approaches. Furthermore, we demonstrated the feasibility of the proposed framework on the original CMAPSS dataset, thereby confirming its superior performance.

@article{cao_kspmi_2022,
	title = {{KSPMI}: {A} {Knowledge}-based {System} for {Predictive} {Maintenance} in {Industry} 4.0},
	volume = {74},
	issn = {0736-5845},
	shorttitle = {{KSPMI}},
	url = {https://www.sciencedirect.com/science/article/pii/S0736584521001617},
	doi = {10.1016/j.rcim.2021.102281},
	abstract = {In the context of Industry 4.0, smart factories use advanced sensing and data analytic technologies to understand and monitor the manufacturing processes. To enhance production efficiency and reliability, statistical Artificial Intelligence (AI) technologies such as machine learning and data mining are used to detect and predict potential anomalies within manufacturing processes. However, due to the heterogeneous nature of industrial data, sometimes the knowledge extracted from industrial data is presented in a complex structure. This brings the semantic gap issue which stands for the lack of interoperability among different manufacturing systems. Furthermore, as the Cyber-Physical Systems (CPS) are becoming more knowledge-intensive, uniform knowledge representation of physical resources and real-time reasoning capabilities for analytic tasks are needed to automate the decision-making processes for these systems. These requirements highlight the potential of using symbolic AI for predictive maintenance. To automate and facilitate predictive analytics in Industry 4.0, in this paper, we present a novel Knowledge-based System for Predictive Maintenance in Industry 4.0 (KSPMI). KSPMI is developed based on a novel hybrid approach that leverages both statistical and symbolic AI technologies. The hybrid approach involves using statistical AI technologies such as machine learning and chronicle mining (a special type of sequential pattern mining approach) to extract machine degradation models from industrial data. On the other hand, symbolic AI technologies, especially domain ontologies and logic rules, will use the extracted chronicle patterns to query and reason on system input data with rich domain and contextual knowledge. This hybrid approach uses Semantic Web Rule Language (SWRL) rules generated from chronicle patterns together with domain ontologies to perform ontology reasoning, which enables the automatic detection of machinery anomalies and the prediction of future events’ occurrence. KSPMI is evaluated and tested on both real-world and synthetic data sets.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Robotics and Computer-Integrated Manufacturing},
	author = {Cao, Qiushi and Zanni-Merk, Cecilia and Samet, Ahmed and Reich, Christoph and Beuvron, François de Bertrand de and Beckmann, Arnold and Giannetti, Cinzia},
	month = apr,
	year = {2022},
	keywords = {Chronicle mining, Industry 4.0, Knowledge-based system, Ontology reasoning, Predictive maintenance},
	pages = {102281},
}

In the context of Industry 4.0, smart factories use advanced sensing and data analytic technologies to understand and monitor the manufacturing processes. To enhance production efficiency and reliability, statistical Artificial Intelligence (AI) technologies such as machine learning and data mining are used to detect and predict potential anomalies within manufacturing processes. However, due to the heterogeneous nature of industrial data, sometimes the knowledge extracted from industrial data is presented in a complex structure. This brings the semantic gap issue which stands for the lack of interoperability among different manufacturing systems. Furthermore, as the Cyber-Physical Systems (CPS) are becoming more knowledge-intensive, uniform knowledge representation of physical resources and real-time reasoning capabilities for analytic tasks are needed to automate the decision-making processes for these systems. These requirements highlight the potential of using symbolic AI for predictive maintenance. To automate and facilitate predictive analytics in Industry 4.0, in this paper, we present a novel Knowledge-based System for Predictive Maintenance in Industry 4.0 (KSPMI). KSPMI is developed based on a novel hybrid approach that leverages both statistical and symbolic AI technologies. The hybrid approach involves using statistical AI technologies such as machine learning and chronicle mining (a special type of sequential pattern mining approach) to extract machine degradation models from industrial data. On the other hand, symbolic AI technologies, especially domain ontologies and logic rules, will use the extracted chronicle patterns to query and reason on system input data with rich domain and contextual knowledge. This hybrid approach uses Semantic Web Rule Language (SWRL) rules generated from chronicle patterns together with domain ontologies to perform ontology reasoning, which enables the automatic detection of machinery anomalies and the prediction of future events’ occurrence. KSPMI is evaluated and tested on both real-world and synthetic data sets.

@article{giordano_data-driven_2022,
	title = {Data-driven strategies for predictive maintenance: {Lesson} learned from an automotive use case},
	volume = {134},
	issn = {0166-3615},
	shorttitle = {Data-driven strategies for predictive maintenance},
	url = {https://www.sciencedirect.com/science/article/pii/S0166361521001615},
	doi = {10.1016/j.compind.2021.103554},
	abstract = {Predictive maintenance is an ever-growing topic of interest, spanning different fields and approaches. In the automotive domain, thanks to on-board sensors and the possibility to transmit collected data to the cloud, car manufacturers can deploy predictive maintenance solutions to prevent components malfunctioning and eventually recall to the service the vehicle before the customer experiences the failure. In this paper we present PREPIPE, a data-driven pipeline for predictive maintenance. Given the raw time series of signals recorded by the on-board engine control unit of diesel engines, we exploit PREPIPE to predict the clogging status of the oxygen sensor, a key component of the exhaust system to control combustion efficiency and pollutant emissions. In the design of PREPIPE we deeply investigate: (i) how to choose the best subset of signals to best capture the sensor status, (ii) how much data needs to be collected to make the most accurate prediction, (iii) how to transform the original time series into features suitable for state-of-art classifiers, (iv) how to select the most important features, (v) how to include historical features to predict the clogging status of the sensor. We thoroughly assess PREPIPE performance and compare it with state-of-art deep learning architectures. Our results show that PREPIPE correctly identifies critical situations before the sensor reaches critical conditions. Furthermore, PREPIPE supports domain experts in optimizing the design of data-driven predictive maintenance pipelines with performance comparable to deep learning methodologies while keeping a degree of interpretability.},
	language = {en},
	urldate = {2021-11-15},
	journal = {Computers in Industry},
	author = {Giordano, Danilo and Giobergia, Flavio and Pastor, Eliana and La Macchia, Antonio and Cerquitelli, Tania and Baralis, Elena and Mellia, Marco and Tricarico, Davide},
	month = jan,
	year = {2022},
	keywords = {Automotive, Data-driven, Machine learning, Predictive maintenance, Time series},
	pages = {103554},
}

Predictive maintenance is an ever-growing topic of interest, spanning different fields and approaches. In the automotive domain, thanks to on-board sensors and the possibility to transmit collected data to the cloud, car manufacturers can deploy predictive maintenance solutions to prevent components malfunctioning and eventually recall to the service the vehicle before the customer experiences the failure. In this paper we present PREPIPE, a data-driven pipeline for predictive maintenance. Given the raw time series of signals recorded by the on-board engine control unit of diesel engines, we exploit PREPIPE to predict the clogging status of the oxygen sensor, a key component of the exhaust system to control combustion efficiency and pollutant emissions. In the design of PREPIPE we deeply investigate: (i) how to choose the best subset of signals to best capture the sensor status, (ii) how much data needs to be collected to make the most accurate prediction, (iii) how to transform the original time series into features suitable for state-of-art classifiers, (iv) how to select the most important features, (v) how to include historical features to predict the clogging status of the sensor. We thoroughly assess PREPIPE performance and compare it with state-of-art deep learning architectures. Our results show that PREPIPE correctly identifies critical situations before the sensor reaches critical conditions. Furthermore, PREPIPE supports domain experts in optimizing the design of data-driven predictive maintenance pipelines with performance comparable to deep learning methodologies while keeping a degree of interpretability.

@article{arena_novel_2022,
	title = {A novel decision support system for managing predictive maintenance strategies based on machine learning approaches},
	volume = {146},
	issn = {0925-7535},
	url = {https://www.sciencedirect.com/science/article/pii/S0925753521003726},
	doi = {10.1016/j.ssci.2021.105529},
	abstract = {Nowadays, the industrial environment is characterised by growing competitiveness, short response times, cost reduction and reliability of production to meet customer needs. Thus, the new industrial paradigm of Industry 4.0 has gained interest worldwide, leading many manufacturers to a significant digital transformation. Digital technologies have enabled a novel approach to decision-making processes based on data-driven strategies, where knowledge extraction relies on the analysis of a large amount of data from sensor-equipped factories. In this context, Predictive Maintenance (PdM) based on Machine Learning (ML) is one of the most prominent data-driven analytical approaches for monitoring industrial systems aiming to maximise reliability and efficiency. In fact, PdM aims not only to reduce equipment failure rates but also to minimise operating costs by maximising equipment life. When considering industrial applications, industries deal with different issues and constraints relating to process digitalisation. The main purpose of this study is to develop a new decision support system based on decision trees (DTs) that guides the decision-making process of PdM implementation, considering context-aware information, quality and maturity of collected data, severity, occurrence and detectability of potential failures (identified through FMECA analysis) and direct and indirect maintenance costs. The decision trees allow the study of different scenarios to identify the conditions under which a PdM policy, based on the ML algorithm, is economically profitable compared to corrective maintenance, considered to be the current scenario. The results show that the proposed methodology is a simple and easy way to implement tool to support the decision process by assessing the different levels of occurrence and severity of failures. For each level, savings and the potential costs have been evaluated at leaf nodes of the trees aimed at defining the most suitable maintenance strategy implementation. Finally, the proposed DTs are applied to a real industrial case to illustrate their applicability and robustness.},
	language = {en},
	urldate = {2021-10-25},
	journal = {Safety Science},
	author = {Arena, S. and Florian, E. and Zennaro, I. and Orrù, P. F. and Sgarbossa, F.},
	month = feb,
	year = {2022},
	keywords = {Decision support system, Decision tree, FMECA, Machine learning, Predictive maintenance, Profitability},
	pages = {105529},
}

Nowadays, the industrial environment is characterised by growing competitiveness, short response times, cost reduction and reliability of production to meet customer needs. Thus, the new industrial paradigm of Industry 4.0 has gained interest worldwide, leading many manufacturers to a significant digital transformation. Digital technologies have enabled a novel approach to decision-making processes based on data-driven strategies, where knowledge extraction relies on the analysis of a large amount of data from sensor-equipped factories. In this context, Predictive Maintenance (PdM) based on Machine Learning (ML) is one of the most prominent data-driven analytical approaches for monitoring industrial systems aiming to maximise reliability and efficiency. In fact, PdM aims not only to reduce equipment failure rates but also to minimise operating costs by maximising equipment life. When considering industrial applications, industries deal with different issues and constraints relating to process digitalisation. The main purpose of this study is to develop a new decision support system based on decision trees (DTs) that guides the decision-making process of PdM implementation, considering context-aware information, quality and maturity of collected data, severity, occurrence and detectability of potential failures (identified through FMECA analysis) and direct and indirect maintenance costs. The decision trees allow the study of different scenarios to identify the conditions under which a PdM policy, based on the ML algorithm, is economically profitable compared to corrective maintenance, considered to be the current scenario. The results show that the proposed methodology is a simple and easy way to implement tool to support the decision process by assessing the different levels of occurrence and severity of failures. For each level, savings and the potential costs have been evaluated at leaf nodes of the trees aimed at defining the most suitable maintenance strategy implementation. Finally, the proposed DTs are applied to a real industrial case to illustrate their applicability and robustness.

@article{davari_survey_2021,
	title = {A {Survey} on {Data}-{Driven} {Predictive} {Maintenance} for the {Railway} {Industry}},
	volume = {21},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	issn = {1424-8220},
	url = {https://www.mdpi.com/1424-8220/21/17/5739},
	doi = {10.3390/s21175739},
	abstract = {In the last few years, many works have addressed Predictive Maintenance (PdM) by the use of Machine Learning (ML) and Deep Learning (DL) solutions, especially the latter. The monitoring and logging of industrial equipment events, like temporal behavior and fault events—anomaly detection in time-series—can be obtained from records generated by sensors installed in different parts of an industrial plant. However, such progress is incipient because we still have many challenges, and the performance of applications depends on the appropriate choice of the method. This article presents a survey of existing ML and DL techniques for handling PdM in the railway industry. This survey discusses the main approaches for this specific application within a taxonomy defined by the type of task, employed methods, metrics of evaluation, the specific equipment or process, and datasets. Lastly, we conclude and outline some suggestions for future research.},
	language = {en},
	number = {17},
	urldate = {2023-10-26},
	journal = {Sensors},
	author = {Davari, Narjes and Veloso, Bruno and Costa, Gustavo de Assis and Pereira, Pedro Mota and Ribeiro, Rita P. and Gama, João},
	month = jan,
	year = {2021},
	note = {Number: 17
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {artificial intelligence, condition-based maintenance, deep learning, machine learning, predictive maintenance, railway industry},
	pages = {5739},
}

In the last few years, many works have addressed Predictive Maintenance (PdM) by the use of Machine Learning (ML) and Deep Learning (DL) solutions, especially the latter. The monitoring and logging of industrial equipment events, like temporal behavior and fault events—anomaly detection in time-series—can be obtained from records generated by sensors installed in different parts of an industrial plant. However, such progress is incipient because we still have many challenges, and the performance of applications depends on the appropriate choice of the method. This article presents a survey of existing ML and DL techniques for handling PdM in the railway industry. This survey discusses the main approaches for this specific application within a taxonomy defined by the type of task, employed methods, metrics of evaluation, the specific equipment or process, and datasets. Lastly, we conclude and outline some suggestions for future research.

@article{ji_autonomous_2021,
	title = {Autonomous underwater vehicle fault diagnosis dataset},
	volume = {39},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340921007587},
	doi = {10.1016/j.dib.2021.107477},
	abstract = {The dataset contains 1225 data samples for 5 fault types (labels). We divided the dataset into the training set and the test set through random stratified sampling. The test set accounted for 20\% of the total dataset. Our experimental subject is ‘Haizhe’, which is a small quadrotor AUV developed in the laboratory. For each fault type, ‘Haizhe’ was tested several times. For each time, ‘Haizhe’ ran the same program and sailed underwater for 10–20 s to ensure that state data was long enough. The state data recorded in each test were then used as a data sample, and the corresponding fault type was the true label of the data sample. The dataset was used to validate a model-free fault diagnosis method proposed in our paper [1] and the complete dynamic model of ‘Haizhe’ AUV was reported in [2].},
	urldate = {2023-10-04},
	journal = {Data in Brief},
	author = {Ji, Daxiong and Yao, Xin and Li, Shuo and Tang, Yuangui and Tian, Yu},
	month = dec,
	year = {2021},
	keywords = {Autonomous underwater vehicles (AUV), Fault diagnosis, Fault type, Model-free, State data},
	pages = {107477},
}

The dataset contains 1225 data samples for 5 fault types (labels). We divided the dataset into the training set and the test set through random stratified sampling. The test set accounted for 20% of the total dataset. Our experimental subject is ‘Haizhe’, which is a small quadrotor AUV developed in the laboratory. For each fault type, ‘Haizhe’ was tested several times. For each time, ‘Haizhe’ ran the same program and sailed underwater for 10–20 s to ensure that state data was long enough. The state data recorded in each test were then used as a data sample, and the corresponding fault type was the true label of the data sample. The dataset was used to validate a model-free fault diagnosis method proposed in our paper [1] and the complete dynamic model of ‘Haizhe’ AUV was reported in [2].

@article{inturi_comprehensive_2021,
	title = {Comprehensive fault diagnostics of wind turbine gearbox through adaptive condition monitoring scheme},
	volume = {174},
	issn = {0003-682X},
	url = {https://www.sciencedirect.com/science/article/pii/S0003682X20308422},
	doi = {10.1016/j.apacoust.2020.107738},
	abstract = {The current work reports a multi-level classification to envisage the location, type/category and severity level of local defects at different stages of speed in a wind turbine gearbox with minimal human intervention. Experiments are conducted by subjecting a three-stage gearbox to fluctuating speeds with multiple sensors recording the real-time information generated. Wavelet coefficients are employed to extract the statistical features from the raw signatures decomposed through wavelet transform. A decision tree algorithm is used to identify features of significance and an integrated multi-variable feature data set is devised based on feature-level data fusion. The intended multi-level classification on the integrated feature data set is accomplished with the help of machine-learning algorithms. The results reveal that the adaptive neuro-fuzzy inference system (ANFIS) performs the intended four-level classification on the wind turbine gearbox with a classification accuracy of 92\%. Thus, the integration of multi-sensor information in conjunction with ANFIS as a classification algorithm, owing to its efficiency in predicting every possible detail about the health/condition of the different gearbox components, demonstrates its potential to be used as an adaptive condition monitoring as it.},
	language = {en},
	urldate = {2023-06-07},
	journal = {Applied Acoustics},
	author = {Inturi, Vamsi and Shreyas, N. and Chetti, Karthick and Sabareesh, G. R.},
	month = mar,
	year = {2021},
	keywords = {ANFIS, Condition monitoring, Fault diagnosis, Multi-level classification, Wind turbine gearbox},
	pages = {107738},
}

The current work reports a multi-level classification to envisage the location, type/category and severity level of local defects at different stages of speed in a wind turbine gearbox with minimal human intervention. Experiments are conducted by subjecting a three-stage gearbox to fluctuating speeds with multiple sensors recording the real-time information generated. Wavelet coefficients are employed to extract the statistical features from the raw signatures decomposed through wavelet transform. A decision tree algorithm is used to identify features of significance and an integrated multi-variable feature data set is devised based on feature-level data fusion. The intended multi-level classification on the integrated feature data set is accomplished with the help of machine-learning algorithms. The results reveal that the adaptive neuro-fuzzy inference system (ANFIS) performs the intended four-level classification on the wind turbine gearbox with a classification accuracy of 92%. Thus, the integration of multi-sensor information in conjunction with ANFIS as a classification algorithm, owing to its efficiency in predicting every possible detail about the health/condition of the different gearbox components, demonstrates its potential to be used as an adaptive condition monitoring as it.

@article{mutlu_virtual_2021,
	title = {The virtual and the physical: two frames of mind},
	volume = {24},
	issn = {2589-0042},
	shorttitle = {The virtual and the physical},
	url = {https://www.sciencedirect.com/science/article/pii/S2589004220311627},
	doi = {10.1016/j.isci.2020.101965},
	abstract = {Virtual and physical embodiments of interactive artificial agents utilize similar core technologies for perception, planning, and interaction and engage with people in similar ways. Thus, designers have typically considered these embodiments to be broadly interchangeable, and the choice of embodiment primarily depends on the practical demands of an application. This paper makes the case that virtual and physical embodiments elicit fundamentally different “frames of mind” in the users of the technology and follow different metaphors for interaction, resulting in diverging expectations, forms of engagement, and eventually interaction outcomes. It illustrates these differences through the lens of five key mechanisms: “situativity, interactivity, agency, proxemics, and believability”. It also outlines the design implications of the two frames of mind, arguing for different domains of interaction serving as appropriate context for virtual and physical embodiments.},
	language = {en},
	number = {2},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Mutlu, Bilge},
	month = feb,
	year = {2021},
	keywords = {Artificial Intelligence, Cognitive Neuroscience, Psychology, Social Sciences},
	pages = {101965},
}

Virtual and physical embodiments of interactive artificial agents utilize similar core technologies for perception, planning, and interaction and engage with people in similar ways. Thus, designers have typically considered these embodiments to be broadly interchangeable, and the choice of embodiment primarily depends on the practical demands of an application. This paper makes the case that virtual and physical embodiments elicit fundamentally different “frames of mind” in the users of the technology and follow different metaphors for interaction, resulting in diverging expectations, forms of engagement, and eventually interaction outcomes. It illustrates these differences through the lens of five key mechanisms: “situativity, interactivity, agency, proxemics, and believability”. It also outlines the design implications of the two frames of mind, arguing for different domains of interaction serving as appropriate context for virtual and physical embodiments.

@article{vinanzi_collaborative_2021,
	title = {The collaborative mind: intention reading and trust in human-robot interaction},
	volume = {24},
	issn = {2589-0042},
	shorttitle = {The collaborative mind},
	url = {https://www.sciencedirect.com/science/article/pii/S2589004221000985},
	doi = {10.1016/j.isci.2021.102130},
	abstract = {Robots are likely to become important social actors in our future and so require more human-like ways of assisting us. We state that collaboration between humans and robots is fostered by two cognitive skills: intention reading and trust. An agent possessing these abilities would be able to infer the non-verbal intentions of others and to evaluate how likely they are to achieve their goals, jointly understanding what kind and which degree of collaboration they require. For this reason, we propose a developmental artificial cognitive architecture that integrates unsupervised machine learning and probabilistic models to imbue a humanoid robot with intention reading and trusting capabilities. Our experimental results show that the synergistic implementation of these cognitive skills enable the robot to cooperate in a meaningful way, with the intention reading model allowing a correct goal prediction and with the trust component enhancing the likelihood of a positive outcome for the task.},
	language = {en},
	number = {2},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Vinanzi, Samuele and Cangelosi, Angelo and Goerick, Christian},
	month = feb,
	year = {2021},
	keywords = {Artificial Intelligence, Human-Centered Computing, Human-Computer Interaction},
	pages = {102130},
}

Robots are likely to become important social actors in our future and so require more human-like ways of assisting us. We state that collaboration between humans and robots is fostered by two cognitive skills: intention reading and trust. An agent possessing these abilities would be able to infer the non-verbal intentions of others and to evaluate how likely they are to achieve their goals, jointly understanding what kind and which degree of collaboration they require. For this reason, we propose a developmental artificial cognitive architecture that integrates unsupervised machine learning and probabilistic models to imbue a humanoid robot with intention reading and trusting capabilities. Our experimental results show that the synergistic implementation of these cognitive skills enable the robot to cooperate in a meaningful way, with the intention reading model allowing a correct goal prediction and with the trust component enhancing the likelihood of a positive outcome for the task.

@article{lepri_ethical_2021,
	title = {Ethical machines: {The} human-centric use of artificial intelligence},
	volume = {24},
	issn = {2589-0042},
	shorttitle = {Ethical machines},
	url = {https://www.sciencedirect.com/science/article/pii/S2589004221002170},
	doi = {10.1016/j.isci.2021.102249},
	abstract = {Today's increased availability of large amounts of human behavioral data and advances in artificial intelligence (AI) are contributing to a growing reliance on algorithms to make consequential decisions for humans, including those related to access to credit or medical treatments, hiring, etc. Algorithmic decision-making processes might lead to more objective decisions than those made by humans who may be influenced by prejudice, conflicts of interest, or fatigue. However, algorithmic decision-making has been criticized for its potential to lead to privacy invasion, information asymmetry, opacity, and discrimination. In this paper, we describe available technical solutions in three large areas that we consider to be of critical importance to achieve a human-centric AI: (1) privacy and data ownership; (2) accountability and transparency; and (3) fairness. We also highlight the criticality and urgency to engage multi-disciplinary teams of researchers, practitioners, policy makers, and citizens to co-develop and evaluate in the real-world algorithmic decision-making processes designed to maximize fairness, accountability, and transparency while respecting privacy.},
	language = {en},
	number = {3},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Lepri, Bruno and Oliver, Nuria and Pentland, Alex},
	month = mar,
	year = {2021},
	keywords = {Algorithms, Artificial Intelligence, Computer Privacy},
	pages = {102249},
}

Today's increased availability of large amounts of human behavioral data and advances in artificial intelligence (AI) are contributing to a growing reliance on algorithms to make consequential decisions for humans, including those related to access to credit or medical treatments, hiring, etc. Algorithmic decision-making processes might lead to more objective decisions than those made by humans who may be influenced by prejudice, conflicts of interest, or fatigue. However, algorithmic decision-making has been criticized for its potential to lead to privacy invasion, information asymmetry, opacity, and discrimination. In this paper, we describe available technical solutions in three large areas that we consider to be of critical importance to achieve a human-centric AI: (1) privacy and data ownership; (2) accountability and transparency; and (3) fairness. We also highlight the criticality and urgency to engage multi-disciplinary teams of researchers, practitioners, policy makers, and citizens to co-develop and evaluate in the real-world algorithmic decision-making processes designed to maximize fairness, accountability, and transparency while respecting privacy.

@article{franklin_blaming_2021,
	title = {Blaming automated vehicles in difficult situations},
	volume = {24},
	issn = {2589-0042},
	url = {https://www.sciencedirect.com/science/article/pii/S2589004221002200},
	doi = {10.1016/j.isci.2021.102252},
	abstract = {Automated vehicles (AVs) have made huge strides toward large-scale deployment. Despite this progress, AVs continue to make mistakes, some resulting in death. Although some mistakes are avoidable, others are hard to avoid even by highly skilled drivers. As these mistakes continue to shape attitudes toward AVs, we need to understand whether people differentiate between them. We ask the following two questions. When an AV makes a mistake, does the perceived difficulty or novelty of the situation predict blame attributed to it? How does that blame attribution compare to a human driving a car? Through two studies, we find that the amount of blame people attribute to AVs and human drivers is sensitive to situation difficulty. However, while some situations could be more difficult for AVs and others for human drivers, people blamed AVs more, regardless. Our results provide novel insights in understanding psychological barriers influencing the public's view of AVs.},
	language = {en},
	number = {4},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Franklin, Matija and Awad, Edmond and Lagnado, David},
	month = apr,
	year = {2021},
	keywords = {Artificial Intelligence, Psychology, Research Methodology Social Sciences},
	pages = {102252},
}

Automated vehicles (AVs) have made huge strides toward large-scale deployment. Despite this progress, AVs continue to make mistakes, some resulting in death. Although some mistakes are avoidable, others are hard to avoid even by highly skilled drivers. As these mistakes continue to shape attitudes toward AVs, we need to understand whether people differentiate between them. We ask the following two questions. When an AV makes a mistake, does the perceived difficulty or novelty of the situation predict blame attributed to it? How does that blame attribution compare to a human driving a car? Through two studies, we find that the amount of blame people attribute to AVs and human drivers is sensitive to situation difficulty. However, while some situations could be more difficult for AVs and others for human drivers, people blamed AVs more, regardless. Our results provide novel insights in understanding psychological barriers influencing the public's view of AVs.

@article{nonaka_brain_2021,
	title = {Brain hierarchy score: {Which} deep neural networks are hierarchically brain-like?},
	volume = {24},
	issn = {2589-0042},
	shorttitle = {Brain hierarchy score},
	url = {https://www.sciencedirect.com/science/article/pii/S2589004221009810},
	doi = {10.1016/j.isci.2021.103013},
	abstract = {Achievement of human-level image recognition by deep neural networks (DNNs) has spurred interest in whether and how DNNs are brain-like. Both DNNs and the visual cortex perform hierarchical processing, and correspondence has been shown between hierarchical visual areas and DNN layers in representing visual features. Here, we propose the brain hierarchy (BH) score as a metric to quantify the degree of hierarchical correspondence based on neural decoding and encoding analyses where DNN unit activations and human brain activity are predicted from each other. We find that BH scores for 29 pre-trained DNNs with various architectures are negatively correlated with image recognition performance, thus indicating that recently developed high-performance DNNs are not necessarily brain-like. Experimental manipulations of DNN models suggest that single-path sequential feedforward architecture with broad spatial integration is critical to brain-like hierarchy. Our method may provide new ways to design DNNs in light of their representational homology to the brain.},
	language = {en},
	number = {9},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Nonaka, Soma and Majima, Kei and Aoki, Shuntaro C. and Kamitani, Yukiyasu},
	month = sep,
	year = {2021},
	keywords = {Human-centered computing, Neural networks, Neuroscience},
	pages = {103013},
}

Achievement of human-level image recognition by deep neural networks (DNNs) has spurred interest in whether and how DNNs are brain-like. Both DNNs and the visual cortex perform hierarchical processing, and correspondence has been shown between hierarchical visual areas and DNN layers in representing visual features. Here, we propose the brain hierarchy (BH) score as a metric to quantify the degree of hierarchical correspondence based on neural decoding and encoding analyses where DNN unit activations and human brain activity are predicted from each other. We find that BH scores for 29 pre-trained DNNs with various architectures are negatively correlated with image recognition performance, thus indicating that recently developed high-performance DNNs are not necessarily brain-like. Experimental manipulations of DNN models suggest that single-path sequential feedforward architecture with broad spatial integration is critical to brain-like hierarchy. Our method may provide new ways to design DNNs in light of their representational homology to the brain.

@inproceedings{turgis_prognostic_2021,
	address = {Anger, France},
	title = {Prognostic {Expert} {System} for {Railway} {Fleet} {Maintenance}},
	doi = {http://dx.doi.org/10.3850/978-981-18-2016-8_391-cd},
	booktitle = {Proceedings of the 31st {European} {Safety} and {Reliability} {Conference}},
	author = {Turgis, Fabien and Audier, Pierre and Marion, Rémy},
	month = sep,
	year = {2021},
	doi = {10.3850/978-981-18-2016-8_391-cd},
	pages = {2111},
}

@misc{lee_predicting_2021,
	title = {Predicting {What} {You} {Already} {Know} {Helps}: {Provable} {Self}-{Supervised} {Learning}},
	shorttitle = {Predicting {What} {You} {Already} {Know} {Helps}},
	url = {http://arxiv.org/abs/2008.01064},
	doi = {10.48550/arXiv.2008.01064},
	abstract = {Self-supervised representation learning solves auxiliary prediction tasks (known as pretext tasks) without requiring labeled data to learn useful semantic representations. These pretext tasks are created solely using the input features, such as predicting a missing image patch, recovering the color channels of an image from context, or predicting missing words in text; yet predicting this {\textbackslash}textit\{known\} information helps in learning representations effective for downstream prediction tasks. We posit a mechanism exploiting the statistical connections between certain \{{\textbackslash}em reconstruction-based\} pretext tasks that guarantee to learn a good representation. Formally, we quantify how the approximate independence between the components of the pretext task (conditional on the label and latent variables) allows us to learn representations that can solve the downstream task by just training a linear layer on top of the learned representation. We prove the linear layer yields small approximation error even for complex ground truth function class and will drastically reduce labeled sample complexity. Next, we show a simple modification of our method leads to nonlinear CCA, analogous to the popular SimSiam algorithm, and show similar guarantees for nonlinear CCA.},
	urldate = {2022-12-07},
	publisher = {arXiv},
	author = {Lee, Jason D. and Lei, Qi and Saunshi, Nikunj and Zhuo, Jiacheng},
	month = nov,
	year = {2021},
	note = {arXiv:2008.01064 [cs, stat]},
	keywords = {Computer Science - Machine Learning, Statistics - Machine Learning},
}

Self-supervised representation learning solves auxiliary prediction tasks (known as pretext tasks) without requiring labeled data to learn useful semantic representations. These pretext tasks are created solely using the input features, such as predicting a missing image patch, recovering the color channels of an image from context, or predicting missing words in text; yet predicting this \textit\known\ information helps in learning representations effective for downstream prediction tasks. We posit a mechanism exploiting the statistical connections between certain \\em reconstruction-based\ pretext tasks that guarantee to learn a good representation. Formally, we quantify how the approximate independence between the components of the pretext task (conditional on the label and latent variables) allows us to learn representations that can solve the downstream task by just training a linear layer on top of the learned representation. We prove the linear layer yields small approximation error even for complex ground truth function class and will drastically reduce labeled sample complexity. Next, we show a simple modification of our method leads to nonlinear CCA, analogous to the popular SimSiam algorithm, and show similar guarantees for nonlinear CCA.

@article{putina_online_2021,
	title = {Online {Anomaly} {Detection} {Leveraging} {Stream}-{Based} {Clustering} and {Real}-{Time} {Telemetry}},
	volume = {18},
	issn = {1932-4537},
	doi = {10.1109/TNSM.2020.3037019},
	abstract = {Recent technology evolution allows network equipment to continuously stream a wealth of “telemetry” information, which pertains to multiple protocols and layers of the stack, at a very fine spatial-grain and high-frequency. This deluge of telemetry data clearly offers new opportunities for network control and troubleshooting, but also poses a serious challenge for what concerns its real-time processing. We tackle this challenge by applying streaming machine-learning techniques to the continuous flow of control and data-plane telemetry data, with the purpose of real-time detection of anomalies. In particular, we implement an anomaly detection engine that leverages DenStream, an unsupervised clustering technique, and apply it to features collected from a large-scale testbed comprising tens of routers traversed up to 3Terabit/s worth of real application traffic. We contrast DenStream with offline algorithms such as DBScan and Local Outlier Factor (LOF), as well as online algorithms such as the windowed version of DBScan, ExactSTORM, Continuous Outlier Detection (COD) and Robust Random Cut Forest (RRCF). Our experimental campaign compares these seven algorithms under both accuracy and computational complexity viewpoints: results testify that DenStream (i) achieves detection results on par with RRCF, the best performing algorithm and (ii) is significantly faster than other approaches, notably over two orders of magnitude faster than RRCF. In spirit with the recent trend toward reproducibility of results, we make our code available as open source to the scientific community.},
	number = {1},
	journal = {IEEE Transactions on Network and Service Management},
	author = {Putina, Andrian and Rossi, Dario},
	month = mar,
	year = {2021},
	note = {Conference Name: IEEE Transactions on Network and Service Management},
	keywords = {Anomaly detection, Anomaly detection algorithms, Feature extraction, Principal component analysis, Protocols, Real-time systems, Support vector machines, Telemetry, machine learning, model driven telemetry, network monitoring and measurements, stream learning},
	pages = {839--854},
}

Recent technology evolution allows network equipment to continuously stream a wealth of “telemetry” information, which pertains to multiple protocols and layers of the stack, at a very fine spatial-grain and high-frequency. This deluge of telemetry data clearly offers new opportunities for network control and troubleshooting, but also poses a serious challenge for what concerns its real-time processing. We tackle this challenge by applying streaming machine-learning techniques to the continuous flow of control and data-plane telemetry data, with the purpose of real-time detection of anomalies. In particular, we implement an anomaly detection engine that leverages DenStream, an unsupervised clustering technique, and apply it to features collected from a large-scale testbed comprising tens of routers traversed up to 3Terabit/s worth of real application traffic. We contrast DenStream with offline algorithms such as DBScan and Local Outlier Factor (LOF), as well as online algorithms such as the windowed version of DBScan, ExactSTORM, Continuous Outlier Detection (COD) and Robust Random Cut Forest (RRCF). Our experimental campaign compares these seven algorithms under both accuracy and computational complexity viewpoints: results testify that DenStream (i) achieves detection results on par with RRCF, the best performing algorithm and (ii) is significantly faster than other approaches, notably over two orders of magnitude faster than RRCF. In spirit with the recent trend toward reproducibility of results, we make our code available as open source to the scientific community.

@inproceedings{winkler_big_2021,
	title = {Big {Data} {Needs} and {Challenges} in {Smart} {Manufacturing}: {An} {Industry}-{Academia} {Survey}},
	shorttitle = {Big {Data} {Needs} and {Challenges} in {Smart} {Manufacturing}},
	doi = {10.1109/ETFA45728.2021.9613600},
	abstract = {The increasing availability of data in Smart Manufacturing opens new challenges and required capabilities in the area of big data in industry and academia. Various organizations have started initiatives to collect and analyse data in their individual contexts with specific goals, e.g., for monitoring, optimization, or decision support in order to reduce risks and costs in their manufacturing systems. However, the variety of available application areas require to focus on most promising activities. Therefore, we see the need for investigating common challenges and priorities in academia and industry from expert and management perspective to identify the state of the practice and promising application areas for driving future research directions. The goal of this paper is to report on an industry-academia survey to capture the current state of the art, required capabilities and priorities in the area of big data applications. Therefore, we conducted a survey in winter 2020/21 in industry and academia. We received 22 responses from different application domains highlighting the need for supporting (a) fault detection and (b) fault classification based on (c) historical and (d) real-time data analysis concepts. Therefore, the survey results reveals current and upcoming challenges in big data applications, such as defect handling based on historical and real-time data.},
	booktitle = {2021 26th {IEEE} {International} {Conference} on {Emerging} {Technologies} and {Factory} {Automation} ({ETFA} )},
	author = {Winkler, Dietmar and Korobeinykov, Alexander and Novák, Petr and Lüder, Arndt and Biffl, Stefan},
	month = sep,
	year = {2021},
	keywords = {Big Data Application, Big Data applications, Fault detection, Industries, Monitoring, Optimization, Organizations, Real-time systems, Required Capabilities, Smart Manufacturing, State of the Practice, Survey},
	pages = {1--8},
}

The increasing availability of data in Smart Manufacturing opens new challenges and required capabilities in the area of big data in industry and academia. Various organizations have started initiatives to collect and analyse data in their individual contexts with specific goals, e.g., for monitoring, optimization, or decision support in order to reduce risks and costs in their manufacturing systems. However, the variety of available application areas require to focus on most promising activities. Therefore, we see the need for investigating common challenges and priorities in academia and industry from expert and management perspective to identify the state of the practice and promising application areas for driving future research directions. The goal of this paper is to report on an industry-academia survey to capture the current state of the art, required capabilities and priorities in the area of big data applications. Therefore, we conducted a survey in winter 2020/21 in industry and academia. We received 22 responses from different application domains highlighting the need for supporting (a) fault detection and (b) fault classification based on (c) historical and (d) real-time data analysis concepts. Therefore, the survey results reveals current and upcoming challenges in big data applications, such as defect handling based on historical and real-time data.

@article{pappaterra_systematic_2021,
	title = {A {Systematic} {Review} of {Artificial} {Intelligence} {Public} {Datasets} for {Railway} {Applications}},
	volume = {6},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	issn = {2412-3811},
	url = {https://www.mdpi.com/2412-3811/6/10/136},
	doi = {10.3390/infrastructures6100136},
	abstract = {The aim of this paper is to review existing publicly available and open artificial intelligence (AI) oriented datasets in different domains and subdomains of the railway sector. The contribution of this paper is an overview of AI-oriented railway data published under Creative Commons (CC) or any other copyright type that entails public availability and freedom of use. These data are of great value for open research and publications related to the application of AI in the railway sector. This paper includes insights on the public railway data: we distinguish different subdomains, including maintenance and inspection, traffic planning and management, safety and security and type of data including numerical, string, image and other. The datasets reviewed cover the last three decades, from January 1990 to January 2021. The study revealed that the number of open datasets is very small in comparison with the available literature related to AI applications in the railway industry. Another shortcoming is the lack of documentation and metadata on public datasets, including information related to missing data, collection schemes and other limitations. This study also presents quantitative data, such as the number of available open datasets divided by railway application, type of data and year of publication. This review also reveals that there are openly available APIs—maintained by government organizations and train operating companies (TOCs)—that can be of great use for data harvesting and can facilitate the creation of large public datasets. These data are usually well-curated real-time data that can greatly contribute to the accuracy of AI models. Furthermore, we conclude that the extension of AI applications in the railway sector merits a centralized hub for publicly available datasets and open APIs.},
	language = {en},
	number = {10},
	urldate = {2022-05-15},
	journal = {Infrastructures},
	author = {Pappaterra, Mauro José and Flammini, Francesco and Vittorini, Valeria and Bešinović, Nikola},
	month = oct,
	year = {2021},
	note = {Number: 10
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {intelligent transportation, machine learning, predictive maintenance, public datasets, railways},
	pages = {136},
}

The aim of this paper is to review existing publicly available and open artificial intelligence (AI) oriented datasets in different domains and subdomains of the railway sector. The contribution of this paper is an overview of AI-oriented railway data published under Creative Commons (CC) or any other copyright type that entails public availability and freedom of use. These data are of great value for open research and publications related to the application of AI in the railway sector. This paper includes insights on the public railway data: we distinguish different subdomains, including maintenance and inspection, traffic planning and management, safety and security and type of data including numerical, string, image and other. The datasets reviewed cover the last three decades, from January 1990 to January 2021. The study revealed that the number of open datasets is very small in comparison with the available literature related to AI applications in the railway industry. Another shortcoming is the lack of documentation and metadata on public datasets, including information related to missing data, collection schemes and other limitations. This study also presents quantitative data, such as the number of available open datasets divided by railway application, type of data and year of publication. This review also reveals that there are openly available APIs—maintained by government organizations and train operating companies (TOCs)—that can be of great use for data harvesting and can facilitate the creation of large public datasets. These data are usually well-curated real-time data that can greatly contribute to the accuracy of AI models. Furthermore, we conclude that the extension of AI applications in the railway sector merits a centralized hub for publicly available datasets and open APIs.

@article{lomov_fault_2021,
	title = {Fault detection in {Tennessee} {Eastman} process with temporal deep learning models},
	volume = {23},
	issn = {2452-414X},
	url = {https://www.sciencedirect.com/science/article/pii/S2452414X21000145},
	doi = {10.1016/j.jii.2021.100216},
	abstract = {Automated early process fault detection and prediction remains a challenging problem in industrial processes. Traditionally it has been done by multivariate statistical analysis of sensor readings and, more recently, with the help of machine learning methods. The quality of machine learning models strongly depends on feature engineering, that in turn heavily relies on expertise of the process engineers and model developers. With the recent advent of deep learning neural network methods and abundance of available sensor data, it became possible to develop advanced approaches to early fault detection and prediction that do not require feature engineering and provide more accurate and timely results. In this paper we investigate a wide range of recurrent and convolutional architectures on the publicly available simulated Tennessee Eastman Process extended TEP dataset for the fault detection in chemical processes. We have selected the best architecture for the task and proposed a novel temporal CNN1D2D architecture that achieves overall better performance on the dataset than any referenced method. We have also proposed to use Generative Adversarial Network GAN to extend and enrich data used in training.},
	language = {en},
	urldate = {2022-05-02},
	journal = {Journal of Industrial Information Integration},
	author = {Lomov, Ildar and Lyubimov, Mark and Makarov, Ilya and Zhukov, Leonid E.},
	month = sep,
	year = {2021},
	keywords = {Chemical processes, Deep learning, Fault detection, Generative adversarial networks, Industrial data integration, Industrial data management, Industrial machine learning, Tennessee Eastman process},
	pages = {100216},
}

Automated early process fault detection and prediction remains a challenging problem in industrial processes. Traditionally it has been done by multivariate statistical analysis of sensor readings and, more recently, with the help of machine learning methods. The quality of machine learning models strongly depends on feature engineering, that in turn heavily relies on expertise of the process engineers and model developers. With the recent advent of deep learning neural network methods and abundance of available sensor data, it became possible to develop advanced approaches to early fault detection and prediction that do not require feature engineering and provide more accurate and timely results. In this paper we investigate a wide range of recurrent and convolutional architectures on the publicly available simulated Tennessee Eastman Process extended TEP dataset for the fault detection in chemical processes. We have selected the best architecture for the task and proposed a novel temporal CNN1D2D architecture that achieves overall better performance on the dataset than any referenced method. We have also proposed to use Generative Adversarial Network GAN to extend and enrich data used in training.

@article{zubaroglu_data_2021,
	title = {Data stream clustering: a review},
	volume = {54},
	issn = {1573-7462},
	shorttitle = {Data stream clustering},
	url = {https://doi.org/10.1007/s10462-020-09874-x},
	doi = {10.1007/s10462-020-09874-x},
	abstract = {Number of connected devices is steadily increasing and these devices continuously generate data streams. Real-time processing of data streams is arousing interest despite many challenges. Clustering is one of the most suitable methods for real-time data stream processing, because it can be applied with less prior information about the data and it does not need labeled instances. However, data stream clustering differs from traditional clustering in many aspects and it has several challenging issues. Here, we provide information regarding the concepts and common characteristics of data streams, such as concept drift, data structures for data streams, time window models and outlier detection. We comprehensively review recent data stream clustering algorithms and analyze them in terms of the base clustering technique, computational complexity and clustering accuracy. A comparison of these algorithms is given along with still open problems. We indicate popular data stream repositories and datasets, stream processing tools and platforms. Open problems about data stream clustering are also discussed.},
	language = {en},
	number = {2},
	urldate = {2022-03-25},
	journal = {Artificial Intelligence Review},
	author = {Zubaroğlu, Alaettin and Atalay, Volkan},
	month = feb,
	year = {2021},
	pages = {1201--1236},
}

Number of connected devices is steadily increasing and these devices continuously generate data streams. Real-time processing of data streams is arousing interest despite many challenges. Clustering is one of the most suitable methods for real-time data stream processing, because it can be applied with less prior information about the data and it does not need labeled instances. However, data stream clustering differs from traditional clustering in many aspects and it has several challenging issues. Here, we provide information regarding the concepts and common characteristics of data streams, such as concept drift, data structures for data streams, time window models and outlier detection. We comprehensively review recent data stream clustering algorithms and analyze them in terms of the base clustering technique, computational complexity and clustering accuracy. A comparison of these algorithms is given along with still open problems. We indicate popular data stream repositories and datasets, stream processing tools and platforms. Open problems about data stream clustering are also discussed.

@incollection{cerri_new_2021,
	address = {New York, NY, USA},
	title = {A new self-organizing map based algorithm for multi-label stream classification},
	isbn = {978-1-4503-8104-8},
	url = {https://doi.org/10.1145/3412841.3441922},
	abstract = {Several algorithms have been proposed for offline multi-label classification. However, applications in areas such as traffic monitoring, social networks, and sensors produce data continuously, the so called data streams, posing challenges to batch multi-label learning. With the lack of stationarity in the distribution of data streams, new algorithms are needed to online adapt to such changes (concept drift). Also, in realistic applications, changes occur in scenarios with infinitely delayed labels, where the true classes of the arrival instances are never available. We propose an online unsupervised incremental method based on self-organizing maps for multi-label stream classification in scenarios with infinitely delayed labels. We consider the existence of an initial set of labeled instances to train a self-organizing map for each label. The learned models are then used and adapted in an evolving stream to classify new instances, considering that their classes will never be available. We adapt to incremental concept drifts by online updating the weight vectors of winner neurons and the dataset label cardinality. Predictions are obtained using the Bayes rule and the outputs of each neuron, adapting the prior probabilities and conditional probabilities of the classes in the stream. Experiments using synthetic and real datasets show that our method is highly competitive with several ones from the literature, in both stationary and concept drift scenarios.},
	urldate = {2022-03-17},
	booktitle = {Proceedings of the 36th {Annual} {ACM} {Symposium} on {Applied} {Computing}},
	publisher = {Association for Computing Machinery},
	author = {Cerri, Ricardo and Junior, Joel David C. and Faria, Elaine. R. and Gama, João},
	month = mar,
	year = {2021},
	keywords = {classification, concept drift, data streams, machine learning, multi-label, self-organizing maps},
	pages = {418--426},
}

Several algorithms have been proposed for offline multi-label classification. However, applications in areas such as traffic monitoring, social networks, and sensors produce data continuously, the so called data streams, posing challenges to batch multi-label learning. With the lack of stationarity in the distribution of data streams, new algorithms are needed to online adapt to such changes (concept drift). Also, in realistic applications, changes occur in scenarios with infinitely delayed labels, where the true classes of the arrival instances are never available. We propose an online unsupervised incremental method based on self-organizing maps for multi-label stream classification in scenarios with infinitely delayed labels. We consider the existence of an initial set of labeled instances to train a self-organizing map for each label. The learned models are then used and adapted in an evolving stream to classify new instances, considering that their classes will never be available. We adapt to incremental concept drifts by online updating the weight vectors of winner neurons and the dataset label cardinality. Predictions are obtained using the Bayes rule and the outputs of each neuron, adapting the prior probabilities and conditional probabilities of the classes in the stream. Experiments using synthetic and real datasets show that our method is highly competitive with several ones from the literature, in both stationary and concept drift scenarios.

@inproceedings{bahri_survey_2021,
	address = {Yokohama, Yokohama, Japan},
	series = {{IJCAI}'20},
	title = {Survey on feature transformation techniques for data streams},
	isbn = {978-0-9992411-6-5},
	abstract = {Mining high-dimensional data streams poses a fundamental challenge to machine learning as the presence of high numbers of attributes can remarkably degrade any mining task's performance. In the past several years, dimension reduction (DR) approaches have been successfully applied for different purposes (e.g., visualization). Due to their high-computational costs and numerous passes over large data, these approaches pose a hindrance when processing infinite data streams that are potentially high-dimensional. The latter increases the resource-usage of algorithms that could suffer from the curse of dimensionality. To cope with these issues, some techniques for incremental DR have been proposed. In this paper, we provide a survey on reduction approaches designed to handle data streams and highlight the key benefits of using these approaches for stream mining algorithms.},
	urldate = {2022-03-15},
	booktitle = {Proceedings of the {Twenty}-{Ninth} {International} {Joint} {Conference} on {Artificial} {Intelligence}},
	author = {Bahri, Maroua and Bifet, Albert and Maniu, Silviu and Gomes, Heitor Murilo},
	month = jan,
	year = {2021},
	pages = {4796--4802},
}

Mining high-dimensional data streams poses a fundamental challenge to machine learning as the presence of high numbers of attributes can remarkably degrade any mining task's performance. In the past several years, dimension reduction (DR) approaches have been successfully applied for different purposes (e.g., visualization). Due to their high-computational costs and numerous passes over large data, these approaches pose a hindrance when processing infinite data streams that are potentially high-dimensional. The latter increases the resource-usage of algorithms that could suffer from the curse of dimensionality. To cope with these issues, some techniques for incremental DR have been proposed. In this paper, we provide a survey on reduction approaches designed to handle data streams and highlight the key benefits of using these approaches for stream mining algorithms.

@article{zhong_long_2021,
	title = {Long short-term memory self-adapting online random forests for evolving data stream regression},
	volume = {457},
	issn = {0925-2312},
	url = {https://www.sciencedirect.com/science/article/pii/S0925231221007797},
	doi = {10.1016/j.neucom.2021.05.026},
	abstract = {Evolving data stream, especially with concept drift is generally accepted as a challenging data type for regression task, because it usually makes machine learning models trained on old data not adapting to new data, and leads to dramatic performance degradation as a result. Moreover, the behavior of a data stream may change in different modes and therefore introduces various concept drifts, e.g., abrupt, incremental, gradual, recurring, even more, complex concept drifts. Although there are some algorithms that can adapt to stationary data streams or a specific type of concept drift in non-stationary data streams, a wide range of practical applications call for machine learning regression models to handle multi-type of data streams. In this work, we propose an online learning strategy called adaptive long and short-term memories online Random Forests regression(ALSM-RFR), where an adaptive memory activation mechanism is designed to make the model switch adaptively between long-term and hybrid (long-term plus short-term) memory modes in the face of stationary data streams or non-stationary data streams with different types of concept drift. In particular, leaf and tree weights in random forests are used to learn information at different timescales, namely, long-term and short-term memories. Moreover, we devise an adaptive memory activation mechanism to formulate the switch decision of memory modes as a classification problem. Numerical experiments show remarkable improvements of the proposed method in the adaptability of stream types and predictive accuracy in data streams across several real datasets and synthetic datasets, compared to the state-of-the-art online approaches. Besides, the convergence and the influence of the parameters involved in our method are evaluated.},
	language = {en},
	urldate = {2022-03-14},
	journal = {Neurocomputing},
	author = {Zhong, Yuan and Yang, Hongyu and Zhang, Yanci and Li, Ping and Ren, Cheng},
	month = oct,
	year = {2021},
	keywords = {Adaptive memory activation mechanism, Concept drift, Data stream, Long-term and short-term memory, Online learning, Online random forests regression},
	pages = {265--276},
}

Evolving data stream, especially with concept drift is generally accepted as a challenging data type for regression task, because it usually makes machine learning models trained on old data not adapting to new data, and leads to dramatic performance degradation as a result. Moreover, the behavior of a data stream may change in different modes and therefore introduces various concept drifts, e.g., abrupt, incremental, gradual, recurring, even more, complex concept drifts. Although there are some algorithms that can adapt to stationary data streams or a specific type of concept drift in non-stationary data streams, a wide range of practical applications call for machine learning regression models to handle multi-type of data streams. In this work, we propose an online learning strategy called adaptive long and short-term memories online Random Forests regression(ALSM-RFR), where an adaptive memory activation mechanism is designed to make the model switch adaptively between long-term and hybrid (long-term plus short-term) memory modes in the face of stationary data streams or non-stationary data streams with different types of concept drift. In particular, leaf and tree weights in random forests are used to learn information at different timescales, namely, long-term and short-term memories. Moreover, we devise an adaptive memory activation mechanism to formulate the switch decision of memory modes as a classification problem. Numerical experiments show remarkable improvements of the proposed method in the adaptability of stream types and predictive accuracy in data streams across several real datasets and synthetic datasets, compared to the state-of-the-art online approaches. Besides, the convergence and the influence of the parameters involved in our method are evaluated.

@article{duan_novel_2021,
	title = {A novel {ResNet}-based model structure and its applications in machine health monitoring},
	volume = {27},
	issn = {1077-5463},
	url = {https://doi.org/10.1177/1077546320936506},
	doi = {10.1177/1077546320936506},
	abstract = {Machine health monitoring has become increasingly important in modern manufacturers because of its ability to reduce downtime of the machine and cut down the production cost. Enormous signals acquired from machinery are capable of reflecting current working conditions by in-depth analysis with various data-driven methods. Hand-crafted feature extraction and representation from the traditional methods are essential but daunting tasks, and these methods may not be suitable for these massive data. Compared with traditional methods, deep learning ones are able to extract the best feature combination during model training without any artificial intervention, which makes it easier, more efficient, and more effective to monitor machine health, but the training cost and training time hamper its application. The short-time Fourier transform is adopted as the data preprocessing method to cut down the training cost and boost the training procedure. Inspired by the great achievements of ResNet, the new optimized model based on ResNet has been proposed with layer-by-layer dimension reduction of the feature maps. The proposed model is also able to avoid information loss in the conventional pooling layer. All the potential candidate model blocks are introduced and compared, and the best one is selected as the final one. Repeated model block layers are adapted for the best feature combinations, followed by a two-layer full connection layer for the final targets. The proposed method is validated by conducting experiments on bearing fault diagnosis and tool wear prediction dataset. The final results show that the proposed model achieves the best accuracy rate in the classification task and the lowest root mean squared error in the prediction task.},
	language = {en},
	number = {9-10},
	urldate = {2022-02-09},
	journal = {Journal of Vibration and Control},
	author = {Duan, Jian and Shi, Tielin and Zhou, Hongdi and Xuan, Jianping and Wang, Shuhua},
	month = may,
	year = {2021},
	note = {Publisher: SAGE Publications Ltd STM},
	keywords = {ResNet, bearing, convolution neural network, machine health monitoring, tool Wear},
	pages = {1036--1050},
}

Machine health monitoring has become increasingly important in modern manufacturers because of its ability to reduce downtime of the machine and cut down the production cost. Enormous signals acquired from machinery are capable of reflecting current working conditions by in-depth analysis with various data-driven methods. Hand-crafted feature extraction and representation from the traditional methods are essential but daunting tasks, and these methods may not be suitable for these massive data. Compared with traditional methods, deep learning ones are able to extract the best feature combination during model training without any artificial intervention, which makes it easier, more efficient, and more effective to monitor machine health, but the training cost and training time hamper its application. The short-time Fourier transform is adopted as the data preprocessing method to cut down the training cost and boost the training procedure. Inspired by the great achievements of ResNet, the new optimized model based on ResNet has been proposed with layer-by-layer dimension reduction of the feature maps. The proposed model is also able to avoid information loss in the conventional pooling layer. All the potential candidate model blocks are introduced and compared, and the best one is selected as the final one. Repeated model block layers are adapted for the best feature combinations, followed by a two-layer full connection layer for the final targets. The proposed method is validated by conducting experiments on bearing fault diagnosis and tool wear prediction dataset. The final results show that the proposed model achieves the best accuracy rate in the classification task and the lowest root mean squared error in the prediction task.

@article{zhu_decoupled_2021,
	title = {Decoupled {Feature}-{Temporal} {CNN}: {Explaining} {Deep} {Learning}-{Based} {Machine} {Health} {Monitoring}},
	volume = {70},
	issn = {1557-9662},
	shorttitle = {Decoupled {Feature}-{Temporal} {CNN}},
	doi = {10.1109/TIM.2021.3084310},
	abstract = {Machine learning, especially deep learning, has been extensively applied and studied in the area of machine health monitoring. For machine health monitoring systems (MHMS), major efforts have been put into designing and deploying more and more complex machine learning models. Those black-box models are nontransparent toward their working mechanism. However, this research trend brings huge potential risks in real life. Since machine health monitoring itself belongs to high stake decision applications, the outputs of the autonomous monitoring systems should be trustworthy and reliable, which refers to obtain explainability. Then, it comes to the following key question: why the deployed MHMS predicts what they predict. In this article, we shed some light on this meaningful research direction: explainable MHMSs (EMHMS). In EMHMS, the machine doctor could act like a real doctor who can not only make a diagnosis but also describe the patient's symptoms. First, we propose a specific convolutional neural network (CNN) structure, named DecouplEd Feature-Temporal CNN (DEFT-CNN), to balance precision-explainability tradeoff. Specifically, feature information and temporal information have been encoded in different stages of our model. The spatial attention module is added to boost the performance of the model. Then, to explain the decision of the model, we adopt gradient-based methods to generate features and temporal saliency maps highlighting which kinds of features and time steps are keys for the model's predictions. Finally, we conduct the experimental studies on two real datasets to verify the effectiveness of our proposed framework.},
	journal = {IEEE Transactions on Instrumentation and Measurement},
	author = {Zhu, Chaoyi and Chen, Zhenghua and Zhao, Rui and Wang, Jinjiang and Yan, Ruqiang},
	year = {2021},
	note = {Conference Name: IEEE Transactions on Instrumentation and Measurement},
	keywords = {Attention mechanism, Computational modeling, Data models, Deep learning, Monitoring, Neural networks, Predictive models, Time-frequency analysis, convolutional neural network (CNN), deep learning, ecml, explainable machine learning, machine health monitoring},
	pages = {1--13},
}

Machine learning, especially deep learning, has been extensively applied and studied in the area of machine health monitoring. For machine health monitoring systems (MHMS), major efforts have been put into designing and deploying more and more complex machine learning models. Those black-box models are nontransparent toward their working mechanism. However, this research trend brings huge potential risks in real life. Since machine health monitoring itself belongs to high stake decision applications, the outputs of the autonomous monitoring systems should be trustworthy and reliable, which refers to obtain explainability. Then, it comes to the following key question: why the deployed MHMS predicts what they predict. In this article, we shed some light on this meaningful research direction: explainable MHMSs (EMHMS). In EMHMS, the machine doctor could act like a real doctor who can not only make a diagnosis but also describe the patient's symptoms. First, we propose a specific convolutional neural network (CNN) structure, named DecouplEd Feature-Temporal CNN (DEFT-CNN), to balance precision-explainability tradeoff. Specifically, feature information and temporal information have been encoded in different stages of our model. The spatial attention module is added to boost the performance of the model. Then, to explain the decision of the model, we adopt gradient-based methods to generate features and temporal saliency maps highlighting which kinds of features and time steps are keys for the model's predictions. Finally, we conduct the experimental studies on two real datasets to verify the effectiveness of our proposed framework.

@article{liu_lstm-gan-ae_2021,
	title = {{LSTM}-{GAN}-{AE}: {A} {Promising} {Approach} for {Fault} {Diagnosis} in {Machine} {Health} {Monitoring}},
	issn = {1557-9662},
	shorttitle = {{LSTM}-{GAN}-{AE}},
	doi = {10.1109/TIM.2021.3135328},
	abstract = {Recent years have witnessed that, real-time health monitoring for machine gains more and more importance with the goal of achieving fault diagnosis and predictive maintenance. Conventional diagnosis methods face formidable challenges imposed by the high requirement for expert knowledge and extensive labor. The diagnosis scheme based on deep learning (DL) models has served as a promising solution and achieved great success. However, many of these DL-based models are fail to extract critical temporal information thoroughly. In addition, it is difficult to apply them to machine health monitoring (MHM) in real-time as those methods take long time for diagnosis in practice. To address the aforementioned issues, this paper introduces a novel intelligent fault diagnosis algorithm with three stages for MHM. It is a hybrid framework that combines generative adversarial networks (GAN) and auto-encoder (AE) based on the bi-directional long short-term memory (bi-LSTM). Firstly, GAN is employed to obtain the reconstruction residual and learn the discriminative representation. Then, AE is used to perform the critical temporal features extraction and dimension reduction. Finally, the supervised learning model is constructed to integrate feature information and predict diagnosis results. To verify the effectiveness of the proposed algorithm, typical rolling bearing datasets are taken as trial data. Preliminary simulation results demonstrate that the proposed algorithm achieves superior performance compared to the competing methods.},
	journal = {IEEE Transactions on Instrumentation and Measurement},
	author = {Liu, Haoqiang and Zhao, Hongbo and Wang, Jiayue and Yuan, Shuai and Feng, Wenquan},
	year = {2021},
	note = {Conference Name: IEEE Transactions on Instrumentation and Measurement},
	keywords = {Data mining, Deep learning, Fault diagnosis, Feature extraction, Generative adversarial networks, Logic gates, Monitoring, Support vector machines, ecml, fault diagnosis, long short-term memory (LSTM), machine health monitoring},
	pages = {1--1},
}

Recent years have witnessed that, real-time health monitoring for machine gains more and more importance with the goal of achieving fault diagnosis and predictive maintenance. Conventional diagnosis methods face formidable challenges imposed by the high requirement for expert knowledge and extensive labor. The diagnosis scheme based on deep learning (DL) models has served as a promising solution and achieved great success. However, many of these DL-based models are fail to extract critical temporal information thoroughly. In addition, it is difficult to apply them to machine health monitoring (MHM) in real-time as those methods take long time for diagnosis in practice. To address the aforementioned issues, this paper introduces a novel intelligent fault diagnosis algorithm with three stages for MHM. It is a hybrid framework that combines generative adversarial networks (GAN) and auto-encoder (AE) based on the bi-directional long short-term memory (bi-LSTM). Firstly, GAN is employed to obtain the reconstruction residual and learn the discriminative representation. Then, AE is used to perform the critical temporal features extraction and dimension reduction. Finally, the supervised learning model is constructed to integrate feature information and predict diagnosis results. To verify the effectiveness of the proposed algorithm, typical rolling bearing datasets are taken as trial data. Preliminary simulation results demonstrate that the proposed algorithm achieves superior performance compared to the competing methods.

@article{hou_adaptive_2021,
	title = {Adaptive {Weighted} {Signal} {Preprocessing} {Technique} for {Machine} {Health} {Monitoring}},
	volume = {70},
	issn = {1557-9662},
	doi = {10.1109/TIM.2020.3033471},
	abstract = {Machine health monitoring (MHM) aims to timely detect the incipient faults and monotonically assess the machine degradation tendency for prediction of remaining useful life (RUL), which is the basis of condition-based maintenance. Construction of a health index (HI) is a core step to realize the aforementioned purposes. Among the existing HIs, sparsity measures (SMs), including kurtosis, smoothness index, Gini index, and negative entropy, have shown promising applications in fault diagnosis of rotating machines because they are widely used to quantify the repetitive transients caused by rotating faults. However, drawbacks of SMs for MHM still exist and they are that: 1) SMs are too fluctuating to detect the incipient faults; 2) SMs are prone to be affected by impulsive noise; and 3) SMs might not exhibit monotonic degradation tendency. To enhance the abilities of SMs as HIs for MHM, some improvements on SMs, coined as an adaptive weighted signal preprocessing technique (AWSPT), are proposed in this article. Subsequently, theoretical values of AWSPT-based SMs under healthy states are investigated. Numerical experiments reveal that AWSPT-based SMs can quantify the cyclostationarity and they are robust to the effects of impulsive noise. Bearing and gear run-to-failure data sets are used to show that the proposed AWSPT-based SMs can simultaneously detect the early bearing and gear faults and provide the monotonic degradation tendency. Moreover, AWSPT-based SMs are more effective in selecting an optimal envelope demodulation frequency band than traditional SMs.},
	journal = {IEEE Transactions on Instrumentation and Measurement},
	author = {Hou, Bingchang and Wang, Dong and Wang, Yi and Yan, Tongtong and Peng, Zhike and Tsui, Kwok-Leung},
	year = {2021},
	note = {Conference Name: IEEE Transactions on Instrumentation and Measurement},
	keywords = {Degradation, Gaussian noise, Gears, Health index (HI), Indexes, Monitoring, Probability density function, Transient analysis, ecml, health monitoring, incipient fault detection, monotonic degradation assessment, sparsity measures (SMs)},
	pages = {1--11},
}

Machine health monitoring (MHM) aims to timely detect the incipient faults and monotonically assess the machine degradation tendency for prediction of remaining useful life (RUL), which is the basis of condition-based maintenance. Construction of a health index (HI) is a core step to realize the aforementioned purposes. Among the existing HIs, sparsity measures (SMs), including kurtosis, smoothness index, Gini index, and negative entropy, have shown promising applications in fault diagnosis of rotating machines because they are widely used to quantify the repetitive transients caused by rotating faults. However, drawbacks of SMs for MHM still exist and they are that: 1) SMs are too fluctuating to detect the incipient faults; 2) SMs are prone to be affected by impulsive noise; and 3) SMs might not exhibit monotonic degradation tendency. To enhance the abilities of SMs as HIs for MHM, some improvements on SMs, coined as an adaptive weighted signal preprocessing technique (AWSPT), are proposed in this article. Subsequently, theoretical values of AWSPT-based SMs under healthy states are investigated. Numerical experiments reveal that AWSPT-based SMs can quantify the cyclostationarity and they are robust to the effects of impulsive noise. Bearing and gear run-to-failure data sets are used to show that the proposed AWSPT-based SMs can simultaneously detect the early bearing and gear faults and provide the monotonic degradation tendency. Moreover, AWSPT-based SMs are more effective in selecting an optimal envelope demodulation frequency band than traditional SMs.

@article{han_online_2021,
	title = {Online fault diagnosis for sucker rod pumping well by optimized density peak clustering},
	issn = {0019-0578},
	url = {https://www.sciencedirect.com/science/article/pii/S0019057821001634},
	doi = {10.1016/j.isatra.2021.03.022},
	abstract = {Online diagnosis for sucker rod pumping well has great significances for rapidly grasping operations of the oil well. Feature extraction of the working condition and determination of the online diagnostic algorithm are two indispensable parts. In this paper, five feature vectors are extracted using Freeman chain codes. Then, an optimized density peak clustering (DPC) method is proposed to realize online diagnosis solved by an improved brain storm optimization (BSO) algorithm, in which the cloud model is adopted to generate new solutions in the searching space. During the online diagnosis process, a new cluster updating strategy is used to update the cluster centers online. According to the proposed online diagnostic method, various samples are automatically classified into different classifications by the unsupervised learning. The simulation results verify that the proposed online diagnosis method is satisfactory, which can give a higher and more stable diagnostic accuracy.},
	language = {en},
	urldate = {2022-01-14},
	journal = {ISA Transactions},
	author = {Han, Ying and Li, Kun and Ge, Fawei and Wang, Yi’an and Xu, Wensu},
	month = mar,
	year = {2021},
	keywords = {Density peak clustering, Dynamometer card, Fault diagnosis, Optimization, Sucker rod pumping well},
}

Online diagnosis for sucker rod pumping well has great significances for rapidly grasping operations of the oil well. Feature extraction of the working condition and determination of the online diagnostic algorithm are two indispensable parts. In this paper, five feature vectors are extracted using Freeman chain codes. Then, an optimized density peak clustering (DPC) method is proposed to realize online diagnosis solved by an improved brain storm optimization (BSO) algorithm, in which the cloud model is adopted to generate new solutions in the searching space. During the online diagnosis process, a new cluster updating strategy is used to update the cluster centers online. According to the proposed online diagnostic method, various samples are automatically classified into different classifications by the unsupervised learning. The simulation results verify that the proposed online diagnosis method is satisfactory, which can give a higher and more stable diagnostic accuracy.

@article{pham_data-driven_2021,
	title = {Data-driven fault detection of open circuits in multi-phase inverters based on current polarity using {Auto}-adaptive and {Dynamical} {Clustering}},
	volume = {113},
	issn = {0019-0578},
	url = {https://www.sciencedirect.com/science/article/pii/S0019057820302561},
	doi = {10.1016/j.isatra.2020.06.009},
	abstract = {This paper proposes a data-driven method for the detection and isolation of open-circuit faults in multi-phase inverters using measurements of the motor currents. First, feature variables are formulated in terms of the averages of the phase currents and their absolute values. Next, by using an AUto-adaptive and Dynamical Clustering (AUDyC) based on Gaussian Mixture Models, feature data is clustered into different classes characterizing normal and faulty operation modes. Afterwards, these classes are used for deriving appropriate conditions for detecting and labelling faults. The proposed method requires minimal knowledge about the system operation. Furthermore, it allows us to update our knowledge of existing faults online, thus making it possible to detect unknown faults. Moreover, conditions are formulated to describe the influence of the method parameters on the detection time. Once parameters are tuned, the accuracy of the proposed method is illustrated on various experimental data sets, where single and double faults are detected with detection times in the order of the fundamental signal period.},
	language = {en},
	urldate = {2022-01-14},
	journal = {ISA Transactions},
	author = {Pham, Thanh-Hung and Lefteriu, Sanda and Duviella, Eric and Lecoeuche, Stéphane},
	month = jul,
	year = {2021},
	pages = {185--195},
}

This paper proposes a data-driven method for the detection and isolation of open-circuit faults in multi-phase inverters using measurements of the motor currents. First, feature variables are formulated in terms of the averages of the phase currents and their absolute values. Next, by using an AUto-adaptive and Dynamical Clustering (AUDyC) based on Gaussian Mixture Models, feature data is clustered into different classes characterizing normal and faulty operation modes. Afterwards, these classes are used for deriving appropriate conditions for detecting and labelling faults. The proposed method requires minimal knowledge about the system operation. Furthermore, it allows us to update our knowledge of existing faults online, thus making it possible to detect unknown faults. Moreover, conditions are formulated to describe the influence of the method parameters on the detection time. Once parameters are tuned, the accuracy of the proposed method is illustrated on various experimental data sets, where single and double faults are detected with detection times in the order of the fundamental signal period.

@article{wei_learning_2021,
	title = {Learning the health index of complex systems using dynamic conditional variational autoencoders},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021005147},
	doi = {10.1016/j.ress.2021.108004},
	abstract = {Recent advances in sensing technologies have enabled engineers to collect big data to predict the remaining useful life (RUL) of complex systems. Current modeling techniques for RUL predictions are usually not able to quantify the degradation behavior of a complex system through a health index. Although some studies have been conducted to learn the health index of degradation systems, most of the existing methods are highly dependent on pre-defined assumptions which may not be consistent with the real degradation behaviors. To address this issue, we introduce a time-dependent directed graphical model to characterize the probabilistic relationships among sensor signals, RUL, operational conditions, and health index. Based on the graphical model, a dynamic conditional variational autoencoder is proposed to learn the health index. The experimental results have shown that the proposed method can learn an effective and reliable health index that measures complex system degradation behavior. Moreover, the learned health index improves the accuracy of RUL predictions.},
	language = {en},
	urldate = {2022-01-13},
	journal = {Reliability Engineering \& System Safety},
	author = {Wei, Yupeng and Wu, Dazhong and Terpenny, Janis},
	month = dec,
	year = {2021},
	keywords = {Data fusion, Deep learning, Degradation modeling, Multiple sensors, Remaining useful life, sigkdd-rw},
	pages = {108004},
}

Recent advances in sensing technologies have enabled engineers to collect big data to predict the remaining useful life (RUL) of complex systems. Current modeling techniques for RUL predictions are usually not able to quantify the degradation behavior of a complex system through a health index. Although some studies have been conducted to learn the health index of degradation systems, most of the existing methods are highly dependent on pre-defined assumptions which may not be consistent with the real degradation behaviors. To address this issue, we introduce a time-dependent directed graphical model to characterize the probabilistic relationships among sensor signals, RUL, operational conditions, and health index. Based on the graphical model, a dynamic conditional variational autoencoder is proposed to learn the health index. The experimental results have shown that the proposed method can learn an effective and reliable health index that measures complex system degradation behavior. Moreover, the learned health index improves the accuracy of RUL predictions.

@article{melani_framework_2021,
	title = {A framework to automate fault detection and diagnosis based on moving window principal component analysis and {Bayesian} network},
	volume = {215},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021003574},
	doi = {10.1016/j.ress.2021.107837},
	abstract = {Through Condition-Based Maintenance strategy, planners can monitor the health of the machinery and recommend actions based on the information obtained. Nevertheless, this approach depends on the successful establishment of Fault Detection and Diagnosis (FDD) processes. Although FDD is a research area in full growth with the development of several methods and heuristics, the availability of data from systems under a fault condition is still scarce in many applications, mainly related to complex systems. In many circumstances, only data from the system in healthy conditions is available and the applied FDD method should be able to detect variations in system conditions and diagnose faults without the need for previous labeled fault data. In this context, this article proposes a hybrid framework to automate FDD based on Moving Window Principal Component Analysis (MWPCA) and Bayesian Network (BN). First, the knowledge base on technical systems is organized to support the next steps of the framework. Then, the detection and diagnosis processes are performed sequentially through MWPCA and BN. The framework was implemented in the analysis of a simplified model of a hydrogenerator, considering real and simulated data. The results showed that the proposed method was able to detect and diagnose several simulated failures.},
	language = {en},
	urldate = {2022-01-13},
	journal = {Reliability Engineering \& System Safety},
	author = {Melani, Arthur Henrique de Andrade and Michalski, Miguel Angelo de Carvalho and da Silva, Renan Favarão and de Souza, Gilberto Francisco Martha},
	month = nov,
	year = {2021},
	keywords = {Adaptative principal component analysis, Bayesian network, Fault detection and diagnosis, MWPCA, Principal component analysis, sigkdd-rw},
	pages = {107837},
}

Through Condition-Based Maintenance strategy, planners can monitor the health of the machinery and recommend actions based on the information obtained. Nevertheless, this approach depends on the successful establishment of Fault Detection and Diagnosis (FDD) processes. Although FDD is a research area in full growth with the development of several methods and heuristics, the availability of data from systems under a fault condition is still scarce in many applications, mainly related to complex systems. In many circumstances, only data from the system in healthy conditions is available and the applied FDD method should be able to detect variations in system conditions and diagnose faults without the need for previous labeled fault data. In this context, this article proposes a hybrid framework to automate FDD based on Moving Window Principal Component Analysis (MWPCA) and Bayesian Network (BN). First, the knowledge base on technical systems is organized to support the next steps of the framework. Then, the detection and diagnosis processes are performed sequentially through MWPCA and BN. The framework was implemented in the analysis of a simplified model of a hydrogenerator, considering real and simulated data. The results showed that the proposed method was able to detect and diagnose several simulated failures.

@article{de_lima_healthmon_2021,
	title = {{HealthMon}: {An} approach for monitoring machines degradation using time-series decomposition, clustering, and metaheuristics},
	volume = {162},
	issn = {0360-8352},
	shorttitle = {{HealthMon}},
	url = {https://www.sciencedirect.com/science/article/pii/S0360835221006136},
	doi = {10.1016/j.cie.2021.107709},
	abstract = {Monitoring the degradation of machines to anticipate potential failures represents a significant challenge. In Industry 4.0, this task is critical when the costs associated with maintenance and stoppages on the productive processes are high. Nowadays, many preventive maintenance techniques employ supervised or unsupervised machine learning algorithms. However, the definition of which features should be processed by such algorithms is not a simple task, being crucial to the proposed technique’s success. Against this background, we consider whether unsupervised algorithms combined with time-series decomposition can enhance the estimate of a machine’s health. This article proposes HealthMon as a novel approach to compute a health index of machines based on sensor measurements. HealthMon extracts time-series from such sensors, which are decomposed in an unsupervised way to present the health state along time. The health index is related to the degradation of the considered machine, thus optimizing the machine maintenance schedule. This work advances the state-of-the-art in the following ways: (i) it proposes a novel index of machines health, which yields a more direct and intuitive view of machine degradation; (ii) it devises the first approach capable of estimating the health index of a machine in a completely unsupervised way; (iii) it generalizes vibrating and rotating machines, thus being able to monitor a wide range of industrial equipment. We evaluated our method using both simulated and real data. The results show that the evolution of vibrating machines’ failures can be effectively detected under various input workloads. Finally, through HealthMon, industry decision-makers benefit from the guidelines for preventive actions at appropriate times, thus meeting Industry 4.0.},
	language = {en},
	urldate = {2021-10-04},
	journal = {Computers \& Industrial Engineering},
	author = {de Lima, Miromar Jose and Paredes Crovato, Cesar David and Goytia Mejia, Rodrigo Ivan and da Rosa Righi, Rodrigo and de Oliveira Ramos, Gabriel and André da Costa, Cristiano and Pesenti, Giovani},
	month = dec,
	year = {2021},
	keywords = {Health index, Machine learning, Monitoring, Prediction, Preventive maintenance, Time-series, Unsupervised learning, sigkdd-rw},
	pages = {107709},
}

Monitoring the degradation of machines to anticipate potential failures represents a significant challenge. In Industry 4.0, this task is critical when the costs associated with maintenance and stoppages on the productive processes are high. Nowadays, many preventive maintenance techniques employ supervised or unsupervised machine learning algorithms. However, the definition of which features should be processed by such algorithms is not a simple task, being crucial to the proposed technique’s success. Against this background, we consider whether unsupervised algorithms combined with time-series decomposition can enhance the estimate of a machine’s health. This article proposes HealthMon as a novel approach to compute a health index of machines based on sensor measurements. HealthMon extracts time-series from such sensors, which are decomposed in an unsupervised way to present the health state along time. The health index is related to the degradation of the considered machine, thus optimizing the machine maintenance schedule. This work advances the state-of-the-art in the following ways: (i) it proposes a novel index of machines health, which yields a more direct and intuitive view of machine degradation; (ii) it devises the first approach capable of estimating the health index of a machine in a completely unsupervised way; (iii) it generalizes vibrating and rotating machines, thus being able to monitor a wide range of industrial equipment. We evaluated our method using both simulated and real data. The results show that the evolution of vibrating machines’ failures can be effectively detected under various input workloads. Finally, through HealthMon, industry decision-makers benefit from the guidelines for preventive actions at appropriate times, thus meeting Industry 4.0.

@article{pang_deep_2021,
	title = {Deep {Learning} for {Anomaly} {Detection}: {A} {Review}},
	volume = {54},
	issn = {0360-0300},
	shorttitle = {Deep {Learning} for {Anomaly} {Detection}},
	url = {https://doi.org/10.1145/3439950},
	doi = {10.1145/3439950},
	abstract = {Anomaly detection, a.k.a. outlier detection or novelty detection, has been a lasting yet active research area in various research communities for several decades. There are still some unique problem complexities and challenges that require advanced approaches. In recent years, deep learning enabled anomaly detection, i.e., deep anomaly detection, has emerged as a critical direction. This article surveys the research of deep anomaly detection with a comprehensive taxonomy, covering advancements in 3 high-level categories and 11 fine-grained categories of the methods. We review their key intuitions, objective functions, underlying assumptions, advantages, and disadvantages and discuss how they address the aforementioned challenges. We further discuss a set of possible future opportunities and new perspectives on addressing the challenges.},
	number = {2},
	urldate = {2021-11-27},
	journal = {ACM Computing Surveys},
	author = {Pang, Guansong and Shen, Chunhua and Cao, Longbing and Hengel, Anton Van Den},
	month = mar,
	year = {2021},
	keywords = {Anomaly detection, deep learning, novelty detection, one-class classification, outlier detection},
	pages = {38:1--38:38},
}

Anomaly detection, a.k.a. outlier detection or novelty detection, has been a lasting yet active research area in various research communities for several decades. There are still some unique problem complexities and challenges that require advanced approaches. In recent years, deep learning enabled anomaly detection, i.e., deep anomaly detection, has emerged as a critical direction. This article surveys the research of deep anomaly detection with a comprehensive taxonomy, covering advancements in 3 high-level categories and 11 fine-grained categories of the methods. We review their key intuitions, objective functions, underlying assumptions, advantages, and disadvantages and discuss how they address the aforementioned challenges. We further discuss a set of possible future opportunities and new perspectives on addressing the challenges.

@article{foorthuis_nature_2021,
	title = {On the nature and types of anomalies: a review of deviations in data},
	volume = {12},
	issn = {2364-4168},
	shorttitle = {On the nature and types of anomalies},
	url = {https://doi.org/10.1007/s41060-021-00265-1},
	doi = {10.1007/s41060-021-00265-1},
	abstract = {Anomalies are occurrences in a dataset that are in some way unusual and do not fit the general patterns. The concept of the anomaly is typically ill defined and perceived as vague and domain-dependent. Moreover, despite some 250 years of publications on the topic, no comprehensive and concrete overviews of the different types of anomalies have hitherto been published. By means of an extensive literature review this study therefore offers the first theoretically principled and domain-independent typology of data anomalies and presents a full overview of anomaly types and subtypes. To concretely define the concept of the anomaly and its different manifestations, the typology employs five dimensions: data type, cardinality of relationship, anomaly level, data structure, and data distribution. These fundamental and data-centric dimensions naturally yield 3 broad groups, 9 basic types, and 63 subtypes of anomalies. The typology facilitates the evaluation of the functional capabilities of anomaly detection algorithms, contributes to explainable data science, and provides insights into relevant topics such as local versus global anomalies.},
	language = {en},
	number = {4},
	urldate = {2021-11-27},
	journal = {International Journal of Data Science and Analytics},
	author = {Foorthuis, Ralph},
	month = oct,
	year = {2021},
	pages = {297--331},
}

Anomalies are occurrences in a dataset that are in some way unusual and do not fit the general patterns. The concept of the anomaly is typically ill defined and perceived as vague and domain-dependent. Moreover, despite some 250 years of publications on the topic, no comprehensive and concrete overviews of the different types of anomalies have hitherto been published. By means of an extensive literature review this study therefore offers the first theoretically principled and domain-independent typology of data anomalies and presents a full overview of anomaly types and subtypes. To concretely define the concept of the anomaly and its different manifestations, the typology employs five dimensions: data type, cardinality of relationship, anomaly level, data structure, and data distribution. These fundamental and data-centric dimensions naturally yield 3 broad groups, 9 basic types, and 63 subtypes of anomalies. The typology facilitates the evaluation of the functional capabilities of anomaly detection algorithms, contributes to explainable data science, and provides insights into relevant topics such as local versus global anomalies.

@article{tchaghe_anomaly_2021,
	title = {Anomaly explanation: {A} review},
	issn = {0169-023X},
	shorttitle = {Anomaly explanation},
	url = {https://www.sciencedirect.com/science/article/pii/S0169023X21000720},
	doi = {10.1016/j.datak.2021.101946},
	abstract = {Anomaly detection has been studied intensively by the data mining community for several years. As a result, many methods to detect anomalies have emerged, and others are still under development. But during the recent years, anomaly detection, just like a lot of machine learning tasks, is facing a wall. This wall, erected by the lack of trust of the final users, has slowed down the usage of these algorithms in the real-world situations for which they are designed. Having the best empirical accuracy is not enough anymore; there is a need for algorithms to explain their outputs to the users in order to increase their trust. Consequently, a new expression has emerged recently: eXplainable Artificial Intelligence (XAI). This expression, which gathers all the methods that provide explanations to the output of algorithms has gained popularity, especially with the outbreak of deep learning. A lot of work has been devoted to anomaly detection in the literature, but not as much to anomaly explanation. There is so much work on anomaly detection that several reviews can be found on the topic. In contrast, we were not able to find a survey on anomaly explanation in particular, while there are a lot of surveys on XAI in general or on XAI for neural networks for example. With this paper, we want to provide a comprehensive review of the anomaly explanation field. After a brief recall of some important anomaly detection algorithms, the anomaly explanation methods that we discovered in the literature will be classified according to a taxonomy that we define. This taxonomy stems from an analysis of what is really important when trying to explain anomalies.},
	language = {en},
	urldate = {2021-11-26},
	journal = {Data \& Knowledge Engineering},
	author = {Tchaghe, Véronne Yepmo and Smits, Grégory and Pivert, Olivier},
	month = nov,
	year = {2021},
	keywords = {Anomaly detection, Anomaly explanation, Explainable Artificial Intelligence (XAI), Interpretability, Outlier interpretation},
	pages = {101946},
}

Anomaly detection has been studied intensively by the data mining community for several years. As a result, many methods to detect anomalies have emerged, and others are still under development. But during the recent years, anomaly detection, just like a lot of machine learning tasks, is facing a wall. This wall, erected by the lack of trust of the final users, has slowed down the usage of these algorithms in the real-world situations for which they are designed. Having the best empirical accuracy is not enough anymore; there is a need for algorithms to explain their outputs to the users in order to increase their trust. Consequently, a new expression has emerged recently: eXplainable Artificial Intelligence (XAI). This expression, which gathers all the methods that provide explanations to the output of algorithms has gained popularity, especially with the outbreak of deep learning. A lot of work has been devoted to anomaly detection in the literature, but not as much to anomaly explanation. There is so much work on anomaly detection that several reviews can be found on the topic. In contrast, we were not able to find a survey on anomaly explanation in particular, while there are a lot of surveys on XAI in general or on XAI for neural networks for example. With this paper, we want to provide a comprehensive review of the anomaly explanation field. After a brief recall of some important anomaly detection algorithms, the anomaly explanation methods that we discovered in the literature will be classified according to a taxonomy that we define. This taxonomy stems from an analysis of what is really important when trying to explain anomalies.

@inproceedings{le_nguyen_complete_2021,
	title = {A {Complete} {Streaming} {Pipeline} for {Real}-time {Monitoring} and {Predictive} {Maintenance}},
	doi = {10.3850/978-981-18-2016-8_400-cd},
	booktitle = {Proceedings of the 31st {European} {Safety} and {Reliability} {Conference}},
	author = {Le Nguyen, Minh Huong and Turgis, Fabien and Fayemi, Pierre-Emmanuel and Bifet, Albert},
	year = {2021},
	doi = {10.3850/978-981-18-2016-8_400-cd},
	pages = {2119},
}

@article{balbi_relevance_2021,
	series = {17th {IFAC} {Symposium} on {Information} {Control} {Problems} in {Manufacturing} {INCOM} 2021},
	title = {On the relevance of clustering strategies for collaborative prognostics},
	volume = {54},
	issn = {2405-8963},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896321007023},
	doi = {10.1016/j.ifacol.2021.08.004},
	abstract = {The innovative concept of Social Internet of Industrial Things is opening a promising perspective for collaborative prognostics in order to improve maintenance and operational policies. Given this context, the present work studies the exploitation of historical and collaborative information for on-line prognostic assessment. In particular, while aiming at a cost-effective prognostic algorithm, with an efficient use of the available data and a proper prediction accuracy, the work remarks the relevance of an optimized clustering strategy for the selection of the useful information.},
	language = {en},
	number = {1},
	urldate = {2021-11-15},
	journal = {IFAC-PapersOnLine},
	author = {Balbi, Matteo and Cattaneo, Laura and Nucera, Domenico Daniele and Macchi, Marco},
	month = jan,
	year = {2021},
	keywords = {Collaborative prognostics, RUL prediction, clustering, data-driven prognostics},
	pages = {37--42},
}

The innovative concept of Social Internet of Industrial Things is opening a promising perspective for collaborative prognostics in order to improve maintenance and operational policies. Given this context, the present work studies the exploitation of historical and collaborative information for on-line prognostic assessment. In particular, while aiming at a cost-effective prognostic algorithm, with an efficient use of the available data and a proper prediction accuracy, the work remarks the relevance of an optimized clustering strategy for the selection of the useful information.

@article{nucera_data-driven_2021,
	series = {17th {IFAC} {Symposium} on {Information} {Control} {Problems} in {Manufacturing} {INCOM} 2021},
	title = {Data-{Driven} {State} {Detection} for an asset working at heterogenous regimens⁎⁎{This} work is supported by {Lombardy} funded project {SMART4CPPS} ({ID}: 236789 {CUP}: {E19I18000000009})},
	volume = {54},
	issn = {2405-8963},
	shorttitle = {Data-{Driven} {State} {Detection} for an asset working at heterogenous regimens⁎⁎{This} work is supported by {Lombardy} funded project {SMART4CPPS} ({ID}},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896321009137},
	doi = {10.1016/j.ifacol.2021.08.149},
	abstract = {The current trend of industrial digitalization paved the way to Machine Learning applications which are adding value to data coming from the assets. In this context, the case study of a State Detection in an asset characterized by heterogeneous working regimens is proposed, with the aim of automatically recognizing the type of the ongoing production and of identifying its different operating conditions. The activity is executed by exploiting the data available on the asset controller and applying and comparing two different clustering algorithms, namely K-Means and HDBSCAN. The paper describes hence the application case and the adopted approaches, while providing insights on the most preferable choice for any of the two objectives, in order to pave the ground for condition-based maintenance activities.},
	language = {en},
	number = {1},
	urldate = {2021-11-15},
	journal = {IFAC-PapersOnLine},
	author = {Nucera, Domenico Daniele and Quadrini, Walter and Fumagalli, Luca and Scipioni, Marcello Paolo},
	month = jan,
	year = {2021},
	keywords = {Clustering, HDBSCAN, K-Means, Production activity control, Quality assurance, State Detection, maintenance},
	pages = {1248--1253},
}

The current trend of industrial digitalization paved the way to Machine Learning applications which are adding value to data coming from the assets. In this context, the case study of a State Detection in an asset characterized by heterogeneous working regimens is proposed, with the aim of automatically recognizing the type of the ongoing production and of identifying its different operating conditions. The activity is executed by exploiting the data available on the asset controller and applying and comparing two different clustering algorithms, namely K-Means and HDBSCAN. The paper describes hence the application case and the adopted approaches, while providing insights on the most preferable choice for any of the two objectives, in order to pave the ground for condition-based maintenance activities.

@article{wellsandt_towards_2021,
	series = {17th {IFAC} {Symposium} on {Information} {Control} {Problems} in {Manufacturing} {INCOM} 2021},
	title = {Towards {Using} {Digital} {Intelligent} {Assistants} to {Put} {Humans} in the {Loop} of {Predictive} {Maintenance} {Systems}},
	volume = {54},
	issn = {2405-8963},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896321007047},
	doi = {10.1016/j.ifacol.2021.08.005},
	abstract = {Predictive maintenance systems are socio-technical systems where the interaction between maintenance personnel and the technical system is critical to achieving maintenance goals. Employees who use a predictive maintenance system should explore, modify, and verify their analysis and decision-making methods and rules. Conventional modes of interaction make this difficult since they are often hard to understand, obtrusive and unintuitive. Digital Intelligent Assistants (DIAs) provide fast, intuitive, and potentially hands-free access to systems through voice-based interaction and cognitive assistance. This paper introduces a novel approach to interact with predictive maintenance systems through DIAs. The aim is to integrate human knowledge more effectively into the predictive maintenance process to create a hybrid-intelligence system. In such systems, humans and computers complement and evolve together.},
	language = {en},
	number = {1},
	urldate = {2021-11-15},
	journal = {IFAC-PapersOnLine},
	author = {Wellsandt, Stefan and Klein, Konstantin and Hribernik, Karl and Lewandowski, Marco and Bousdekis, Alexandros and Mentzas, Gregoris and Thoben, Klaus-Dieter},
	month = jan,
	year = {2021},
	keywords = {Engineering Applications of Artificial Intelligence, Human-Automation Integration, Hybrid Intelligence Systems, Predictive Maintenance},
	pages = {49--54},
}

Predictive maintenance systems are socio-technical systems where the interaction between maintenance personnel and the technical system is critical to achieving maintenance goals. Employees who use a predictive maintenance system should explore, modify, and verify their analysis and decision-making methods and rules. Conventional modes of interaction make this difficult since they are often hard to understand, obtrusive and unintuitive. Digital Intelligent Assistants (DIAs) provide fast, intuitive, and potentially hands-free access to systems through voice-based interaction and cognitive assistance. This paper introduces a novel approach to interact with predictive maintenance systems through DIAs. The aim is to integrate human knowledge more effectively into the predictive maintenance process to create a hybrid-intelligence system. In such systems, humans and computers complement and evolve together.

@article{davari_survey_2021,
	title = {A {Survey} on {Data}-{Driven} {Predictive} {Maintenance} for the {Railway} {Industry}},
	volume = {21},
	doi = {10.3390/s21175739},
	abstract = {In the last few years, many works have addressed Predictive Maintenance (PdM) by the use of Machine Learning (ML) and Deep Learning (DL) solutions, especially the latter. The monitoring and logging of industrial equipment events, like temporal behavior and fault events-anomaly detection in time-series-can be obtained from records generated by sensors installed in different parts of an industrial plant. However, such progress is incipient because we still have many challenges, and the performance of applications depends on the appropriate choice of the method. This article presents a survey of existing ML and DL techniques for handling PdM in the railway industry. This survey discusses the main approaches for this specific application within a taxonomy defined by the type of task, employed methods, metrics of evaluation, the specific equipment or process, and datasets. Lastly, we conclude and outline some suggestions for future research.},
	journal = {Sensors},
	author = {Davari, Narjes and Veloso, Bruno and De Assis Costa, Gustavo and Pereira, Pedro and Ribeiro, Rita and Gama, João},
	month = sep,
	year = {2021},
	keywords = {data-driven pdm, pdm, railway, survey},
	pages = {5739},
}

In the last few years, many works have addressed Predictive Maintenance (PdM) by the use of Machine Learning (ML) and Deep Learning (DL) solutions, especially the latter. The monitoring and logging of industrial equipment events, like temporal behavior and fault events-anomaly detection in time-series-can be obtained from records generated by sensors installed in different parts of an industrial plant. However, such progress is incipient because we still have many challenges, and the performance of applications depends on the appropriate choice of the method. This article presents a survey of existing ML and DL techniques for handling PdM in the railway industry. This survey discusses the main approaches for this specific application within a taxonomy defined by the type of task, employed methods, metrics of evaluation, the specific equipment or process, and datasets. Lastly, we conclude and outline some suggestions for future research.

@article{michalowska_anomaly_2021,
	series = {13th {IFAC} {Conference} on {Control} {Applications} in {Marine} {Systems}, {Robotics}, and {Vehicles} {CAMS} 2021},
	title = {Anomaly {Detection} with {Unknown} {Anomalies}: {Application} to {Maritime} {Machinery}},
	volume = {54},
	issn = {2405-8963},
	shorttitle = {Anomaly {Detection} with {Unknown} {Anomalies}},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896321014828},
	doi = {10.1016/j.ifacol.2021.10.080},
	abstract = {We present a framework for deriving anomaly detection algorithms on timeseries data when the time and expression of anomalous behaviour is unknown. The framework is suited for problems in which individual machine learning paradigms cannot be directly implemented: supervised learning is not applicable due to the lack of labelled data, unsupervised learning is not effective since the normal operations are insufficiently defined and take complex and diverse forms, and deep learning risks confusing problematic behaviours for expected ones due to the possible repetitiveness of similar anomalies. The proposed approach is comprised of two phases: unsupervised discovery of anomalies, and semi-supervised construction and tuning of the anomaly detection algorithm. By leveraging data exploration methods and expert knowledge, the resulting algorithms are interpretable and detect a wide range of anomalous behaviours. The approach is applied to the early detection of wear and tear of maritime propulsion and manoeuvring machinery. We show that the final algorithm is able to detect different types of anomalies, including an actual internal leakage in a thruster which is otherwise overlooked by the present rule-based alarm system.},
	language = {en},
	number = {16},
	urldate = {2021-11-08},
	journal = {IFAC-PapersOnLine},
	author = {Michałowska, Katarzyna and Riemer-Sørensen, Signe and Sterud, Camilla and Hjellset, Ole Magnus},
	month = jan,
	year = {2021},
	keywords = {anomaly detection, condition-based monitoring, diagnosis, fault detection, grey-box modelling, machine learning, predictive maintenance},
	pages = {105--111},
}

We present a framework for deriving anomaly detection algorithms on timeseries data when the time and expression of anomalous behaviour is unknown. The framework is suited for problems in which individual machine learning paradigms cannot be directly implemented: supervised learning is not applicable due to the lack of labelled data, unsupervised learning is not effective since the normal operations are insufficiently defined and take complex and diverse forms, and deep learning risks confusing problematic behaviours for expected ones due to the possible repetitiveness of similar anomalies. The proposed approach is comprised of two phases: unsupervised discovery of anomalies, and semi-supervised construction and tuning of the anomaly detection algorithm. By leveraging data exploration methods and expert knowledge, the resulting algorithms are interpretable and detect a wide range of anomalous behaviours. The approach is applied to the early detection of wear and tear of maritime propulsion and manoeuvring machinery. We show that the final algorithm is able to detect different types of anomalies, including an actual internal leakage in a thruster which is otherwise overlooked by the present rule-based alarm system.

@article{cai_novel_2021,
	title = {A novel {RUL} prognosis methodology of multilevel system with cascading failure: {Subsea} oil and gas transportation systems as a case study},
	volume = {242},
	issn = {0029-8018},
	shorttitle = {A novel {RUL} prognosis methodology of multilevel system with cascading failure},
	url = {https://www.sciencedirect.com/science/article/pii/S0029801821014621},
	doi = {10.1016/j.oceaneng.2021.110141},
	abstract = {Cascading failure has a great negative impact on the operation of multilevel systems. Although the frequency of this kind of failure is lower than that of general failure, it may cause outage and considerable human and economic losses. In this paper, a novel modeling methodology of cascading failure based on position importance and function importance is proposed by using dynamic Bayesian networks, and the remaining useful life (RUL) of multilevel systems considering cascading failure is estimated. By detecting the working state of the nodes, the position and function importance of the nodes are determined, and the performance of the next layer of related nodes is calculated. Through calculating iteratively to the last level, the number of failure nodes is finally determined, and the overall performance of the multilevel systems is evaluated. A subsea transportation system with three-level network and four nodes in each level is used to demonstrate the application of the proposed methodology, and the feasibility of the methodology is analyzed. The results show that the RUL of the systems is significantly reduced when considering the cascading failure, which is quite different from the degradation of the non-cascading failure mode.},
	language = {en},
	urldate = {2021-11-08},
	journal = {Ocean Engineering},
	author = {Cai, Baoping and Shao, Xiaoyan and Yuan, Xiaobing and Liu, Yonghong and Chen, Guoming and Feng, Qiang and Liu, Yiqi and Ren, Yi},
	month = dec,
	year = {2021},
	keywords = {Cascading failure, Dynamic Bayesian network, Multilevel systems, Position and function importance, Remaining useful life},
	pages = {110141},
}

Cascading failure has a great negative impact on the operation of multilevel systems. Although the frequency of this kind of failure is lower than that of general failure, it may cause outage and considerable human and economic losses. In this paper, a novel modeling methodology of cascading failure based on position importance and function importance is proposed by using dynamic Bayesian networks, and the remaining useful life (RUL) of multilevel systems considering cascading failure is estimated. By detecting the working state of the nodes, the position and function importance of the nodes are determined, and the performance of the next layer of related nodes is calculated. Through calculating iteratively to the last level, the number of failure nodes is finally determined, and the overall performance of the multilevel systems is evaluated. A subsea transportation system with three-level network and four nodes in each level is used to demonstrate the application of the proposed methodology, and the feasibility of the methodology is analyzed. The results show that the RUL of the systems is significantly reduced when considering the cascading failure, which is quite different from the degradation of the non-cascading failure mode.

@phdthesis{aleskog_graph-based_2021,
	address = {Sweden},
	title = {Graph-based {Multi}-view {Clustering} for {Continuous} {Pattern} {Mining}},
	url = {http://urn.kb.se/resolve?urn=urn:nbn:se:bth-21850},
	abstract = {Background. In many smart monitoring applications, such as smart healthcare, smart building, autonomous cars etc., data are collected from multiple sources and contain information about different perspectives/views of the monitored phenomenon, physical object, system. In addition, in many of those applications the availability of relevant labelled data is often low or even non-existing. Inspired by this, in this thesis study we propose a novel algorithm for multi-view stream clustering. The algorithm can be applied for continuous pattern mining and labeling of streaming data.

Objectives. The main objective of this thesis is to develop and implement a novel multi-view stream clustering algorithm. In addition, the potential of the proposed algorithm is studied and evaluated on two datasets: synthetic and real-world. The conducted experiments study the new algorithm’s performance compared to a single-view clustering algorithm and an algorithm without transferring knowledge between chunks. Finally, the obtained results are analyzed, discussed and interpreted.

Methods. Initially, we study the state-of-the-art multi-view (stream) clustering algorithms. Then we develop our multi-view clustering algorithm for streaming data by implementing transfer of knowledge feature. We present and explain in details the developed algorithm by motivating each choice made during the algorithm design phase. Finally, discussion of the algorithm configuration, experimental setup and the datasets chosen for the experiments are presented and motivated.

Results. Different configurations of the proposed algorithm have been studied and evaluated under different experimental scenarios on two different datasets: synthetic and real-world. The proposed multi-view clustering algorithm has demonstrated higher performance on the synthetic data than on the real-world dataset. This is mainly due to not very good quality of the used real-world data.

Conclusions. The proposed algorithm has demonstrated higher performance results on the synthetic dataset than on the real-world dataset. It can generate high-quality clustering solutions with respect to the used evaluation metrics. In addition, the transfer of knowledge feature has been shown to have a positive effect on the algorithm performance. A further study of the proposed algorithm on other richer and more suitable datasets, e.g., data collected from numerous sensors used for monitoring some phenomenon, is planned to be conducted in the future work.},
	language = {eng},
	urldate = {2021-11-07},
	school = {Blekinge Insitute of Technology},
	author = {Åleskog, Christoffer},
	year = {2021},
}

Background. In many smart monitoring applications, such as smart healthcare, smart building, autonomous cars etc., data are collected from multiple sources and contain information about different perspectives/views of the monitored phenomenon, physical object, system. In addition, in many of those applications the availability of relevant labelled data is often low or even non-existing. Inspired by this, in this thesis study we propose a novel algorithm for multi-view stream clustering. The algorithm can be applied for continuous pattern mining and labeling of streaming data. Objectives. The main objective of this thesis is to develop and implement a novel multi-view stream clustering algorithm. In addition, the potential of the proposed algorithm is studied and evaluated on two datasets: synthetic and real-world. The conducted experiments study the new algorithm’s performance compared to a single-view clustering algorithm and an algorithm without transferring knowledge between chunks. Finally, the obtained results are analyzed, discussed and interpreted. Methods. Initially, we study the state-of-the-art multi-view (stream) clustering algorithms. Then we develop our multi-view clustering algorithm for streaming data by implementing transfer of knowledge feature. We present and explain in details the developed algorithm by motivating each choice made during the algorithm design phase. Finally, discussion of the algorithm configuration, experimental setup and the datasets chosen for the experiments are presented and motivated. Results. Different configurations of the proposed algorithm have been studied and evaluated under different experimental scenarios on two different datasets: synthetic and real-world. The proposed multi-view clustering algorithm has demonstrated higher performance on the synthetic data than on the real-world dataset. This is mainly due to not very good quality of the used real-world data. Conclusions. The proposed algorithm has demonstrated higher performance results on the synthetic dataset than on the real-world dataset. It can generate high-quality clustering solutions with respect to the used evaluation metrics. In addition, the transfer of knowledge feature has been shown to have a positive effect on the algorithm performance. A further study of the proposed algorithm on other richer and more suitable datasets, e.g., data collected from numerous sensors used for monitoring some phenomenon, is planned to be conducted in the future work.

@article{ang_efficient_2021,
	title = {Efficient linear predictive model with short term features for lithium-ion batteries state of health estimation},
	volume = {44},
	issn = {2352-152X},
	url = {https://www.sciencedirect.com/science/article/pii/S2352152X21010951},
	doi = {10.1016/j.est.2021.103409},
	abstract = {The need to predict the State of Health (SoH) of lithium-ion batteries accurately and efficiently is rising with growing use of such batteries in safety critical applications. In this paper, an intuitive and efficient predictive algorithm that can estimate the SoH of lithium-ion batteries with accuracy on par with more complex and computationally demanding models is presented. The predictive algorithm uses the battery's temperature and voltage time discharge profile to predict its current SoH with root mean square error (RMSE) of 1\%. It is shown that the crux in achieving good prediction accuracy lies in data preprocessing, which are cleansing, normalization and retaining of key features that are rich in information from the raw measured data. Finally, a simplified version of the algorithm using only voltage time profiles for model training is proposed that provides less than 12\% RMSE error, comparable with current state of the art. This algorithm can be easily implemented in most applications since only measured voltage data is required. Throughout this paper, the proposed algorithm is tested with publicly available dataset and comparison is done with existing literature results to benchmark the proposed algorithm performance.},
	language = {en},
	urldate = {2021-10-25},
	journal = {Journal of Energy Storage},
	author = {Ang, Elisa Y. M. and Paw, Yew Chai},
	month = dec,
	year = {2021},
	keywords = {Battery state of health, Data analytics, Linear predictive model, Lithium-ion battery, Machine learning, Predictive maintenance},
	pages = {103409},
}

The need to predict the State of Health (SoH) of lithium-ion batteries accurately and efficiently is rising with growing use of such batteries in safety critical applications. In this paper, an intuitive and efficient predictive algorithm that can estimate the SoH of lithium-ion batteries with accuracy on par with more complex and computationally demanding models is presented. The predictive algorithm uses the battery's temperature and voltage time discharge profile to predict its current SoH with root mean square error (RMSE) of 1%. It is shown that the crux in achieving good prediction accuracy lies in data preprocessing, which are cleansing, normalization and retaining of key features that are rich in information from the raw measured data. Finally, a simplified version of the algorithm using only voltage time profiles for model training is proposed that provides less than 12% RMSE error, comparable with current state of the art. This algorithm can be easily implemented in most applications since only measured voltage data is required. Throughout this paper, the proposed algorithm is tested with publicly available dataset and comparison is done with existing literature results to benchmark the proposed algorithm performance.

@article{dikshit_explainable_2021,
	title = {Explainable {AI} in drought forecasting},
	issn = {2666-8270},
	url = {https://www.sciencedirect.com/science/article/pii/S2666827021000967},
	doi = {10.1016/j.mlwa.2021.100192},
	abstract = {Droughts are one of the disastrous natural hazards which has severe impacts on agricultural production, economy, and society. One of the critical steps for effective drought management is developing a robust forecasting model and understanding how the variables affect the model outcomes. The present study forecasts SPI-12 at a lead time of 3 months, using the Long Short-Term Memory (LSTM) model, and further interprets the spatial and temporal relationship between variables and forecasting results using SHapley Additive exPlanations (SHAP). The developed model is tested in four different regions in New South Wales (NSW), Australia. SPI-12 was computed using monthly rainfall data collected from Scientific Information for Land Owners (SILO) for 1901–2018. The model was trained from 1901–2000 and tested from 2001–2018, and the performance was measured using Coefficient of Determination (R2), Nash–SutcliffeEfficiency (NSE) and Root-Mean-Square-Error (RMSE). To understand the underlying impact of variables on the model outcomes, SHAPley values were calculated for the entire testing period and also at three different temporal ranges, which are during the Millennium Drought (2001–2010), post drought period (2011–2018) and at a seasonal scale (summer months). The comparison of the results shows a significant variation in the impact of variables on forecasting, both temporally and spatially. It also shows the need to study the model outcomes for specific regions and for a shorter duration than the entire testing period. This is a first of its study towards interpreting the forecasting model in drought studies, which could help understand the behaviour of drought variables.},
	language = {en},
	urldate = {2021-10-25},
	journal = {Machine Learning with Applications},
	author = {Dikshit, Abhirup and Pradhan, Biswajeet},
	month = oct,
	year = {2021},
	keywords = {Deep learning, Drought forecasting, Explainable AI, Standard precipitation index},
	pages = {100192},
}

Droughts are one of the disastrous natural hazards which has severe impacts on agricultural production, economy, and society. One of the critical steps for effective drought management is developing a robust forecasting model and understanding how the variables affect the model outcomes. The present study forecasts SPI-12 at a lead time of 3 months, using the Long Short-Term Memory (LSTM) model, and further interprets the spatial and temporal relationship between variables and forecasting results using SHapley Additive exPlanations (SHAP). The developed model is tested in four different regions in New South Wales (NSW), Australia. SPI-12 was computed using monthly rainfall data collected from Scientific Information for Land Owners (SILO) for 1901–2018. The model was trained from 1901–2000 and tested from 2001–2018, and the performance was measured using Coefficient of Determination (R2), Nash–SutcliffeEfficiency (NSE) and Root-Mean-Square-Error (RMSE). To understand the underlying impact of variables on the model outcomes, SHAPley values were calculated for the entire testing period and also at three different temporal ranges, which are during the Millennium Drought (2001–2010), post drought period (2011–2018) and at a seasonal scale (summer months). The comparison of the results shows a significant variation in the impact of variables on forecasting, both temporally and spatially. It also shows the need to study the model outcomes for specific regions and for a shorter duration than the entire testing period. This is a first of its study towards interpreting the forecasting model in drought studies, which could help understand the behaviour of drought variables.

@inproceedings{mistry_aedbscanadaptive_2021,
	address = {Singapore},
	series = {Advances in {Intelligent} {Systems} and {Computing}},
	title = {{AEDBSCAN}—{Adaptive} {Epsilon} {Density}-{Based} {Spatial} {Clustering} of {Applications} with {Noise}},
	isbn = {9789811563539},
	doi = {10.1007/978-981-15-6353-9_20},
	abstract = {The objectives of this research are related to study the DBSCAN algorithm and engineer an enhancement to this algorithm addressing its flaws. DBSCAN is criticized for its requirement to input two parameters, namely—epsilon radius (ϵ) and minimum number of points (MinPts). It is difficult to know beforehand the optimum value of both parameters, and hence many trials are required until desired clusters are obtained. Also, in a dataset, a cluster’s density can vary. DBSCAN fails to identify clusters with density variations present. The proposed algorithm Adaptive Epsilon DBSCAN (AEDBSCAN), generates epsilon dynamically in accordance with the neighborhood of a point and thereafter adopts DBSCAN clustering with the corresponding epsilon to obtain the clusters. Experimental results are obtained from testing AEDBSCAN on artificial datasets. The experimental results confirm that the proposed AEDBSCAN algorithm efficiently carries out multi-density clustering than the original DBSCAN.},
	language = {en},
	booktitle = {Progress in {Advanced} {Computing} and {Intelligent} {Engineering}},
	publisher = {Springer},
	author = {Mistry, Vidhi and Pandya, Urja and Rathwa, Anjana and Kachroo, Himani and Jivani, Anjali},
	editor = {Panigrahi, Chhabi Rani and Pati, Bibudhendu and Mohapatra, Prasant and Buyya, Rajkumar and Li, Kuan-Ching},
	year = {2021},
	keywords = {Adaptive epsilon, DBSCAN, Data mining, Density-based clustering, Multi-density},
	pages = {213--226},
}

The objectives of this research are related to study the DBSCAN algorithm and engineer an enhancement to this algorithm addressing its flaws. DBSCAN is criticized for its requirement to input two parameters, namely—epsilon radius (ϵ) and minimum number of points (MinPts). It is difficult to know beforehand the optimum value of both parameters, and hence many trials are required until desired clusters are obtained. Also, in a dataset, a cluster’s density can vary. DBSCAN fails to identify clusters with density variations present. The proposed algorithm Adaptive Epsilon DBSCAN (AEDBSCAN), generates epsilon dynamically in accordance with the neighborhood of a point and thereafter adopts DBSCAN clustering with the corresponding epsilon to obtain the clusters. Experimental results are obtained from testing AEDBSCAN on artificial datasets. The experimental results confirm that the proposed AEDBSCAN algorithm efficiently carries out multi-density clustering than the original DBSCAN.

@article{zhang_multi-label_2021,
	title = {Multi-label learning with label-specific features via weighting and label entropy guided clustering ensemble},
	volume = {419},
	issn = {0925-2312},
	url = {https://www.sciencedirect.com/science/article/pii/S0925231220313059},
	doi = {10.1016/j.neucom.2020.07.107},
	abstract = {Multi-label learning has attracted more and more researchers’ attention. It deals with the problem where each instance is associated with multiple labels simultaneously. Some methods improve the performance by constructing label-specific features. Specifically, the LIFTACE method constructs label-specific features by clustering ensemble techniques, which ignores the importance of label vectors and does not explore label correlations when constructing the classification model. In this paper, we propose a multi-label learning method called LF-LELC, which considers the importance of label vectors and constructs the classification model by considering label correlations. Firstly, it performs clustering on the positive instances and negative instances respectively. The number of clusters is set by the information contained in the label vectors. After that, it employs clustering ensemble techniques that consider label correlations to make the clustering results more stable and effective. Then, it constructs label-specific features for each label. Finally, it builds the classification model by exploring label correlations. The label set for each test example is predicted by the classification model. Experiments show that LF-LELC can achieve better performance by considering the importance of label vectors and the correlations among labels.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Neurocomputing},
	author = {Zhang, Chunyu and Li, Zhanshan},
	month = jan,
	year = {2021},
	keywords = {Label correlation, Label entropy, Label-specific features, Multi-label learning},
	pages = {59--69},
}

Multi-label learning has attracted more and more researchers’ attention. It deals with the problem where each instance is associated with multiple labels simultaneously. Some methods improve the performance by constructing label-specific features. Specifically, the LIFTACE method constructs label-specific features by clustering ensemble techniques, which ignores the importance of label vectors and does not explore label correlations when constructing the classification model. In this paper, we propose a multi-label learning method called LF-LELC, which considers the importance of label vectors and constructs the classification model by considering label correlations. Firstly, it performs clustering on the positive instances and negative instances respectively. The number of clusters is set by the information contained in the label vectors. After that, it employs clustering ensemble techniques that consider label correlations to make the clustering results more stable and effective. Then, it constructs label-specific features for each label. Finally, it builds the classification model by exploring label correlations. The label set for each test example is predicted by the classification model. Experiments show that LF-LELC can achieve better performance by considering the importance of label vectors and the correlations among labels.

@article{phan_dataset_2021,
	title = {Dataset of {Vietnamese} students’ academic perfectionism and school alienation},
	volume = {39},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340921007459},
	doi = {10.1016/j.dib.2021.107463},
	abstract = {Across the steadily superseding world, the younger generation is coming under pressure for an increase in the standard and a highly growing demand on their life themselves. This could lead to a variety of problems, including academic perfectionism and school alienation. To gain more insights into these phenomena, we conducted two research projects on students from eight upper secondary schools in Ho Chi Minh City, Vietnam using online surveys, and obtained two datasets. Dataset A covers (i) the level of students' perfectionism; (ii) belief in school meritocracy; (iii) The competitiveness among students; and (iv) the intrinsic motivation to achieve. Dataset B contains students' self-reports about (i) their perceptions of parents' and teachers' academic conditional regard; (ii) academic contingent self-worth; and (iii) school alienation. The numbers of respondents of dataset A and dataset B are 2942 and 2970, respectively.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Data in Brief},
	author = {Phan, Thanh-Thao Thi and Nguyen, Linh-Chi and Nguyen, Ngoc-Quang and Nguyen, Yen-Chi},
	month = dec,
	year = {2021},
	keywords = {Classroom competitiveness, Contingent self-worth, Meritocracy belief, Parental conditional regard, School alienation, Student academic perfectionism, Teacher conditional regard, Vietnam},
	pages = {107463},
}

Across the steadily superseding world, the younger generation is coming under pressure for an increase in the standard and a highly growing demand on their life themselves. This could lead to a variety of problems, including academic perfectionism and school alienation. To gain more insights into these phenomena, we conducted two research projects on students from eight upper secondary schools in Ho Chi Minh City, Vietnam using online surveys, and obtained two datasets. Dataset A covers (i) the level of students' perfectionism; (ii) belief in school meritocracy; (iii) The competitiveness among students; and (iv) the intrinsic motivation to achieve. Dataset B contains students' self-reports about (i) their perceptions of parents' and teachers' academic conditional regard; (ii) academic contingent self-worth; and (iii) school alienation. The numbers of respondents of dataset A and dataset B are 2942 and 2970, respectively.

@article{hoi_online_2021,
	title = {Online learning: {A} comprehensive survey},
	volume = {459},
	issn = {0925-2312},
	shorttitle = {Online learning},
	url = {https://www.sciencedirect.com/science/article/pii/S0925231221006706},
	doi = {10.1016/j.neucom.2021.04.112},
	abstract = {Online learning represents a family of machine learning methods, where a learner attempts to tackle some predictive (or any type of decision-making) task by learning from a sequence of data instances one by one at each time. The goal of online learning is to maximize the accuracy/correctness for the sequence of predictions/decisions made by the online learner given the knowledge of correct answers to previous prediction/learning tasks and possibly additional information. This is in contrast to traditional batch or offline machine learning methods that are often designed to learn a model from the entire training data set at once. Online learning has become a promising technique for learning from continuous streams of data in many real-world applications. This survey aims to provide a comprehensive survey of the online machine learning literature through a systematic review of basic ideas and key principles and a proper categorization of different algorithms and techniques. Generally speaking, according to the types of learning tasks and the forms of feedback information, the existing online learning works can be classified into three major categories: (i) online supervised learning where full feedback information is always available, (ii) online learning with limited feedback, and (iii) online unsupervised learning where no feedback is available. Due to space limitation, the survey will be mainly focused on the first category, but also briefly cover some basics of the other two categories. Finally, we also discuss some open issues and attempt to shed light on potential future research directions in this field.},
	language = {en},
	urldate = {2021-10-15},
	journal = {Neurocomputing},
	author = {Hoi, Steven C. H. and Sahoo, Doyen and Lu, Jing and Zhao, Peilin},
	month = oct,
	year = {2021},
	keywords = {Online convex optimization, Online learning, Sequential decision making},
	pages = {249--289},
}

Online learning represents a family of machine learning methods, where a learner attempts to tackle some predictive (or any type of decision-making) task by learning from a sequence of data instances one by one at each time. The goal of online learning is to maximize the accuracy/correctness for the sequence of predictions/decisions made by the online learner given the knowledge of correct answers to previous prediction/learning tasks and possibly additional information. This is in contrast to traditional batch or offline machine learning methods that are often designed to learn a model from the entire training data set at once. Online learning has become a promising technique for learning from continuous streams of data in many real-world applications. This survey aims to provide a comprehensive survey of the online machine learning literature through a systematic review of basic ideas and key principles and a proper categorization of different algorithms and techniques. Generally speaking, according to the types of learning tasks and the forms of feedback information, the existing online learning works can be classified into three major categories: (i) online supervised learning where full feedback information is always available, (ii) online learning with limited feedback, and (iii) online unsupervised learning where no feedback is available. Due to space limitation, the survey will be mainly focused on the first category, but also briefly cover some basics of the other two categories. Finally, we also discuss some open issues and attempt to shed light on potential future research directions in this field.

@article{lv_application_2021,
	title = {Application of {Multilayer} {Network} {Models} in {Bioinformatics}},
	volume = {12},
	issn = {1664-8021},
	url = {https://www.frontiersin.org/article/10.3389/fgene.2021.664860},
	doi = {10.3389/fgene.2021.664860},
	abstract = {Multilayer networks provide an efficient tool for studying complex systems, and with current, dramatic development of bioinformatics tools and accumulation of data, researchers have applied network concepts to all aspects of research problems in the field of biology. Addressing the combination of multilayer networks and bioinformatics, through summarizing the applications of multilayer network models in bioinformatics, this review classifies applications and presents a summary of the latest results. Among them, we classify the applications of multilayer networks according to the object of study. Furthermore, because of the systemic nature of biology, we classify the subjects into several hierarchical categories, such as cells, tissues, organs, and groups, according to the hierarchical nature of biological composition. On the basis of the complexity of biological systems, we selected brain research for a detailed explanation. We describe the application of multilayer networks and chronological networks in brain research to demonstrate the primary ideas associated with the application of multilayer networks in biological studies. Finally, we mention a quality assessment method focusing on multilayer and single-layer networks as an evaluation method emphasizing network studies.},
	urldate = {2021-10-14},
	journal = {Frontiers in Genetics},
	author = {Lv, Yuanyuan and Huang, Shan and Zhang, Tianjiao and Gao, Bo},
	year = {2021},
	pages = {380},
}

Multilayer networks provide an efficient tool for studying complex systems, and with current, dramatic development of bioinformatics tools and accumulation of data, researchers have applied network concepts to all aspects of research problems in the field of biology. Addressing the combination of multilayer networks and bioinformatics, through summarizing the applications of multilayer network models in bioinformatics, this review classifies applications and presents a summary of the latest results. Among them, we classify the applications of multilayer networks according to the object of study. Furthermore, because of the systemic nature of biology, we classify the subjects into several hierarchical categories, such as cells, tissues, organs, and groups, according to the hierarchical nature of biological composition. On the basis of the complexity of biological systems, we selected brain research for a detailed explanation. We describe the application of multilayer networks and chronological networks in brain research to demonstrate the primary ideas associated with the application of multilayer networks in biological studies. Finally, we mention a quality assessment method focusing on multilayer and single-layer networks as an evaluation method emphasizing network studies.

@article{galvez_fault_2021,
	title = {Fault {Detection} and {RUL} {Estimation} for {Railway} {HVAC} {Systems} {Using} a {Hybrid} {Model}-{Based} {Approach}},
	volume = {13},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	url = {https://www.mdpi.com/2071-1050/13/12/6828},
	doi = {10.3390/su13126828},
	abstract = {Heating, ventilation, and air conditioning (HVAC) systems installed in a passenger train carriage are critical systems, whose failures can affect people or the environment. This, together with restrictive regulations, results in the replacement of critical components in initial stages of degradation, as well as a lack of data on advanced stages of degradation. This paper proposes a hybrid model-based approach (HyMA) to overcome the lack of failure data on a HVAC system installed in a passenger train carriage. The proposed HyMA combines physics-based models with data-driven models to deploy diagnostic and prognostic processes for a complex and critical system. The physics-based model generates data on healthy and faulty working conditions; the faults are generated in different levels of degradation and can appear individually or together. A fusion of synthetic data and measured data is used to train, validate, and test the proposed hybrid model (HyM) for fault detection and diagnostics (FDD) of the HVAC system. The model obtains an accuracy of 92.60\%. In addition, the physics-based model generates run-to-failure data for the HVAC air filter to develop a remaining useful life (RUL) prediction model, the RUL estimations performed obtained an accuracy in the range of 95.21–97.80\% Both models obtain a remarkable accuracy. The development presented will result in a tool which provides relevant information on the health state of the HVAC system, extends its useful life, reduces its life cycle cost, and improves its reliability and availability; thus enhancing the sustainability of the system.},
	language = {en},
	number = {12},
	urldate = {2021-10-12},
	journal = {Sustainability},
	author = {Gálvez, Antonio and Diez-Olivan, Alberto and Seneviratne, Dammika and Galar, Diego},
	month = jan,
	year = {2021},
	note = {Number: 12
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {HVAC systems, fault detection, fault modelling, hvac, hybrid modelling, predictive maintenance, railway, rul, soft sensing, synthetic data},
	pages = {6828},
}

Heating, ventilation, and air conditioning (HVAC) systems installed in a passenger train carriage are critical systems, whose failures can affect people or the environment. This, together with restrictive regulations, results in the replacement of critical components in initial stages of degradation, as well as a lack of data on advanced stages of degradation. This paper proposes a hybrid model-based approach (HyMA) to overcome the lack of failure data on a HVAC system installed in a passenger train carriage. The proposed HyMA combines physics-based models with data-driven models to deploy diagnostic and prognostic processes for a complex and critical system. The physics-based model generates data on healthy and faulty working conditions; the faults are generated in different levels of degradation and can appear individually or together. A fusion of synthetic data and measured data is used to train, validate, and test the proposed hybrid model (HyM) for fault detection and diagnostics (FDD) of the HVAC system. The model obtains an accuracy of 92.60%. In addition, the physics-based model generates run-to-failure data for the HVAC air filter to develop a remaining useful life (RUL) prediction model, the RUL estimations performed obtained an accuracy in the range of 95.21–97.80% Both models obtain a remarkable accuracy. The development presented will result in a tool which provides relevant information on the health state of the HVAC system, extends its useful life, reduces its life cycle cost, and improves its reliability and availability; thus enhancing the sustainability of the system.

@article{ciani_condition-based_2021,
	title = {Condition-{Based} {Maintenance} of {HVAC} on a {High}-{Speed} {Train} for {Fault} {Detection}},
	volume = {10},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	url = {https://www.mdpi.com/2079-9292/10/12/1418},
	doi = {10.3390/electronics10121418},
	abstract = {Reliability-centered maintenance (RCM) is a well-established method for preventive maintenance planning. This paper focuses on the optimization of a maintenance plan for an HVAC (heating, ventilation and air conditioning) system located on high-speed trains. The first steps of the RCM procedure help in identifying the most critical items of the system in terms of safety and availability by means of a failure modes and effects analysis. Then, RMC proposes the optimal maintenance tasks for each item making up the system. However, the decision-making diagram that leads to the maintenance choice is extremely generic, with a consequent high subjectivity in the task selection. This paper proposes a new fuzzy-based decision-making diagram to minimize the subjectivity of the task choice and preserve the cost-efficiency of the procedure. It uses a case from the railway industry to illustrate the suggested approach, but the procedure could be easily applied to different industrial and technological fields. The results of the proposed fuzzy approach highlight the importance of an accurate diagnostics (with an overall 86\% of the task as diagnostic-based maintenance) and condition monitoring strategy (covering 54\% of the tasks) to optimize the maintenance plan and to minimize the system availability. The findings show that the framework strongly mitigates the issues related to the classical RCM procedure, notably the high subjectivity of experts. It lays the groundwork for a general fuzzy-based reliability-centered maintenance method.},
	language = {en},
	number = {12},
	urldate = {2021-10-12},
	journal = {Electronics},
	author = {Ciani, Lorenzo and Guidi, Giulia and Patrizi, Gabriele and Galar, Diego},
	month = jan,
	year = {2021},
	note = {Number: 12
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {cbm, condition-based maintenance, fault detection, fuzzy logic, hvac, railway, rcm, reliability, reliability-centered maintenance},
	pages = {1418},
}

Reliability-centered maintenance (RCM) is a well-established method for preventive maintenance planning. This paper focuses on the optimization of a maintenance plan for an HVAC (heating, ventilation and air conditioning) system located on high-speed trains. The first steps of the RCM procedure help in identifying the most critical items of the system in terms of safety and availability by means of a failure modes and effects analysis. Then, RMC proposes the optimal maintenance tasks for each item making up the system. However, the decision-making diagram that leads to the maintenance choice is extremely generic, with a consequent high subjectivity in the task selection. This paper proposes a new fuzzy-based decision-making diagram to minimize the subjectivity of the task choice and preserve the cost-efficiency of the procedure. It uses a case from the railway industry to illustrate the suggested approach, but the procedure could be easily applied to different industrial and technological fields. The results of the proposed fuzzy approach highlight the importance of an accurate diagnostics (with an overall 86% of the task as diagnostic-based maintenance) and condition monitoring strategy (covering 54% of the tasks) to optimize the maintenance plan and to minimize the system availability. The findings show that the framework strongly mitigates the issues related to the classical RCM procedure, notably the high subjectivity of experts. It lays the groundwork for a general fuzzy-based reliability-centered maintenance method.

@article{tan_multi-label_2021,
	title = {Multi-label classification for simultaneous fault diagnosis of marine machinery: {A} comparative study},
	volume = {239},
	issn = {0029-8018},
	shorttitle = {Multi-label classification for simultaneous fault diagnosis of marine machinery},
	url = {https://www.sciencedirect.com/science/article/pii/S0029801821010921},
	doi = {10.1016/j.oceaneng.2021.109723},
	abstract = {Fault diagnosis of marine machinery is of utmost importance in modern ships. The widely used machine learning techniques have made it possible to realize intelligent diagnosis by using large amounts of sensory data. However, the detection of simultaneous faults is still a challenge in the absence of simultaneous fault data. Multi-label classification has recently gained popularity in simultaneous fault diagnosis with promising results. The contribution of this work is to carry out a comparative study of several state-of-the-art multi-label classification algorithms for simultaneous fault diagnosis of marine machinery based on single fault data. The proposed method is experimentally validated with a dataset generated from a real data validated simulator of a Frigate. The experimental results show the effectiveness of the proposed method, which can provide decision support for the application of multi-label classification in the simultaneous fault diagnosis of similar marine systems.},
	language = {en},
	urldate = {2021-10-11},
	journal = {Ocean Engineering},
	author = {Tan, Yanghui and Zhang, Jundong and Tian, Hui and Jiang, Dingyu and Guo, Lei and Wang, Gaoming and Lin, Yejin},
	month = nov,
	year = {2021},
	keywords = {Marine machinery, Multi-label classification, Simultaneous fault diagnosis},
	pages = {109723},
}

Fault diagnosis of marine machinery is of utmost importance in modern ships. The widely used machine learning techniques have made it possible to realize intelligent diagnosis by using large amounts of sensory data. However, the detection of simultaneous faults is still a challenge in the absence of simultaneous fault data. Multi-label classification has recently gained popularity in simultaneous fault diagnosis with promising results. The contribution of this work is to carry out a comparative study of several state-of-the-art multi-label classification algorithms for simultaneous fault diagnosis of marine machinery based on single fault data. The proposed method is experimentally validated with a dataset generated from a real data validated simulator of a Frigate. The experimental results show the effectiveness of the proposed method, which can provide decision support for the application of multi-label classification in the simultaneous fault diagnosis of similar marine systems.

@article{manjunath_time-distributed_2021,
	title = {Time-{Distributed} {Feature} {Learning} in {Network} {Traffic} {Classification} for {Internet} of {Things}},
	url = {http://arxiv.org/abs/2109.14696},
	abstract = {The plethora of Internet of Things (IoT) devices leads to explosive network traffic. The network traffic classification (NTC) is an essential tool to explore behaviours of network flows, and NTC is required for Internet service providers (ISPs) to manage the performance of the IoT network. We propose a novel network data representation, treating the traffic data as a series of images. Thus, the network data is realized as a video stream to employ time-distributed (TD) feature learning. The intra-temporal information within the network statistical data is learned using convolutional neural networks (CNN) and long short-term memory (LSTM), and the inter pseudo-temporal feature among the flows is learned by TD multi-layer perceptron (MLP). We conduct experiments using a large data-set with more number of classes. The experimental result shows that the TD feature learning elevates the network classification performance by 10\%.},
	urldate = {2021-10-07},
	journal = {arXiv:2109.14696 [cs]},
	author = {Manjunath, Yoga Suhas Kuruba and Zhao, Sihao and Zhang, Xiao-Ping},
	month = sep,
	year = {2021},
	note = {arXiv: 2109.14696},
	keywords = {Computer Science - Machine Learning, Computer Science - Networking and Internet Architecture},
}

The plethora of Internet of Things (IoT) devices leads to explosive network traffic. The network traffic classification (NTC) is an essential tool to explore behaviours of network flows, and NTC is required for Internet service providers (ISPs) to manage the performance of the IoT network. We propose a novel network data representation, treating the traffic data as a series of images. Thus, the network data is realized as a video stream to employ time-distributed (TD) feature learning. The intra-temporal information within the network statistical data is learned using convolutional neural networks (CNN) and long short-term memory (LSTM), and the inter pseudo-temporal feature among the flows is learned by TD multi-layer perceptron (MLP). We conduct experiments using a large data-set with more number of classes. The experimental result shows that the TD feature learning elevates the network classification performance by 10%.

@inproceedings{sandhya_madhuri_review_2021,
	address = {Singapore},
	series = {Lecture {Notes} in {Networks} and {Systems}},
	title = {Review {Paper} on {Anomaly} {Detection} in {Data} {Streams}},
	isbn = {9789811619410},
	doi = {10.1007/978-981-16-1941-0_72},
	abstract = {Anomaly is in general defined as deviation or diversion from the normal. The word anomaly came from the Greek word anomalia which means “uneven” or “irregular”. In our day-to-day lives, we have seen many such irregularities or deviations from normalcy. For example, a condition monitoring system beeps an alarm when it detects any value or parameter of the machine away from the minimum value limit to the maximum value limit, or a credit card fraud alerts the bank and the customer immediately. Now, the crucial task here is how we detect anomalies in data streams. When there is streaming data that is continuously generated from any source, it is called a data stream. The task of finding anomalies from such a stream of data will be a challenging job. In this paper, we will discuss elaborately about data streams and anomaly detection in data streams by reviewing several papers and articles written on this topic.},
	language = {en},
	booktitle = {Proceedings of the 2nd {International} {Conference} on {Computational} and {Bio} {Engineering}},
	publisher = {Springer},
	author = {Sandhya Madhuri, G. and {Yamuna} and Usha Rani, M.},
	editor = {Jyothi, S. and Mamatha, D. M. and Zhang, Yu-Dong and Raju, K. Srujan},
	year = {2021},
	keywords = {Anomaly, Anomaly detection algorithms, Data streams, Outliers},
	pages = {721--728},
}

Anomaly is in general defined as deviation or diversion from the normal. The word anomaly came from the Greek word anomalia which means “uneven” or “irregular”. In our day-to-day lives, we have seen many such irregularities or deviations from normalcy. For example, a condition monitoring system beeps an alarm when it detects any value or parameter of the machine away from the minimum value limit to the maximum value limit, or a credit card fraud alerts the bank and the customer immediately. Now, the crucial task here is how we detect anomalies in data streams. When there is streaming data that is continuously generated from any source, it is called a data stream. The task of finding anomalies from such a stream of data will be a challenging job. In this paper, we will discuss elaborately about data streams and anomaly detection in data streams by reviewing several papers and articles written on this topic.

@article{sejr_explainable_2021,
	title = {Explainable outlier detection: {What}, for {Whom} and {Why}?},
	issn = {2666-8270},
	shorttitle = {Explainable outlier detection},
	url = {https://www.sciencedirect.com/science/article/pii/S2666827021000864},
	doi = {10.1016/j.mlwa.2021.100172},
	abstract = {Outlier algorithms are becoming increasingly complex. Thereby, they become much less interpretable to the data scientists applying the algorithms in real-life settings and to end-users using their predictions. We argue that outliers are context-dependent and, therefore, can only be detected via domain knowledge, algorithm insight, and interaction with end-users. As outlier detection is equivalent to unsupervised semantic binary classification, at the core of interpreting an outlier algorithm we find the semantics of the classes, i.e., the algorithm’s conceptual outlier definition. We investigate current interpretable and explainable outlier algorithms: what they are, for whom they are, and what their value proposition is. We then discuss how interpretation and explanation and user involvement have the potential to provide the missing link to bring modern complex outlier algorithms from computer science labs into real-life applications and the challenges they induce.},
	language = {en},
	urldate = {2021-10-04},
	journal = {Machine Learning with Applications},
	author = {Sejr, Jonas Herskind and Schneider-Kamp, Anna},
	month = oct,
	year = {2021},
	keywords = {Explainable artificial intelligence, Unsupervised outlier detection},
	pages = {100172},
}

Outlier algorithms are becoming increasingly complex. Thereby, they become much less interpretable to the data scientists applying the algorithms in real-life settings and to end-users using their predictions. We argue that outliers are context-dependent and, therefore, can only be detected via domain knowledge, algorithm insight, and interaction with end-users. As outlier detection is equivalent to unsupervised semantic binary classification, at the core of interpreting an outlier algorithm we find the semantics of the classes, i.e., the algorithm’s conceptual outlier definition. We investigate current interpretable and explainable outlier algorithms: what they are, for whom they are, and what their value proposition is. We then discuss how interpretation and explanation and user involvement have the potential to provide the missing link to bring modern complex outlier algorithms from computer science labs into real-life applications and the challenges they induce.

@article{jia_condition_2021,
	title = {Condition monitoring and performance forecasting of wind turbines based on denoising autoencoder and novel convolutional neural networks},
	volume = {7},
	issn = {2352-4847},
	url = {https://www.sciencedirect.com/science/article/pii/S2352484721008854},
	doi = {10.1016/j.egyr.2021.09.080},
	abstract = {With the proportion of wind power in the grid increasing, the monitoring and maintenance of wind turbines is becoming more and more important for the reliability of the grid. In this study, a data-driven modelling framework based on deep convolutional neural networks is constructed for wind turbines condition monitoring (CM) and performance forecasting (PF). For CM, a robust denoising autoencoder (DAE) model is introduced to output the reconstruction error (RE) of raw signals. The RE is processed to a state indicator by exponentially weighted moving average (EWMA) and monitored on a control chart. For PF, two multi-steps ahead forecasting models are constructed for the forecasting of generator bearing and transformer temperature. To prevent overfitting caused by abundant features, the marginal effect analysis based on random forests is implemented to measure the importance of features. Besides, novel residual attention module (RAM) and training strategies are used improve their representation power of DAE and PF models. Experiments on a real wind turbine dataset prove the effectiveness of the proposed models and methods.},
	language = {en},
	urldate = {2021-10-04},
	journal = {Energy Reports},
	author = {Jia, Xiongjie and Han, Yang and Li, Yanjun and Sang, Yichen and Zhang, Guolei},
	month = nov,
	year = {2021},
	keywords = {Condition monitoring, Denoising autoencoder, Performance forecasting, Residual attention module, Wind turbine},
	pages = {6354--6365},
}

With the proportion of wind power in the grid increasing, the monitoring and maintenance of wind turbines is becoming more and more important for the reliability of the grid. In this study, a data-driven modelling framework based on deep convolutional neural networks is constructed for wind turbines condition monitoring (CM) and performance forecasting (PF). For CM, a robust denoising autoencoder (DAE) model is introduced to output the reconstruction error (RE) of raw signals. The RE is processed to a state indicator by exponentially weighted moving average (EWMA) and monitored on a control chart. For PF, two multi-steps ahead forecasting models are constructed for the forecasting of generator bearing and transformer temperature. To prevent overfitting caused by abundant features, the marginal effect analysis based on random forests is implemented to measure the importance of features. Besides, novel residual attention module (RAM) and training strategies are used improve their representation power of DAE and PF models. Experiments on a real wind turbine dataset prove the effectiveness of the proposed models and methods.

@article{hoque_data_2021,
	title = {Data driven analysis of lithium-ion battery internal resistance towards reliable state of health prediction},
	volume = {513},
	issn = {0378-7753},
	url = {https://www.sciencedirect.com/science/article/pii/S037877532101020X},
	doi = {10.1016/j.jpowsour.2021.230519},
	abstract = {Accurately predicting the lifetime of lithium-ion batteries in the early stage is critical for faster battery production, tuning the production line, and predictive maintenance of energy storage systems and battery-powered devices. Diverse usage patterns, variability in the devices housing the batteries, and diversity in their operating conditions pose significant challenges for this task. The contributions of this paper are three-fold. First, a public dataset is used to characterize the behavior of battery internal resistance. Internal resistance has non-linear dynamics as the battery ages, making it an excellent candidate for reliable battery health prediction during early cycles. Second, using these findings, battery health prediction models for different operating conditions are developed. The best models are more than 95\% accurate in predicting battery health using the internal resistance dynamics of 100 cycles at room temperature. Thirdly, instantaneous voltage drops due to multiple pulse discharge loads are shown to be capable of characterizing battery heterogeneity in as few as five cycles. The results pave the way toward improved battery models and better efficiency within the production and use of lithium-ion batteries.},
	language = {en},
	urldate = {2021-10-04},
	journal = {Journal of Power Sources},
	author = {Hoque, Mohammad A. and Nurmi, Petteri and Kumar, Arjun and Varjonen, Samu and Song, Junehwa and Pecht, Michael G. and Tarkoma, Sasu},
	month = nov,
	year = {2021},
	keywords = {Battery capacity, Internal resistance, Lithium-ion battery, State of health, health prediction, physical models},
	pages = {230519},
}

Accurately predicting the lifetime of lithium-ion batteries in the early stage is critical for faster battery production, tuning the production line, and predictive maintenance of energy storage systems and battery-powered devices. Diverse usage patterns, variability in the devices housing the batteries, and diversity in their operating conditions pose significant challenges for this task. The contributions of this paper are three-fold. First, a public dataset is used to characterize the behavior of battery internal resistance. Internal resistance has non-linear dynamics as the battery ages, making it an excellent candidate for reliable battery health prediction during early cycles. Second, using these findings, battery health prediction models for different operating conditions are developed. The best models are more than 95% accurate in predicting battery health using the internal resistance dynamics of 100 cycles at room temperature. Thirdly, instantaneous voltage drops due to multiple pulse discharge loads are shown to be capable of characterizing battery heterogeneity in as few as five cycles. The results pave the way toward improved battery models and better efficiency within the production and use of lithium-ion batteries.

@article{xiang_multicellular_2021,
	title = {Multicellular {LSTM}-based deep learning model for aero-engine remaining useful life prediction},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004439},
	doi = {10.1016/j.ress.2021.107927},
	abstract = {The prediction of aero-engine remaining useful life (RUL) is helpful for its operation and maintenance. Aiming at the challenge that most neural networks (NNs), including long short-term memory (LSTM), cannot process the input data in different update modes based on its importance degree, a novel variant of LSTM named multicellular LSTM (MCLSTM) is constructed. The level division unit is proposed to determine the importance degree of input data, and then multiple cellular units are designed to update the cell states according to the data level. Thus, MCLSTM can well mine different levels of degradation trends. Based on MCLSTM and a deep NN (DNN), a deep learning model for RUL prediction is set up, where MCLSTM and a branch of the DNN is used to extract health indicators (HIs) of aero-engine from raw data, and the other part of the DNN is applied to generate the HIs from human-made features and predict the RUL based on the concatenated HIs. The proposed RUL prediction model is successfully applied to predict the RULs of aero-engines via the Commercial Modular Aero Propulsion System Simulation datasets, and the comparative results show that it has a better comprehensive prediction performance than the commonly-used machine learning methods.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Xiang, Sheng and Qin, Yi and Luo, Jun and Pu, Huayan and Tang, Baoping},
	month = dec,
	year = {2021},
	keywords = {Data level, Degradation trend, Health feature, Multi-resource data, RUL prediction},
	pages = {107927},
}

The prediction of aero-engine remaining useful life (RUL) is helpful for its operation and maintenance. Aiming at the challenge that most neural networks (NNs), including long short-term memory (LSTM), cannot process the input data in different update modes based on its importance degree, a novel variant of LSTM named multicellular LSTM (MCLSTM) is constructed. The level division unit is proposed to determine the importance degree of input data, and then multiple cellular units are designed to update the cell states according to the data level. Thus, MCLSTM can well mine different levels of degradation trends. Based on MCLSTM and a deep NN (DNN), a deep learning model for RUL prediction is set up, where MCLSTM and a branch of the DNN is used to extract health indicators (HIs) of aero-engine from raw data, and the other part of the DNN is applied to generate the HIs from human-made features and predict the RUL based on the concatenated HIs. The proposed RUL prediction model is successfully applied to predict the RULs of aero-engines via the Commercial Modular Aero Propulsion System Simulation datasets, and the comparative results show that it has a better comprehensive prediction performance than the commonly-used machine learning methods.

@article{chen_two-phase_2021,
	title = {Two-phase degradation data analysis with change-point detection based on {Gaussian} process degradation model},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004324},
	doi = {10.1016/j.ress.2021.107916},
	abstract = {Degradation paths of the products exhibiting two-phase patterns are commonly seen in practice due to the changeable internal mechanisms and external environments. In this paper, we propose a two-phase Gaussian process (TPGP) degradation model with a change-point, which comprises the Wiener process-based change-point models as special cases, to describe the degradation paths with two-phase patterns. The change-point is used to represent the transition of degradation characteristics. The degradation rates and variations in the two phases are assumed to be different. Therefore, both monotonically increasing and decreasing or nonmonotonic dispersion trends and complicated auto-correlations in the degradation measurements can be captured by TPGP. Joint methods of the parameter estimation and change-point detection is developed for two different engineering scenarios. The distributions of the first passage time and the remaining useful life are derived in closed-form to promote the mathematical trackability and the applicability of the TPGP model. A comprehensive simulation study shows the effectiveness and validity of the proposed model and method. Finally, we use two real applications to demonstrate the proposed models and methods.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Chen, Zhen and Li, Yaping and Zhou, Di and Xia, Tangbin and Pan, Ershun},
	month = dec,
	year = {2021},
	keywords = {Change-point, First passage time, Gaussian process, Two-phase degradation},
	pages = {107916},
}

Degradation paths of the products exhibiting two-phase patterns are commonly seen in practice due to the changeable internal mechanisms and external environments. In this paper, we propose a two-phase Gaussian process (TPGP) degradation model with a change-point, which comprises the Wiener process-based change-point models as special cases, to describe the degradation paths with two-phase patterns. The change-point is used to represent the transition of degradation characteristics. The degradation rates and variations in the two phases are assumed to be different. Therefore, both monotonically increasing and decreasing or nonmonotonic dispersion trends and complicated auto-correlations in the degradation measurements can be captured by TPGP. Joint methods of the parameter estimation and change-point detection is developed for two different engineering scenarios. The distributions of the first passage time and the remaining useful life are derived in closed-form to promote the mathematical trackability and the applicability of the TPGP model. A comprehensive simulation study shows the effectiveness and validity of the proposed model and method. Finally, we use two real applications to demonstrate the proposed models and methods.

@article{braga_multivariate_2021,
	title = {Multivariate statistical aggregation and dimensionality reduction techniques to improve monitoring and maintenance in railways: {The} wheelset component},
	volume = {216},
	issn = {0951-8320},
	shorttitle = {Multivariate statistical aggregation and dimensionality reduction techniques to improve monitoring and maintenance in railways},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004488},
	doi = {10.1016/j.ress.2021.107932},
	abstract = {Reliable monitoring and assessment of wear evolutions are critical for performing effective railway maintenance. Several characteristics and variables are used to quantify a worn condition of railway wheelsets. To measure all these wear quantities, emerging inspection technologies are being designed with increasingly complex architectures, working mechanisms and associated high costs. Moreover, data-driven models to support condition-based maintenance to the wheelset easily increase their complexity when too many variables are taken into account and may not provide a straightforward guideline to maintenance decision-makers. The purpose of this paper is to reduce the complexity when describing the wear level, by applying multivariate statistical techniques to real degradation data from railway wheelsets. Several wheelset condition variables and their relationships are analysed. Variables are synthetized through a principal component analysis (PCA) where the varimax rotation effect can be observed. A cluster analysis, which uses the principal components, allows identifying characteristics that lead to different wear evolutions. A strong correlation between the flange thickness and flange slope in the wear process is identified. Differences in wear trajectories between motor and trailer wheelsets are strongly significant. The findings are expected to support the improvement of state monitoring techniques and predictive maintenance optimization models.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Braga, Joaquim A. P. and Andrade, António R.},
	month = dec,
	year = {2021},
	keywords = {Cluster analysis, Condition monitoring, Principal component analysis, Railway maintenance, Wheelset inspection, Wheelset wear},
	pages = {107932},
}

Reliable monitoring and assessment of wear evolutions are critical for performing effective railway maintenance. Several characteristics and variables are used to quantify a worn condition of railway wheelsets. To measure all these wear quantities, emerging inspection technologies are being designed with increasingly complex architectures, working mechanisms and associated high costs. Moreover, data-driven models to support condition-based maintenance to the wheelset easily increase their complexity when too many variables are taken into account and may not provide a straightforward guideline to maintenance decision-makers. The purpose of this paper is to reduce the complexity when describing the wear level, by applying multivariate statistical techniques to real degradation data from railway wheelsets. Several wheelset condition variables and their relationships are analysed. Variables are synthetized through a principal component analysis (PCA) where the varimax rotation effect can be observed. A cluster analysis, which uses the principal components, allows identifying characteristics that lead to different wear evolutions. A strong correlation between the flange thickness and flange slope in the wear process is identified. Differences in wear trajectories between motor and trailer wheelsets are strongly significant. The findings are expected to support the improvement of state monitoring techniques and predictive maintenance optimization models.

@article{sun_hierarchical_2021,
	title = {A hierarchical modeling approach for degradation data with mixed-type covariates and latent heterogeneity},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004440},
	doi = {10.1016/j.ress.2021.107928},
	abstract = {Successful modeling of degradation data with covariates is essential for accurate reliability assessment of highly reliable product units. Due to the influences of different types of covariates, such as the external factors (e.g. accelerated operating conditions) and the internal factors (e.g. material microstructure characteristics), as well as latent heterogeneity due to the influences of the unobserved or unknown factors shared within each product unit, the degradation measurements of product units are highly heterogeneous over time. Many of existing degradation models often failed to simultaneously consider the influences of (i) both external accelerated conditions and internal material information, (ii) latent heterogeneity, and (iii) multiple material types. In this work, we propose a generic degradation modeling approach with mixed-type (e.g. both scalar and functional) covariates and latent heterogeneity to account for both the influences of observed internal and external factors as well as their interaction, and the influences of unobserved factors. Effective estimation algorithm is developed under expectation–maximization framework to jointly quantify the influences of mixed-type covariates and individual latent heterogeneity. The proposed algorithms further enables closed-form updating of model parameters at each iteration to ensure the estimation convenience. A real case study is provided to illustrate the proposed modeling approach and to demonstrate its effectiveness from both model prediction and interpretation perspectives.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Sun, Xuxue and Cai, Wenjun and Li, Mingyang},
	month = dec,
	year = {2021},
	keywords = {Data augmentation, Degradation data, Functional data analysis, Latent heterogeneity, Mixed-type covariates},
	pages = {107928},
}

Successful modeling of degradation data with covariates is essential for accurate reliability assessment of highly reliable product units. Due to the influences of different types of covariates, such as the external factors (e.g. accelerated operating conditions) and the internal factors (e.g. material microstructure characteristics), as well as latent heterogeneity due to the influences of the unobserved or unknown factors shared within each product unit, the degradation measurements of product units are highly heterogeneous over time. Many of existing degradation models often failed to simultaneously consider the influences of (i) both external accelerated conditions and internal material information, (ii) latent heterogeneity, and (iii) multiple material types. In this work, we propose a generic degradation modeling approach with mixed-type (e.g. both scalar and functional) covariates and latent heterogeneity to account for both the influences of observed internal and external factors as well as their interaction, and the influences of unobserved factors. Effective estimation algorithm is developed under expectation–maximization framework to jointly quantify the influences of mixed-type covariates and individual latent heterogeneity. The proposed algorithms further enables closed-form updating of model parameters at each iteration to ensure the estimation convenience. A real case study is provided to illustrate the proposed modeling approach and to demonstrate its effectiveness from both model prediction and interpretation perspectives.

@article{fecarotti_mathematical_2021,
	title = {A mathematical programming model to select maintenance strategies in railway networks},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004518},
	doi = {10.1016/j.ress.2021.107940},
	abstract = {This paper presents a nonlinear integer programming model to support the selection of maintenance strategies to implement on different segments of a railway network. Strategies are selected which collectively minimise the impact of sections’ conditions on service, given network availability and budget constraints. Different metrics related to the network topology, sections’ availability, service frequency, performance requirements and maintenance costs, are combined into a quantitative approach with a holistic view. The main contribution is to provide a simple yet effective modelling approach and solution method which are suitable for large networks and make use of standard solvers. Both an ad hoc heuristic solution and relaxation methods are developed, the latter enabling the quality of the heuristic solution to be estimated. The availability of railway lines is computed by exploiting the analogy with series–parallel networks. By varying the model parameters, a scenario analysis is performed to give insight into the influence of the system parameters on the selection of strategies, thus enabling more informed decisions. For its simple structure, the model is versatile to address similar problems arising in the maintenance of other types of networks, such as road and bridges networks, when deciding on the strategic allocation of maintenance efforts.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Fecarotti, Claudia and Andrews, John and Pesenti, Raffaele},
	month = dec,
	year = {2021},
	keywords = {Availability, Maintenance optimisation, Mathematical programming, Railway networks},
	pages = {107940},
}

This paper presents a nonlinear integer programming model to support the selection of maintenance strategies to implement on different segments of a railway network. Strategies are selected which collectively minimise the impact of sections’ conditions on service, given network availability and budget constraints. Different metrics related to the network topology, sections’ availability, service frequency, performance requirements and maintenance costs, are combined into a quantitative approach with a holistic view. The main contribution is to provide a simple yet effective modelling approach and solution method which are suitable for large networks and make use of standard solvers. Both an ad hoc heuristic solution and relaxation methods are developed, the latter enabling the quality of the heuristic solution to be estimated. The availability of railway lines is computed by exploiting the analogy with series–parallel networks. By varying the model parameters, a scenario analysis is performed to give insight into the influence of the system parameters on the selection of strategies, thus enabling more informed decisions. For its simple structure, the model is versatile to address similar problems arising in the maintenance of other types of networks, such as road and bridges networks, when deciding on the strategic allocation of maintenance efforts.

@article{zhuang_temporal_2021,
	title = {Temporal convolution-based transferable cross-domain adaptation approach for remaining useful life estimation under variable failure behaviors},
	volume = {216},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021004592},
	doi = {10.1016/j.ress.2021.107946},
	abstract = {Many data-driven models normally assume that the training and test data are independent and identically distributed to predict the remaining useful life (RUL) of industrial machines. However, different failure models caused by variable failure behaviors may lead to a domain shift. Meanwhile, conventional methods lack comprehensive attention to temporal information, resulting in a limitation. To handle the aforementioned challenges, a transferable cross-domain approach for RUL estimation is proposed. The hidden features are extracted adaptively by a temporal convolution network in which residual self-attention is able to fully consider the contextual degradation information. Furthermore, a new cross-domain adaption architecture with the contrastive loss and multi-kernel maximum mean discrepancy is designed to learn the domain invariant features. The effectiveness and superiority of the proposed method are proved by the case study on IEEE PHM challenge 2012 bearing dataset and the comparison with other methods.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhuang, Jichao and Jia, Minping and Ding, Yifei and Ding, Peng},
	month = dec,
	year = {2021},
	keywords = {Cross-domain adaptation, Remaining useful life estimation, Rolling bearing, Temporal convolutional network, Transfer learning, Variable failure behaviors},
	pages = {107946},
}

Many data-driven models normally assume that the training and test data are independent and identically distributed to predict the remaining useful life (RUL) of industrial machines. However, different failure models caused by variable failure behaviors may lead to a domain shift. Meanwhile, conventional methods lack comprehensive attention to temporal information, resulting in a limitation. To handle the aforementioned challenges, a transferable cross-domain approach for RUL estimation is proposed. The hidden features are extracted adaptively by a temporal convolution network in which residual self-attention is able to fully consider the contextual degradation information. Furthermore, a new cross-domain adaption architecture with the contrastive loss and multi-kernel maximum mean discrepancy is designed to learn the domain invariant features. The effectiveness and superiority of the proposed method are proved by the case study on IEEE PHM challenge 2012 bearing dataset and the comparison with other methods.

@article{duraj_outlier_2021,
	series = {Knowledge-{Based} and {Intelligent} {Information} \& {Engineering} {Systems}: {Proceedings} of the 25th {International} {Conference} {KES2021}},
	title = {Outlier {Detection} in {Data} {Streams} — {A} {Comparative} {Study} of {Selected} {Methods}},
	volume = {192},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050921017841},
	doi = {10.1016/j.procs.2021.09.047},
	abstract = {Outlier detection is an increasingly important and intensively developing area of research. This paper focuses on the problem of outlier detection in data streams. It presents a performance comparison of selected statistical algorithms: AutoRegressive Integrated Moving Average (ARIMA), Exponential Smoothing State Space Model (ETS), Seasonal Hybrid Extreme Studentized Deviation (SHESD), Non-parametric methodology (NMV), and Chen-Liu method (CHL). Based on four data streams from the Kaggle Repository and DataHub Repository, the study provides results concerning the number of outliers detected by each algorithm and the algorithms’ operation times. The experiments were performed on data streams of different lengths (from a few hundred to 1200 records), characterized by the presence of different types of outliers.},
	language = {en},
	urldate = {2021-10-02},
	journal = {Procedia Computer Science},
	author = {Duraj, Agnieszka and Szczepaniak, Piotr S.},
	month = jan,
	year = {2021},
	keywords = {comparative study, outlier detection, stream data analysis},
	pages = {2769--2778},
}

Outlier detection is an increasingly important and intensively developing area of research. This paper focuses on the problem of outlier detection in data streams. It presents a performance comparison of selected statistical algorithms: AutoRegressive Integrated Moving Average (ARIMA), Exponential Smoothing State Space Model (ETS), Seasonal Hybrid Extreme Studentized Deviation (SHESD), Non-parametric methodology (NMV), and Chen-Liu method (CHL). Based on four data streams from the Kaggle Repository and DataHub Repository, the study provides results concerning the number of outliers detected by each algorithm and the algorithms’ operation times. The experiments were performed on data streams of different lengths (from a few hundred to 1200 records), characterized by the presence of different types of outliers.

@article{dubey_comprehensive_2021,
	title = {A {Comprehensive} {Survey} and {Performance} {Analysis} of {Activation} {Functions} in {Deep} {Learning}},
	url = {http://arxiv.org/abs/2109.14545},
	abstract = {Neural networks have shown tremendous growth in recent years to solve numerous problems. Various types of neural networks have been introduced to deal with different types of problems. However, the main goal of any neural network is to transform the non-linearly separable input data into more linearly separable abstract features using a hierarchy of layers. These layers are combinations of linear and nonlinear functions. The most popular and common non-linearity layers are activation functions (AFs), such as Logistic Sigmoid, Tanh, ReLU, ELU, Swish and Mish. In this paper, a comprehensive overview and survey is presented for AFs in neural networks for deep learning. Different classes of AFs such as Logistic Sigmoid and Tanh based, ReLU based, ELU based, and Learning based are covered. Several characteristics of AFs such as output range, monotonicity, and smoothness are also pointed out. A performance comparison is also performed among 18 state-of-the-art AFs with different networks on different types of data. The insights of AFs are presented to benefit the researchers for doing further research and practitioners to select among different choices. The code used for experimental comparison is released at: {\textbackslash}url\{https://github.com/shivram1987/ActivationFunctions\}.},
	urldate = {2021-10-01},
	journal = {arXiv:2109.14545 [cs]},
	author = {Dubey, Shiv Ram and Singh, Satish Kumar and Chaudhuri, Bidyut Baran},
	month = sep,
	year = {2021},
	note = {arXiv: 2109.14545},
	keywords = {Computer Science - Machine Learning, Computer Science - Neural and Evolutionary Computing},
}

Neural networks have shown tremendous growth in recent years to solve numerous problems. Various types of neural networks have been introduced to deal with different types of problems. However, the main goal of any neural network is to transform the non-linearly separable input data into more linearly separable abstract features using a hierarchy of layers. These layers are combinations of linear and nonlinear functions. The most popular and common non-linearity layers are activation functions (AFs), such as Logistic Sigmoid, Tanh, ReLU, ELU, Swish and Mish. In this paper, a comprehensive overview and survey is presented for AFs in neural networks for deep learning. Different classes of AFs such as Logistic Sigmoid and Tanh based, ReLU based, ELU based, and Learning based are covered. Several characteristics of AFs such as output range, monotonicity, and smoothness are also pointed out. A performance comparison is also performed among 18 state-of-the-art AFs with different networks on different types of data. The insights of AFs are presented to benefit the researchers for doing further research and practitioners to select among different choices. The code used for experimental comparison is released at: \url\https://github.com/shivram1987/ActivationFunctions\.

@inproceedings{mallouk_machine_2021,
	title = {Machine learning approach for predictive maintenance of transport systems},
	doi = {10.1109/TST52996.2021.00023},
	abstract = {Transportation companies must face to a huge competition and must reduce downtime and the associated costs. This can be achieved through predictive maintenance (PM), which defines maintenance actions based on the health of the system and its environment. Relevant information can be extracted from massive data related to health prognosis and management (PHM) by applying artificial intelligence (AI) techniques. This paper proposes a Machine Learning approach to develop a prediction model based on a supervised learning by comparing several regression algorithms. The model is then applied to the Remaining useful mileage prediction of trucks tires for a transport application of dangerous substances.},
	booktitle = {2021 {Third} {International} {Conference} on {Transportation} and {Smart} {Technologies} ({TST})},
	author = {Mallouk, Issam and Sallez, Yves and El Majd, Badr Abou},
	month = may,
	year = {2021},
	keywords = {Machine Learning, Machine learning, Prediction algorithms, Prediction model, Predictive models, Stakeholders, Supervised learning, Tires, Transportation, Useful life, regression, transportation},
	pages = {96--100},
}

Transportation companies must face to a huge competition and must reduce downtime and the associated costs. This can be achieved through predictive maintenance (PM), which defines maintenance actions based on the health of the system and its environment. Relevant information can be extracted from massive data related to health prognosis and management (PHM) by applying artificial intelligence (AI) techniques. This paper proposes a Machine Learning approach to develop a prediction model based on a supervised learning by comparing several regression algorithms. The model is then applied to the Remaining useful mileage prediction of trucks tires for a transport application of dangerous substances.

@article{de_pater_predictive_2021,
	title = {Predictive maintenance for multi-component systems of repairables with {Remaining}-{Useful}-{Life} prognostics and a limited stock of spare components},
	volume = {214},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S095183202100288X},
	doi = {10.1016/j.ress.2021.107761},
	abstract = {Aircraft maintenance is undergoing a paradigm shift towards predictive maintenance, where the use of sensor data and Remaining-Useful-Life prognostics are central. This paper proposes an integrated approach for predictive aircraft maintenance planning for multiple multi-component systems, where the components are repairables. First, model-based Remaining-Useful-Life prognostics are developed. These prognostics are updated over time, as more sensor data become available. Then, a rolling horizon integer linear program is developed for the maintenance planning of multiple multi-component systems. This model integrates the Remaining-Useful-Life prognostics with the management of a limited stock of spare repairable components. The maintenance of the multiple systems is linked through the availability of spare components and shared maintenance time slots. Our approach is illustrated for a fleet of aircraft, each equipped with a Cooling System consisting of four Cooling Units. For an aircraft to be operational, a minimum of two Cooling Units out of the four need to be operational. The maintenance planning results show that our integrated approach reduces the costs with maintenance by 48\% relative to a corrective maintenance strategy and by 30\% relative to a preventive maintenance strategy. Moreover, using predictive maintenance, components are replaced in anticipation of failure without wasting their useful life. In general, our approach provides a roadmap from Remaining-Useful-Life prognostics to maintenance planning for multiple multi-component systems of repairables with a limited stock of spares.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Reliability Engineering \& System Safety},
	author = {de Pater, Ingeborg and Mitici, Mihaela},
	month = oct,
	year = {2021},
	keywords = {Aircraft Cooling Units, Aircraft predictive maintenance of repairables, Management of spare components, Multiple multi-component systems, RUL prognostics},
	pages = {107761},
}

Aircraft maintenance is undergoing a paradigm shift towards predictive maintenance, where the use of sensor data and Remaining-Useful-Life prognostics are central. This paper proposes an integrated approach for predictive aircraft maintenance planning for multiple multi-component systems, where the components are repairables. First, model-based Remaining-Useful-Life prognostics are developed. These prognostics are updated over time, as more sensor data become available. Then, a rolling horizon integer linear program is developed for the maintenance planning of multiple multi-component systems. This model integrates the Remaining-Useful-Life prognostics with the management of a limited stock of spare repairable components. The maintenance of the multiple systems is linked through the availability of spare components and shared maintenance time slots. Our approach is illustrated for a fleet of aircraft, each equipped with a Cooling System consisting of four Cooling Units. For an aircraft to be operational, a minimum of two Cooling Units out of the four need to be operational. The maintenance planning results show that our integrated approach reduces the costs with maintenance by 48% relative to a corrective maintenance strategy and by 30% relative to a preventive maintenance strategy. Moreover, using predictive maintenance, components are replaced in anticipation of failure without wasting their useful life. In general, our approach provides a roadmap from Remaining-Useful-Life prognostics to maintenance planning for multiple multi-component systems of repairables with a limited stock of spares.

@article{thuerey_physics-based_2021,
	title = {Physics-based {Deep} {Learning}},
	url = {http://arxiv.org/abs/2109.05237},
	abstract = {This digital book contains a practical and comprehensive introduction of everything related to deep learning in the context of physical simulations. As much as possible, all topics come with hands-on code examples in the form of Jupyter notebooks to quickly get started. Beyond standard supervised learning from data, we'll look at physical loss constraints, more tightly coupled learning algorithms with differentiable simulations, as well as reinforcement learning and uncertainty modeling. We live in exciting times: these methods have a huge potential to fundamentally change what computer simulations can achieve.},
	urldate = {2021-09-15},
	journal = {arXiv:2109.05237 [physics]},
	author = {Thuerey, Nils and Holl, Philipp and Mueller, Maximilian and Schnell, Patrick and Trost, Felix and Um, Kiwon},
	month = sep,
	year = {2021},
	note = {arXiv: 2109.05237},
	keywords = {Computer Science - Machine Learning, Physics - Computational Physics},
}

This digital book contains a practical and comprehensive introduction of everything related to deep learning in the context of physical simulations. As much as possible, all topics come with hands-on code examples in the form of Jupyter notebooks to quickly get started. Beyond standard supervised learning from data, we'll look at physical loss constraints, more tightly coupled learning algorithms with differentiable simulations, as well as reinforcement learning and uncertainty modeling. We live in exciting times: these methods have a huge potential to fundamentally change what computer simulations can achieve.

@article{chen_combining_2021,
	title = {Combining empirical mode decomposition and deep recurrent neural networks for predictive maintenance of lithium-ion battery},
	volume = {50},
	issn = {1474-0346},
	url = {https://www.sciencedirect.com/science/article/pii/S1474034621001579},
	doi = {10.1016/j.aei.2021.101405},
	abstract = {Predictive maintenance of lithium-ion batteries has been one of the popular research subjects in recent years. Lithium-ion batteries can be used as the energy supply for industrial equipment, such as automated guided vehicles and battery electric vehicles. Predictive maintenance plays an important role in the application of smart manufacturing. This mechanism can provide different levels of pre-diagnosis for machines or components. Remaining useful life (RUL) prediction is crucial for the implementation of predictive maintenance strategies. RUL refers to the estimated useful life remaining before the machine cannot operate after a certain period of operation. This study develops a hybrid data science model based on empirical mode decomposition (EMD), grey relational analysis (GRA), and deep recurrent neural networks (RNN) for the RUL prediction of lithium-ion batteries. The EMD and GRA methods are first adopted to extract the characteristics of time series data. Then, various deep RNNs, including vanilla RNN, gated recurrent unit, long short-term memory network (LSTM), and bidirectional LSTM, are established to forecast state of health (SOH) and the RUL of lithium-ion batteries. Bayesian optimization is also used to find the best hyperparameters of deep RNNs. Experimental results with the lithium-ion batteries data of NASA Ames Prognostics Data Repository show that the proposed hybrid data science model can accurately predict the SOH and RUL of lithium-ion batteries. The LSTM network has the optimal results. The proposed hybrid data science model with multiple artificial intelligence-based technologies also demonstrates critical digital-technology enablers for digital transformation of smart manufacturing and transportation.},
	language = {en},
	urldate = {2021-09-13},
	journal = {Advanced Engineering Informatics},
	author = {Chen, James C. and Chen, Tzu-Li and Liu, Wei-Jun and Cheng, C. C. and Li, Meng-Gung},
	month = oct,
	year = {2021},
	keywords = {Bayesian optimization, Deep recurrent neural network, Empirical mode decomposition, Lithium-Ion battery, Predictive maintenance, Remaining useful life},
	pages = {101405},
}

Predictive maintenance of lithium-ion batteries has been one of the popular research subjects in recent years. Lithium-ion batteries can be used as the energy supply for industrial equipment, such as automated guided vehicles and battery electric vehicles. Predictive maintenance plays an important role in the application of smart manufacturing. This mechanism can provide different levels of pre-diagnosis for machines or components. Remaining useful life (RUL) prediction is crucial for the implementation of predictive maintenance strategies. RUL refers to the estimated useful life remaining before the machine cannot operate after a certain period of operation. This study develops a hybrid data science model based on empirical mode decomposition (EMD), grey relational analysis (GRA), and deep recurrent neural networks (RNN) for the RUL prediction of lithium-ion batteries. The EMD and GRA methods are first adopted to extract the characteristics of time series data. Then, various deep RNNs, including vanilla RNN, gated recurrent unit, long short-term memory network (LSTM), and bidirectional LSTM, are established to forecast state of health (SOH) and the RUL of lithium-ion batteries. Bayesian optimization is also used to find the best hyperparameters of deep RNNs. Experimental results with the lithium-ion batteries data of NASA Ames Prognostics Data Repository show that the proposed hybrid data science model can accurately predict the SOH and RUL of lithium-ion batteries. The LSTM network has the optimal results. The proposed hybrid data science model with multiple artificial intelligence-based technologies also demonstrates critical digital-technology enablers for digital transformation of smart manufacturing and transportation.

@article{muller_online_2021,
	series = {16th {IFAC} {Symposium} on {Advanced} {Control} of {Chemical} {Processes} {ADCHEM} 2021},
	title = {Online wear detection for joints in progressing cavity pumps},
	volume = {54},
	issn = {2405-8963},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896321010119},
	doi = {10.1016/j.ifacol.2021.08.240},
	abstract = {The wear and failure of pin joints integrated in coupling rods of process equipment can lead to unpredictable breakdown with severe consequences. We present a simple algorithm for a non-invasive online detection of wear in pin joints where a progressing cavity pump serves as an example. The algorithm only requires a pressure sensor and a binary speed signal of the motor. We verify the algorithm with a laboratory test setup and an embedded system. The results show that the proposed algorithm can reliably monitor joint wear during regular pump operation.},
	language = {en},
	number = {3},
	urldate = {2021-09-13},
	journal = {IFAC-PapersOnLine},
	author = {Müller, J. and Leonow, S. and Schulz, J. and Hansen, C. and Mönnigmann, M.},
	month = jan,
	year = {2021},
	keywords = {fault detection, joint wear, phase-locked loop, positive displacement pumps, predictive maintenance, progressing cavity pump},
	pages = {188--193},
}

The wear and failure of pin joints integrated in coupling rods of process equipment can lead to unpredictable breakdown with severe consequences. We present a simple algorithm for a non-invasive online detection of wear in pin joints where a progressing cavity pump serves as an example. The algorithm only requires a pressure sensor and a binary speed signal of the motor. We verify the algorithm with a laboratory test setup and an embedded system. The results show that the proposed algorithm can reliably monitor joint wear during regular pump operation.

@article{budd_survey_2021,
	title = {A survey on active learning and human-in-the-loop deep learning for medical image analysis},
	volume = {71},
	issn = {1361-8415},
	url = {https://www.sciencedirect.com/science/article/pii/S1361841521001080},
	doi = {10.1016/j.media.2021.102062},
	abstract = {Fully automatic deep learning has become the state-of-the-art technique for many tasks including image acquisition, analysis and interpretation, and for the extraction of clinically useful information for computer-aided detection, diagnosis, treatment planning, intervention and therapy. However, the unique challenges posed by medical image analysis suggest that retaining a human end-user in any deep learning enabled system will be beneficial. In this review we investigate the role that humans might play in the development and deployment of deep learning enabled diagnostic applications and focus on techniques that will retain a significant input from a human end user. Human-in-the-Loop computing is an area that we see as increasingly important in future research due to the safety-critical nature of working in the medical domain. We evaluate four key areas that we consider vital for deep learning in the clinical practice: (1) Active Learning to choose the best data to annotate for optimal model performance; (2) Interaction with model outputs - using iterative feedback to steer models to optima for a given prediction and offering meaningful ways to interpret and respond to predictions; (3) Practical considerations - developing full scale applications and the key considerations that need to be made before deployment; (4) Future Prospective and Unanswered Questions - knowledge gaps and related research fields that will benefit human-in-the-loop computing as they evolve. We offer our opinions on the most promising directions of research and how various aspects of each area might be unified towards common goals.},
	language = {en},
	urldate = {2021-09-13},
	journal = {Medical Image Analysis},
	author = {Budd, Samuel and Robinson, Emma C. and Kainz, Bernhard},
	month = jul,
	year = {2021},
	keywords = {Active learning, Deep Learning, Human-in-the-Loop, Medical image analysis},
	pages = {102062},
}

Fully automatic deep learning has become the state-of-the-art technique for many tasks including image acquisition, analysis and interpretation, and for the extraction of clinically useful information for computer-aided detection, diagnosis, treatment planning, intervention and therapy. However, the unique challenges posed by medical image analysis suggest that retaining a human end-user in any deep learning enabled system will be beneficial. In this review we investigate the role that humans might play in the development and deployment of deep learning enabled diagnostic applications and focus on techniques that will retain a significant input from a human end user. Human-in-the-Loop computing is an area that we see as increasingly important in future research due to the safety-critical nature of working in the medical domain. We evaluate four key areas that we consider vital for deep learning in the clinical practice: (1) Active Learning to choose the best data to annotate for optimal model performance; (2) Interaction with model outputs - using iterative feedback to steer models to optima for a given prediction and offering meaningful ways to interpret and respond to predictions; (3) Practical considerations - developing full scale applications and the key considerations that need to be made before deployment; (4) Future Prospective and Unanswered Questions - knowledge gaps and related research fields that will benefit human-in-the-loop computing as they evolve. We offer our opinions on the most promising directions of research and how various aspects of each area might be unified towards common goals.

@article{he_automl_2021,
	title = {{AutoML}: {A} {Survey} of the {State}-of-the-{Art}},
	volume = {212},
	issn = {09507051},
	shorttitle = {{AutoML}},
	url = {http://arxiv.org/abs/1908.00709},
	doi = {10.1016/j.knosys.2020.106622},
	abstract = {Deep learning (DL) techniques have penetrated all aspects of our lives and brought us great convenience. However, building a high-quality DL system for a specific task highly relies on human expertise, hindering the applications of DL to more areas. Automated machine learning (AutoML) becomes a promising solution to build a DL system without human assistance, and a growing number of researchers focus on AutoML. In this paper, we provide a comprehensive and up-to-date review of the state-of-the-art (SOTA) in AutoML. First, we introduce AutoML methods according to the pipeline, covering data preparation, feature engineering, hyperparameter optimization, and neural architecture search (NAS). We focus more on NAS, as it is currently very hot sub-topic of AutoML. We summarize the performance of the representative NAS algorithms on the CIFAR-10 and ImageNet datasets and further discuss several worthy studying directions of NAS methods: one/two-stage NAS, one-shot NAS, and joint hyperparameter and architecture optimization. Finally, we discuss some open problems of the existing AutoML methods for future research.},
	urldate = {2021-08-19},
	journal = {Knowledge-Based Systems},
	author = {He, Xin and Zhao, Kaiyong and Chu, Xiaowen},
	month = jan,
	year = {2021},
	note = {arXiv: 1908.00709},
	keywords = {Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning, Statistics - Machine Learning},
	pages = {106622},
}

Deep learning (DL) techniques have penetrated all aspects of our lives and brought us great convenience. However, building a high-quality DL system for a specific task highly relies on human expertise, hindering the applications of DL to more areas. Automated machine learning (AutoML) becomes a promising solution to build a DL system without human assistance, and a growing number of researchers focus on AutoML. In this paper, we provide a comprehensive and up-to-date review of the state-of-the-art (SOTA) in AutoML. First, we introduce AutoML methods according to the pipeline, covering data preparation, feature engineering, hyperparameter optimization, and neural architecture search (NAS). We focus more on NAS, as it is currently very hot sub-topic of AutoML. We summarize the performance of the representative NAS algorithms on the CIFAR-10 and ImageNet datasets and further discuss several worthy studying directions of NAS methods: one/two-stage NAS, one-shot NAS, and joint hyperparameter and architecture optimization. Finally, we discuss some open problems of the existing AutoML methods for future research.

@article{alfeo_degradation_2021,
	title = {Degradation stage classification via interpretable feature learning},
	issn = {0278-6125},
	url = {https://www.sciencedirect.com/science/article/pii/S027861252100100X},
	doi = {10.1016/j.jmsy.2021.05.003},
	abstract = {Predictive maintenance (PdM) advocates for the usage of machine learning technologies to monitor asset's health conditions and plan maintenance activities accordingly. However, according to the specific degradation process, some health-related measures (e.g. temperature) may be not informative enough to reliably assess the health stage. Moreover, each measure needs to be properly treated to extract the information linked to the health stage. Those issues are usually addressed by performing a manual feature engineering, which results in high management cost and poor generalization capability of those approaches. In this work, we address this issue by coupling a health stage classifier with a feature learning mechanism. With feature learning, minimally processed data are automatically transformed into informative features. Many effective feature learning approaches are based on deep learning. With those, the features are obtained as a non-linear combination of the inputs, thus it is difficult to understand the input's contribution to the classification outcome and so the reasoning behind the model. Still, these insights are increasingly required to interpret the results and assess the reliability of the model. In this regard, we propose a feature learning approach able to (i) effectively extract high-quality features by processing different input signals, and (ii) provide useful insights about the most informative domain transformations (e.g. Fourier transform or probability density function) of the input signals (e.g. vibration or temperature). The effectiveness of the proposed approach is tested with publicly available real-world datasets about bearings' progressive deterioration and compared with the traditional feature engineering approach.},
	language = {en},
	urldate = {2021-05-22},
	journal = {Journal of Manufacturing Systems},
	author = {Alfeo, Antonio L. and Cimino, Mario G. C. A. and Vaglini, Gigliola},
	month = may,
	year = {2021},
	keywords = {Autoencoder, Deep learning, Explainable artificial intelligence, Feature learning, Interpretable machine learning, Predictive maintenance},
}

Predictive maintenance (PdM) advocates for the usage of machine learning technologies to monitor asset's health conditions and plan maintenance activities accordingly. However, according to the specific degradation process, some health-related measures (e.g. temperature) may be not informative enough to reliably assess the health stage. Moreover, each measure needs to be properly treated to extract the information linked to the health stage. Those issues are usually addressed by performing a manual feature engineering, which results in high management cost and poor generalization capability of those approaches. In this work, we address this issue by coupling a health stage classifier with a feature learning mechanism. With feature learning, minimally processed data are automatically transformed into informative features. Many effective feature learning approaches are based on deep learning. With those, the features are obtained as a non-linear combination of the inputs, thus it is difficult to understand the input's contribution to the classification outcome and so the reasoning behind the model. Still, these insights are increasingly required to interpret the results and assess the reliability of the model. In this regard, we propose a feature learning approach able to (i) effectively extract high-quality features by processing different input signals, and (ii) provide useful insights about the most informative domain transformations (e.g. Fourier transform or probability density function) of the input signals (e.g. vibration or temperature). The effectiveness of the proposed approach is tested with publicly available real-world datasets about bearings' progressive deterioration and compared with the traditional feature engineering approach.

@article{guo_study_2021,
	title = {Study on {Landscape} {Architecture} {Model} {Design} {Based} on {Big} {Data} {Intelligence}},
	issn = {2214-5796},
	url = {https://www.sciencedirect.com/science/article/pii/S2214579621000368},
	doi = {10.1016/j.bdr.2021.100219},
	abstract = {Because of the rapid development of Internet technology in recent years, the speed of information data growth is faster and faster, through the use of Internet information data to landscape architecture analysis and research, has become the main direction of industry development. Landscape planners are involved in a wide range of projects, and architectural objects are often very complex in layout. Projects are usually composed of multiple components. For the point clouds corresponding to these objects, the direct reconstruction of them is relatively complex, which requires macroscopic and reasonable planning on the regional scale. Generally, plots should be designed in three-dimensional space. The contents presented by planning and design of different scales are also different. In dealing with multi-level regional planning and design, this paper mainly focuses on the parametric design technology research of landscape architecture. First, it starts from the design of large areas, analyzes the status quo of regional scale, and then starts to design with the help of three-dimensional model. First will get the clusters of 3 d model the coarse segmentation class continue to split into more rules part, according to the characteristics of the model data, the first of 3 d model to differential information to estimate the point cloud data, calculate the normal vector and curvature of point cloud data, through a point cloud registration technology will point cloud unified under different Angle of view to the same coordinate system, through the big data landscape algorithm for geometric feature and image feature point detection, detailed point cloud data processing algorithms in the process of coarse segmentation of regional scale in the same cluster of different object segmentation, again through the panoramic image segmentation for interior point cloud of geometric structure information. The data can be classified intelligently, the corresponding point cloud data can be marked, and the geometric structure information can be marked into the point cloud data to be segmented through point cloud matching, so as to enhance knowledge reserves, find problems and solve problems timely.},
	language = {en},
	urldate = {2021-02-28},
	journal = {Big Data Research},
	author = {Guo, Shiyun and Tang, Jinping and Liu, Huabin and Gu, Xinren},
	month = feb,
	year = {2021},
	keywords = {Big data, Cluster analysis, Geographic information systems, Landscape architecture, Parameter model construction, Programming},
	pages = {100219},
}

Because of the rapid development of Internet technology in recent years, the speed of information data growth is faster and faster, through the use of Internet information data to landscape architecture analysis and research, has become the main direction of industry development. Landscape planners are involved in a wide range of projects, and architectural objects are often very complex in layout. Projects are usually composed of multiple components. For the point clouds corresponding to these objects, the direct reconstruction of them is relatively complex, which requires macroscopic and reasonable planning on the regional scale. Generally, plots should be designed in three-dimensional space. The contents presented by planning and design of different scales are also different. In dealing with multi-level regional planning and design, this paper mainly focuses on the parametric design technology research of landscape architecture. First, it starts from the design of large areas, analyzes the status quo of regional scale, and then starts to design with the help of three-dimensional model. First will get the clusters of 3 d model the coarse segmentation class continue to split into more rules part, according to the characteristics of the model data, the first of 3 d model to differential information to estimate the point cloud data, calculate the normal vector and curvature of point cloud data, through a point cloud registration technology will point cloud unified under different Angle of view to the same coordinate system, through the big data landscape algorithm for geometric feature and image feature point detection, detailed point cloud data processing algorithms in the process of coarse segmentation of regional scale in the same cluster of different object segmentation, again through the panoramic image segmentation for interior point cloud of geometric structure information. The data can be classified intelligently, the corresponding point cloud data can be marked, and the geometric structure information can be marked into the point cloud data to be segmented through point cloud matching, so as to enhance knowledge reserves, find problems and solve problems timely.

@article{maciag_unsupervised_2021,
	title = {Unsupervised {Anomaly} {Detection} in stream data with {Online} evolving {Spiking} {Neural} {Networks}},
	issn = {0893-6080},
	url = {https://www.sciencedirect.com/science/article/pii/S0893608021000599},
	doi = {10.1016/j.neunet.2021.02.017},
	abstract = {Unsupervised anomaly discovery in stream data is a research topic with many practical applications. However, in many cases, it is not easy to collect enough training data with labeled anomalies for supervised learning of an anomaly detector in order to deploy it later for identification of real anomalies in streaming data. It is thus important to design anomalies detectors that can correctly detect anomalies without access to labeled training data. Our idea is to adapt the Online evolving Spiking Neural Network (OeSNN) classifier to the anomaly detection task. As a result, we offer an Online evolving Spiking Neural Network for Unsupervised Anomaly Detection algorithm (OeSNN-UAD), which, unlike OeSNN, works in an unsupervised way and does not separate output neurons into disjoint decision classes. OeSNN-UAD uses our proposed new two-step anomaly detection method. Also, we derive new theoretical properties of neuronal model and input layer encoding of OeSNN, which enable more effective and efficient detection of anomalies in our OeSNN-UAD approach. The proposed OeSNN-UAD detector was experimentally compared with state-of-the-art unsupervised and semi-supervised detectors of anomalies in stream data from the Numenta Anomaly Benchmark and Yahoo Anomaly Datasets repositories. Our approach outperforms the other solutions provided in the literature in the case of data streams from the Numenta Anomaly Benchmark repository. Also, in the case of real data files of the Yahoo Anomaly Benchmark repository, OeSNN-UAD outperforms other selected algorithms, whereas in the case of Yahoo Anomaly Benchmark synthetic data files, it provides competitive results to the results recently reported in the literature.},
	language = {en},
	urldate = {2021-02-28},
	journal = {Neural Networks},
	author = {Maciąg, Piotr S. and Kryszkiewicz, Marzena and Bembenik, Robert and Lobo, Jesus L. and Del Ser, Javier},
	month = feb,
	year = {2021},
	keywords = {Anomaly detection, Evolving Spiking Neural Networks, Online learning, Outliers detection, Stream data, Time series data, Unsupervised anomaly detection},
}

Unsupervised anomaly discovery in stream data is a research topic with many practical applications. However, in many cases, it is not easy to collect enough training data with labeled anomalies for supervised learning of an anomaly detector in order to deploy it later for identification of real anomalies in streaming data. It is thus important to design anomalies detectors that can correctly detect anomalies without access to labeled training data. Our idea is to adapt the Online evolving Spiking Neural Network (OeSNN) classifier to the anomaly detection task. As a result, we offer an Online evolving Spiking Neural Network for Unsupervised Anomaly Detection algorithm (OeSNN-UAD), which, unlike OeSNN, works in an unsupervised way and does not separate output neurons into disjoint decision classes. OeSNN-UAD uses our proposed new two-step anomaly detection method. Also, we derive new theoretical properties of neuronal model and input layer encoding of OeSNN, which enable more effective and efficient detection of anomalies in our OeSNN-UAD approach. The proposed OeSNN-UAD detector was experimentally compared with state-of-the-art unsupervised and semi-supervised detectors of anomalies in stream data from the Numenta Anomaly Benchmark and Yahoo Anomaly Datasets repositories. Our approach outperforms the other solutions provided in the literature in the case of data streams from the Numenta Anomaly Benchmark repository. Also, in the case of real data files of the Yahoo Anomaly Benchmark repository, OeSNN-UAD outperforms other selected algorithms, whereas in the case of Yahoo Anomaly Benchmark synthetic data files, it provides competitive results to the results recently reported in the literature.

@article{zhang_transfer_2021,
	title = {Transfer learning using deep representation regularization in remaining useful life prediction across operating conditions},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021001095},
	doi = {10.1016/j.ress.2021.107556},
	abstract = {Intelligent data-driven system prognostic methods have been popularly developed in the recent years. Despite the promising results, most approaches assume the training and testing data are from the same operating condition. In the real industries, it is quite common that different machine entities work under different scenarios, that results in performance deteriorations of the data-driven prognostic methods. This paper proposes a transfer learning method for remaining useful life predictions using deep representation regularization. The practical and challenging scenario is investigated, where the training and testing data are from different machinery operating conditions, and no target-domain run-to-failure data is available for training. In the deep learning framework, data alignment schemes are proposed in the representation sub-space, including healthy state alignment, degradation direction alignment, degradation level regularization and degradation fusion. In this way, the life-cycle data of different machine entities across domains can follow the same degradation trace, thus achieving prognostic knowledge transfer. Extensive experiments on the aero-engine dataset validate the effectiveness of the proposed method, which offers a promising solution for industrial prognostics.},
	language = {en},
	urldate = {2021-02-23},
	journal = {Reliability Engineering \& System Safety},
	author = {Zhang, Wei and Li, Xiang and Ma, Hui and Luo, Zhong and Li, Xu},
	month = feb,
	year = {2021},
	keywords = {Data alignment, Deep learning, Prognosis, Remaining useful life prediction, Representation learning},
	pages = {107556},
}

Intelligent data-driven system prognostic methods have been popularly developed in the recent years. Despite the promising results, most approaches assume the training and testing data are from the same operating condition. In the real industries, it is quite common that different machine entities work under different scenarios, that results in performance deteriorations of the data-driven prognostic methods. This paper proposes a transfer learning method for remaining useful life predictions using deep representation regularization. The practical and challenging scenario is investigated, where the training and testing data are from different machinery operating conditions, and no target-domain run-to-failure data is available for training. In the deep learning framework, data alignment schemes are proposed in the representation sub-space, including healthy state alignment, degradation direction alignment, degradation level regularization and degradation fusion. In this way, the life-cycle data of different machine entities across domains can follow the same degradation trace, thus achieving prognostic knowledge transfer. Extensive experiments on the aero-engine dataset validate the effectiveness of the proposed method, which offers a promising solution for industrial prognostics.

@article{munirathinam_drift_2021,
	series = {Proceedings of the 2nd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing} ({ISM} 2020)},
	title = {Drift {Detection} {Analytics} for {IoT} {Sensors}},
	volume = {180},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050921003951},
	doi = {10.1016/j.procs.2021.01.341},
	abstract = {The Industrial Internet of Things (IoT) has an unique opportunity to have a greater impact on the manufacturing sector. Monitoring the health of the expensive equipments in the factory is critical for the business and the opportunities where IoT can be truly transformational. Most often the industries operates on a primitive way of monitoring these equipments on a Statistical Process Control (SPC) Limits. The major flaw in this monitoring system is unable to detect drifts within the static limit and upon triggering of the limits, it is usually too late for the team in the manufacturing floor to take preventive actions before the system goes down. In this paper, we developed a generic model for detecting drifts and identifying potential outliers. The model uses a double linear regression method to identify both aggressive and progressive drift, as well as adjusted boxplot method to detect outliers in both symmetric and skewed distributions. Unlike conventional drift detection approaches, this model has low computational complexity and can be applied to both batch and stream data. This paper will also introduce the infrastructure and architecture on enabling near real-time analytics using the IoT platform and streaming cluster, which reduces the data latency available for analysis to 10 minutes. Enabling real-time monitoring allows the end users to react to the alarms in a timely manner. This system has proven that it is able to provide early detection before the impact was observed as compared to the existing system. Manufacturing operation team could establish a new business process to respond to the early drift alarms by using quality shift left approach.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Procedia Computer Science},
	author = {Munirathinam, Sathyan},
	month = jan,
	year = {2021},
	keywords = {Analytics, Big Data, Concept Drift, Drift, IoT, Machine Learning, Sensor, Statistical Control Limits},
	pages = {903--912},
}

The Industrial Internet of Things (IoT) has an unique opportunity to have a greater impact on the manufacturing sector. Monitoring the health of the expensive equipments in the factory is critical for the business and the opportunities where IoT can be truly transformational. Most often the industries operates on a primitive way of monitoring these equipments on a Statistical Process Control (SPC) Limits. The major flaw in this monitoring system is unable to detect drifts within the static limit and upon triggering of the limits, it is usually too late for the team in the manufacturing floor to take preventive actions before the system goes down. In this paper, we developed a generic model for detecting drifts and identifying potential outliers. The model uses a double linear regression method to identify both aggressive and progressive drift, as well as adjusted boxplot method to detect outliers in both symmetric and skewed distributions. Unlike conventional drift detection approaches, this model has low computational complexity and can be applied to both batch and stream data. This paper will also introduce the infrastructure and architecture on enabling near real-time analytics using the IoT platform and streaming cluster, which reduces the data latency available for analysis to 10 minutes. Enabling real-time monitoring allows the end users to react to the alarms in a timely manner. This system has proven that it is able to provide early detection before the impact was observed as compared to the existing system. Manufacturing operation team could establish a new business process to respond to the early drift alarms by using quality shift left approach.

@article{gallo_industry_2021,
	series = {Proceedings of the 2nd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing} ({ISM} 2020)},
	title = {Industry 4.0 and human factor: {How} is technology changing the role of the maintenance operator?},
	volume = {180},
	issn = {1877-0509},
	shorttitle = {Industry 4.0 and human factor},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050921004415},
	doi = {10.1016/j.procs.2021.01.364},
	abstract = {Industry 4.0 is revolutionizing not only the manufacturing industry but also maintenance strategies. As consequence of the introduction of Industry 4.0 technologies, new skills are demanded to maintenance operators that has to be able to interact, as instance, with Cyber Physical Systems and robots. In this paper, we first investigate the state-of-the-art of Industry 4.0 technologies that are transforming operations and production management and finally we discuss how the role of maintenance operators is changed in a such digitalized environment. We found that, the maintenance Operator 4.0 should be able to find relevant information and predict events by a proper use of Big Data analytics, in addition to the ability of interacting with computers, digital databases and robots. Finally, the ability to rapidly adapt his skills to innovations is also strongly demanded.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Procedia Computer Science},
	author = {Gallo, Tommaso and Santolamazza, Annalisa},
	month = jan,
	year = {2021},
	keywords = {Industry 4.0, maintenance, maintenance Operator 4.0, smart factory},
	pages = {388--393},
}

Industry 4.0 is revolutionizing not only the manufacturing industry but also maintenance strategies. As consequence of the introduction of Industry 4.0 technologies, new skills are demanded to maintenance operators that has to be able to interact, as instance, with Cyber Physical Systems and robots. In this paper, we first investigate the state-of-the-art of Industry 4.0 technologies that are transforming operations and production management and finally we discuss how the role of maintenance operators is changed in a such digitalized environment. We found that, the maintenance Operator 4.0 should be able to find relevant information and predict events by a proper use of Big Data analytics, in addition to the ability of interacting with computers, digital databases and robots. Finally, the ability to rapidly adapt his skills to innovations is also strongly demanded.

@article{bona_implementation_2021,
	series = {Proceedings of the 2nd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing} ({ISM} 2020)},
	title = {Implementation of {Industry} 4.0 technology: {New} opportunities and challenges for maintenance strategy},
	volume = {180},
	issn = {1877-0509},
	shorttitle = {Implementation of {Industry} 4.0 technology},
	url = {https://www.sciencedirect.com/science/article/pii/S187705092100301X},
	doi = {10.1016/j.procs.2021.01.258},
	abstract = {Industry 4.0 is revolutionizing decision-making processes within the manufacturing industry. Maintenance strategies play a crucial role to improve progressively technical performances and economical savings. The introduction of Industry 4.0 technology results in relevant innovations able to condition maintenance policies. Moreover, innovative solutions can be introduced, such as “remote maintenance” and the “self-maintenance”. In this paper, we investigate the state-of-the-art of technologies in a “smart factory” with the aim to understand how Industry 4.0 technologies are affecting maintenance policies and to discuss their implication in strategies. We found important trends in maintenance policies, such as “remote maintenance” and the attractive option of the “autonomous maintenance”. This study represents the first comprehensive investigation in these research themes, and it desires to produce a broader insight and knowledge of current trends and main difficulties, highlighting critical aspects and disadvantages for the implementation of innovative policies.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Procedia Computer Science},
	author = {Bona, Gianpaolo Di and Cesarotti, Vittorio and Arcese, Gabriella and Gallo, Tommaso},
	month = jan,
	year = {2021},
	keywords = {Industry 4.0, maintenance policies, maintenance strategies, smart factory},
	pages = {424--429},
}

Industry 4.0 is revolutionizing decision-making processes within the manufacturing industry. Maintenance strategies play a crucial role to improve progressively technical performances and economical savings. The introduction of Industry 4.0 technology results in relevant innovations able to condition maintenance policies. Moreover, innovative solutions can be introduced, such as “remote maintenance” and the “self-maintenance”. In this paper, we investigate the state-of-the-art of technologies in a “smart factory” with the aim to understand how Industry 4.0 technologies are affecting maintenance policies and to discuss their implication in strategies. We found important trends in maintenance policies, such as “remote maintenance” and the attractive option of the “autonomous maintenance”. This study represents the first comprehensive investigation in these research themes, and it desires to produce a broader insight and knowledge of current trends and main difficulties, highlighting critical aspects and disadvantages for the implementation of innovative policies.

@article{sajid_data_2021,
	title = {Data science applications for predictive maintenance and materials science in context to {Industry} 4.0},
	issn = {2214-7853},
	url = {https://www.sciencedirect.com/science/article/pii/S221478532100448X},
	doi = {10.1016/j.matpr.2021.01.357},
	abstract = {With the revolutionising of the industry to the next generations, machines have become more complicated. If they are not put to regular maintenance then there is more breakdown and disruption in the production line. These days, data science techniques have applications over almost every field and likewise are being applied to Industry 4.0. In this advanced setup, massive data is created and stored every second. Experts with expertise in advanced mathematical and computational skills are in demand to identify root causes of failures and quality deviations of a machine, contributing to minimising a loss in time and money. Moreover, new elements with tailored properties can be discovered with material theories and computational skills. The integration of data science with industry 4.0 will increase efficiency and will be helpful to predict the quality of material minimising the production line cost and time. Different research articles on industry 4.0, data science and predictive maintenance are identified and studied. This paper identifies five critical processes of data scientists for predictive maintenance and discussed briefly through a literature review. Data science uses various processes, scientific methods, and algorithms to extract knowledge from a large amount of data. It can collect a massive amount of industrial data, which is further used to improve the manufacturing systems' efficiency and reliability. It helps analyse the data and become essential for Industry 4.0.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Materials Today: Proceedings},
	author = {Sajid, Sufiyan and Haleem, Abid and Bahl, Shashi and Javaid, Mohd and Goyal, Tarun and Mittal, Manoj},
	month = feb,
	year = {2021},
	keywords = {Data science, Decision making, Industry 4.0, Machine learning, Materials, Predictive maintenance},
}

With the revolutionising of the industry to the next generations, machines have become more complicated. If they are not put to regular maintenance then there is more breakdown and disruption in the production line. These days, data science techniques have applications over almost every field and likewise are being applied to Industry 4.0. In this advanced setup, massive data is created and stored every second. Experts with expertise in advanced mathematical and computational skills are in demand to identify root causes of failures and quality deviations of a machine, contributing to minimising a loss in time and money. Moreover, new elements with tailored properties can be discovered with material theories and computational skills. The integration of data science with industry 4.0 will increase efficiency and will be helpful to predict the quality of material minimising the production line cost and time. Different research articles on industry 4.0, data science and predictive maintenance are identified and studied. This paper identifies five critical processes of data scientists for predictive maintenance and discussed briefly through a literature review. Data science uses various processes, scientific methods, and algorithms to extract knowledge from a large amount of data. It can collect a massive amount of industrial data, which is further used to improve the manufacturing systems' efficiency and reliability. It helps analyse the data and become essential for Industry 4.0.

@article{drakaki_recent_2021,
	series = {Proceedings of the 2nd {International} {Conference} on {Industry} 4.0 and {Smart} {Manufacturing} ({ISM} 2020)},
	title = {Recent {Developments} {Towards} {Industry} 4.0 {Oriented} {Predictive} {Maintenance} in {Induction} {Motors}},
	volume = {180},
	issn = {1877-0509},
	url = {https://www.sciencedirect.com/science/article/pii/S1877050921003999},
	doi = {10.1016/j.procs.2021.01.345},
	abstract = {Predictive maintenance (PdM) for smart manufacturing and Industry 4.0 has been associated with manufacturing intelligence supported by Artificial Intelligence (AI). Therefore, PdM also relies on the smart manufacturing technologies including cyber-physical system (CPS) and big data analytics. The multi-agent system (MAS) technology and deep learning (DL) have shown the capacity to provide efficient tools for the implementation of PdM in a CPS enabled smart industrial production system gaining feedback from big data analytics. Induction motors (IM) constitute the main power source in the industrial production environment and therefore their maintenance and early fault detection and diagnosis (FD/D) is a critical process. Neural network (NN) based FD/D of IM has been widely used in order to identify different fault types. DL methods have recently emerged for FD/D of IM and can efficiently analyze massive data coming from different machine sensors. The MAS has recently been used in combination with artificial NNs as a decision support tool for FD/D of IM. This paper aims to provide a review of recent trends in PdM of IM focusing on MAS and DL based FD/D methods that have emerged in the last 5 years due to their potential to be implemented in a smart manufacturing system. A discussion of the presented methods is given in order to present the recent developments and trends and provide future directions for research.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Procedia Computer Science},
	author = {Drakaki, Maria and Karnavas, Yannis L. and Tzionas, Panagiotis and Chasiotis, Ioannis D.},
	month = jan,
	year = {2021},
	keywords = {Predictive maintenance, deep learning, fault detection, fault diagnosis, induction motor, machinery health management, multi-agent system, neural networks},
	pages = {943--949},
}

Predictive maintenance (PdM) for smart manufacturing and Industry 4.0 has been associated with manufacturing intelligence supported by Artificial Intelligence (AI). Therefore, PdM also relies on the smart manufacturing technologies including cyber-physical system (CPS) and big data analytics. The multi-agent system (MAS) technology and deep learning (DL) have shown the capacity to provide efficient tools for the implementation of PdM in a CPS enabled smart industrial production system gaining feedback from big data analytics. Induction motors (IM) constitute the main power source in the industrial production environment and therefore their maintenance and early fault detection and diagnosis (FD/D) is a critical process. Neural network (NN) based FD/D of IM has been widely used in order to identify different fault types. DL methods have recently emerged for FD/D of IM and can efficiently analyze massive data coming from different machine sensors. The MAS has recently been used in combination with artificial NNs as a decision support tool for FD/D of IM. This paper aims to provide a review of recent trends in PdM of IM focusing on MAS and DL based FD/D methods that have emerged in the last 5 years due to their potential to be implemented in a smart manufacturing system. A discussion of the presented methods is given in order to present the recent developments and trends and provide future directions for research.

@article{han_remaining_2021,
	title = {Remaining useful life prediction and predictive maintenance strategies for multi-state manufacturing systems considering functional dependence},
	volume = {210},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832021001137},
	doi = {10.1016/j.ress.2021.107560},
	abstract = {The performance states of the manufacturing equipment and the quality states of the manufactured products are important indicators for the operational state evaluation and maintenance decision of the multi-state system. Further, the performance degradation of manufacturing components shows some dependence on the decline in product quality. However, the traditional remaining useful life (RUL) prediction and maintenance strategy of manufacturing system are limited to the dependence of the manufacturing components performance degradation. Based on the RUL prediction model that considers the components dependence for product quality requirements, a system predictive maintenance method based on the component functional importance is proposed. First, the connotation of degradation mechanism, functional dependence and RUL for manufacturing system is expounded. Second, a mission reliability oriented RUL prediction method for manufacturing systems is developed based on the functional dependence of components. Third, an approach for average maintenance cost calculation is proposed based on dynamic RUL prediction after each maintenance action, and the functional importance is applied to prioritize the predictive maintenance component-sets. Finally, the case results show that the proposed approach can ensure the ability of manufacturing systems to complete production tasks with high quality product, and reduce the maintenance cost in the production cycle simultaneously.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Reliability Engineering \& System Safety},
	author = {Han, Xiao and Wang, Zili and Xie, Min and He, Yihai and Li, Yao and Wang, Wenzhuo},
	month = jun,
	year = {2021},
	keywords = {Functional dependence, average maintenance cost, mission reliability, predictive maintenance, remaining useful life},
	pages = {107560},
}

The performance states of the manufacturing equipment and the quality states of the manufactured products are important indicators for the operational state evaluation and maintenance decision of the multi-state system. Further, the performance degradation of manufacturing components shows some dependence on the decline in product quality. However, the traditional remaining useful life (RUL) prediction and maintenance strategy of manufacturing system are limited to the dependence of the manufacturing components performance degradation. Based on the RUL prediction model that considers the components dependence for product quality requirements, a system predictive maintenance method based on the component functional importance is proposed. First, the connotation of degradation mechanism, functional dependence and RUL for manufacturing system is expounded. Second, a mission reliability oriented RUL prediction method for manufacturing systems is developed based on the functional dependence of components. Third, an approach for average maintenance cost calculation is proposed based on dynamic RUL prediction after each maintenance action, and the functional importance is applied to prioritize the predictive maintenance component-sets. Finally, the case results show that the proposed approach can ensure the ability of manufacturing systems to complete production tasks with high quality product, and reduce the maintenance cost in the production cycle simultaneously.

@article{ingemarsdotter_challenges_2021,
	title = {Challenges and {Solutions} in condition-based maintenance implementation - a multiple case study},
	issn = {0959-6526},
	url = {https://www.sciencedirect.com/science/article/pii/S0959652621006405},
	doi = {10.1016/j.jclepro.2021.126420},
	abstract = {Previous literature has highlighted many opportunities for digital technologies, such as the Internet of Things (IoT) and data analytics, to enable circular strategies, i.e., strategies which support the transition to a circular economy (CE). As one of the key circular strategies for which the digital opportunities are apparent, maintenance is selected as the focus area for this study. In the field of maintenance, IoT and data analytics enable companies to implement condition-based maintenance (CBM), i.e., maintenance based on monitoring the actual condition of products in the field. CBM can lead to more timely and efficient maintenance, better performing products-in-use, reduced downtime in operations, and longer product lifetimes. Despite these benefits, CBM implementation in practice is still limited. The aim of this research is thus to understand the challenges related to CBM implementation in practice, and to extract solutions which companies have applied to address these challenges. Towards this aim, a multiple case study is conducted at three original equipment manufacturers (OEMs). A framework is derived which allows for a broad analysis of challenges and solutions in the cases. We identify 19 challenges and 16 solutions and translate these into a set of actionable recommendations. Our findings contribute to the field of CBM with a comprehensive view of challenges and solutions in practice, from the OEM’s point of view. Moreover, we contribute to CE literature with a concrete case study about IoT-enabled circular strategy implementation.},
	language = {en},
	urldate = {2021-02-22},
	journal = {Journal of Cleaner Production},
	author = {Ingemarsdotter, Emilia and Kambanou, Marianna Lena and Jamsin, Ella and Sakao, Tomohiko and Balkenende, Ruud},
	month = feb,
	year = {2021},
	keywords = {Case study, Circular Economy, Condition-Based Maintenance, Digitalization, Internet of Things},
	pages = {126420},
}

Previous literature has highlighted many opportunities for digital technologies, such as the Internet of Things (IoT) and data analytics, to enable circular strategies, i.e., strategies which support the transition to a circular economy (CE). As one of the key circular strategies for which the digital opportunities are apparent, maintenance is selected as the focus area for this study. In the field of maintenance, IoT and data analytics enable companies to implement condition-based maintenance (CBM), i.e., maintenance based on monitoring the actual condition of products in the field. CBM can lead to more timely and efficient maintenance, better performing products-in-use, reduced downtime in operations, and longer product lifetimes. Despite these benefits, CBM implementation in practice is still limited. The aim of this research is thus to understand the challenges related to CBM implementation in practice, and to extract solutions which companies have applied to address these challenges. Towards this aim, a multiple case study is conducted at three original equipment manufacturers (OEMs). A framework is derived which allows for a broad analysis of challenges and solutions in the cases. We identify 19 challenges and 16 solutions and translate these into a set of actionable recommendations. Our findings contribute to the field of CBM with a comprehensive view of challenges and solutions in practice, from the OEM’s point of view. Moreover, we contribute to CE literature with a concrete case study about IoT-enabled circular strategy implementation.

@article{soares_unsupervised_2021,
	title = {Unsupervised {Machine} {Learning} {Techniques} to {Prevent} {Faults} in {Railroad} {Switch} {Machines}},
	issn = {1874-5482},
	url = {https://www.sciencedirect.com/science/article/pii/S1874548221000159},
	doi = {10.1016/j.ijcip.2021.100423},
	abstract = {Railroad switch machines are essential electromechanical equipment in a railway network, and the occurrence of failures in such equipment can cause railroad interruptions and lead to potential economic losses. Thus, early diagnosis of these failures can represent a reduction in costs and an increase in productivity. This paper aims to propose a predictive model based on computational intelligence techniques, to solve this problem. The applied methodology includes feature extraction and selection procedures based on hypothesis tests and unsupervised machine learning models. The proposed model was tested in a database made available by a Brazilian railway company and proved to be efficient once it has considered critical operations conducted in the vicinity of the ones classified as faults.},
	language = {en},
	urldate = {2021-02-15},
	journal = {International Journal of Critical Infrastructure Protection},
	author = {Soares, Nielson and Aguiar, Eduardo Pestana de and Souza, Amanda Campos and Goliatt, Leonardo},
	month = feb,
	year = {2021},
	keywords = {Computational Intelligence, Failure Prediction, Machine Learning, Railroad switch},
	pages = {100423},
}

Railroad switch machines are essential electromechanical equipment in a railway network, and the occurrence of failures in such equipment can cause railroad interruptions and lead to potential economic losses. Thus, early diagnosis of these failures can represent a reduction in costs and an increase in productivity. This paper aims to propose a predictive model based on computational intelligence techniques, to solve this problem. The applied methodology includes feature extraction and selection procedures based on hypothesis tests and unsupervised machine learning models. The proposed model was tested in a database made available by a Brazilian railway company and proved to be efficient once it has considered critical operations conducted in the vicinity of the ones classified as faults.

@article{gbadamosi_iot_2021,
	title = {{IoT} for predictive assets monitoring and maintenance: {An} implementation strategy for the {UK} rail industry},
	volume = {122},
	issn = {0926-5805},
	shorttitle = {{IoT} for predictive assets monitoring and maintenance},
	url = {http://www.sciencedirect.com/science/article/pii/S0926580520310669},
	doi = {10.1016/j.autcon.2020.103486},
	abstract = {With about 100\% increase in rail service usage over the last 20 years, it is pertinent that rail infrastructure continues to function at an optimal level to avoid service disruptions, cancellations or delays due to unforeseen asset breakdown. In an endeavour to propose a strategy for the implementation of Internet of Things (IoT) in rail asset maintenance, a qualitative methodology was adopted through a series of focus-group workshops to identify the priority areas and enabling digital technologies for IoT implementation. The methods of data collection included audio recording, note-taking, and concept mapping. The audio records were transcribed and used for thematic analysis, while the concept maps were integrated for conceptual modelling and analysis. This paper presents an implementation strategy for IoT for rail assets maintenance with focus on priority areas such as real-time condition monitoring using IoT sensors, predictive maintenance, remote inspection, and integrated asset data management platform.},
	language = {en},
	urldate = {2020-12-08},
	journal = {Automation in Construction},
	author = {Gbadamosi, Abdul-Quayyum and Oyedele, Lukumon O. and Delgado, Juan Manuel Davila and Kusimo, Habeeb and Akanbi, Lukman and Olawale, Oladimeji and Muhammed-yakubu, Naimah},
	month = feb,
	year = {2021},
	keywords = {Augmented reality, Internet of things, Predictive maintenance, Rail assets, Remote inspection},
	pages = {103486},
}

With about 100% increase in rail service usage over the last 20 years, it is pertinent that rail infrastructure continues to function at an optimal level to avoid service disruptions, cancellations or delays due to unforeseen asset breakdown. In an endeavour to propose a strategy for the implementation of Internet of Things (IoT) in rail asset maintenance, a qualitative methodology was adopted through a series of focus-group workshops to identify the priority areas and enabling digital technologies for IoT implementation. The methods of data collection included audio recording, note-taking, and concept mapping. The audio records were transcribed and used for thematic analysis, while the concept maps were integrated for conceptual modelling and analysis. This paper presents an implementation strategy for IoT for rail assets maintenance with focus on priority areas such as real-time condition monitoring using IoT sensors, predictive maintenance, remote inspection, and integrated asset data management platform.

@article{pinciroli_semi-supervised_2021,
	title = {A semi-supervised method for the characterization of degradation of nuclear power plants steam generators},
	volume = {131},
	issn = {0149-1970},
	url = {http://www.sciencedirect.com/science/article/pii/S0149197020303279},
	doi = {10.1016/j.pnucene.2020.103580},
	abstract = {The digitalization of nuclear power plants, with the rapid growth of information technology, opens the door to the development of new methods of condition-based maintenance. In this work, a semi-supervised method for characterizing the level of degradation of nuclear power plant components using measurements collected during plant operational transients is proposed. It is based on the fusion of selected features extracted from the monitored signals. Feature selection is formulated as a multi-objective optimization problem. The objectives are the maximization of the feature monotonicity and trendability, and the maximization of a novel measure of correlation between the feature values and the results of non-destructive tests performed to assess the component degradation. The features of the Pareto optimal set are normalized and the component degradation level is defined as the median of the obtained values. The developed method is applied to real data collected from steam generators of pressurized water reactors. It is shown able to identify degradation level with errors comparable to those obtained by ad-hoc non-destructive tests.},
	language = {en},
	urldate = {2020-12-08},
	journal = {Progress in Nuclear Energy},
	author = {Pinciroli, Luca and Baraldi, Piero and Shokry, Ahmed and Zio, Enrico and Seraoui, Redouane and Mai, Carole},
	month = jan,
	year = {2021},
	keywords = {Condition-based maintenance, Degradation assessment, Feature selection, Nuclear power plant, Semi-supervised, Steam generator},
	pages = {103580},
}

The digitalization of nuclear power plants, with the rapid growth of information technology, opens the door to the development of new methods of condition-based maintenance. In this work, a semi-supervised method for characterizing the level of degradation of nuclear power plant components using measurements collected during plant operational transients is proposed. It is based on the fusion of selected features extracted from the monitored signals. Feature selection is formulated as a multi-objective optimization problem. The objectives are the maximization of the feature monotonicity and trendability, and the maximization of a novel measure of correlation between the feature values and the results of non-destructive tests performed to assess the component degradation. The features of the Pareto optimal set are normalized and the component degradation level is defined as the median of the obtained values. The developed method is applied to real data collected from steam generators of pressurized water reactors. It is shown able to identify degradation level with errors comparable to those obtained by ad-hoc non-destructive tests.

@article{voronov_forest-based_2021,
	title = {A forest-based algorithm for selecting informative variables using {Variable} {Depth} {Distribution}},
	volume = {97},
	issn = {0952-1976},
	url = {http://www.sciencedirect.com/science/article/pii/S0952197620303341},
	doi = {10.1016/j.engappai.2020.104073},
	abstract = {Predictive maintenance of systems and their components in technical systems is a promising approach to optimize system usage and reduce system downtime. Various sensor data are logged during system operation for different purposes, but sometimes not directly related to the degradation of a specific component. Variable selection algorithms are necessary to reduce model complexity and improve interpretability of diagnostic and prognostic algorithms. This paper presents a forest-based variable selection algorithm that analyzes the distribution of a variable in the decision tree structure, called Variable Depth Distribution, to measure its importance. The proposed variable selection algorithm is developed for datasets with correlated variables that pose problems for existing forest-based variable selection methods. The proposed variable selection method is evaluated and analyzed using three case studies: survival analysis of lead–acid batteries in heavy-duty vehicles, engine misfire detection, and a simulated prognostics dataset. The results show the usefulness of the proposed algorithm, with respect to existing forest-based methods, and its ability to identify important variables in different applications. As an example, the battery prognostics case study shows that similar predictive performance is achieved when only 17\% percent of the variables are used compared to all measured signals.},
	language = {en},
	urldate = {2020-11-30},
	journal = {Engineering Applications of Artificial Intelligence},
	author = {Voronov, Sergii and Jung, Daniel and Frisk, Erik},
	month = jan,
	year = {2021},
	keywords = {Automotive, Random Forest, Random Survival Forest, Variable selection},
	pages = {104073},
}

Predictive maintenance of systems and their components in technical systems is a promising approach to optimize system usage and reduce system downtime. Various sensor data are logged during system operation for different purposes, but sometimes not directly related to the degradation of a specific component. Variable selection algorithms are necessary to reduce model complexity and improve interpretability of diagnostic and prognostic algorithms. This paper presents a forest-based variable selection algorithm that analyzes the distribution of a variable in the decision tree structure, called Variable Depth Distribution, to measure its importance. The proposed variable selection algorithm is developed for datasets with correlated variables that pose problems for existing forest-based variable selection methods. The proposed variable selection method is evaluated and analyzed using three case studies: survival analysis of lead–acid batteries in heavy-duty vehicles, engine misfire detection, and a simulated prognostics dataset. The results show the usefulness of the proposed algorithm, with respect to existing forest-based methods, and its ability to identify important variables in different applications. As an example, the battery prognostics case study shows that similar predictive performance is achieved when only 17% percent of the variables are used compared to all measured signals.

@article{sun_online_2021,
	title = {Online oil debris monitoring of rotating machinery: {A} detailed review of more than three decades},
	volume = {149},
	issn = {0888-3270},
	shorttitle = {Online oil debris monitoring of rotating machinery},
	url = {http://www.sciencedirect.com/science/article/pii/S0888327020307275},
	doi = {10.1016/j.ymssp.2020.107341},
	abstract = {Oil debris monitoring has played an irreplaceable role in ascertaining the health condition of rotating machinery (e.g. engine, gearbox). Although many sensing methods for detecting wear-generated particles of rotating elements have been presented, a comprehensive review paper on these technologies is still missing. To this end, this paper provides a detailed survey of the advances in oil debris monitoring for the online health monitoring of rotating machinery. According to the detection mechanism, these sensing technologies are classified under varying categories (magnetic: magnetic chip detectors and inductive sensors, electrical: resistive-capacitive sensors and electrostatic sensors, optical: photoelectric sensors and imaging sensors, acoustic). The systematic analysis and commentary on each sensing method are conducted, and real applications also be reviewed. These sensing technologies are not confined to research-related activities, of which some have already been patented and commercialized. Finally, future works are presented to meet the challenges faced by online monitoring of rotating machinery.},
	language = {en},
	urldate = {2020-10-19},
	journal = {Mechanical Systems and Signal Processing},
	author = {Sun, Jiayi and Wang, Liming and Li, Jianfeng and Li, Fangyi and Li, Jianyong and Lu, Haiyang},
	month = feb,
	year = {2021},
	keywords = {Lubricating oil, Oil debris monitoring, Online health monitoring, Rotating machinery, Wear debris sensor},
	pages = {107341},
}

Oil debris monitoring has played an irreplaceable role in ascertaining the health condition of rotating machinery (e.g. engine, gearbox). Although many sensing methods for detecting wear-generated particles of rotating elements have been presented, a comprehensive review paper on these technologies is still missing. To this end, this paper provides a detailed survey of the advances in oil debris monitoring for the online health monitoring of rotating machinery. According to the detection mechanism, these sensing technologies are classified under varying categories (magnetic: magnetic chip detectors and inductive sensors, electrical: resistive-capacitive sensors and electrostatic sensors, optical: photoelectric sensors and imaging sensors, acoustic). The systematic analysis and commentary on each sensing method are conducted, and real applications also be reviewed. These sensing technologies are not confined to research-related activities, of which some have already been patented and commercialized. Finally, future works are presented to meet the challenges faced by online monitoring of rotating machinery.

@article{wen_generalized_2021,
	title = {A generalized remaining useful life prediction method for complex systems based on composite health indicator},
	volume = {205},
	issn = {0951-8320},
	url = {http://www.sciencedirect.com/science/article/pii/S0951832020307419},
	doi = {10.1016/j.ress.2020.107241},
	abstract = {As one of the key techniques in Prognostics and Health Management (PHM), accurate Remaining Useful Life (RUL) prediction can effectively reduce the number of downtime maintenance and significantly improve economic benefits. In this paper, a generalized RUL prediction method is proposed for complex systems with multiple Condition Monitoring (CM) signals. A stochastic degradation model is proposed to characterize the system degradation behavior, based on which the respective reliability characteristics such as the RUL and its Confidence Interval (CI) are explicitly derived. Considering the degradation model, two desirable properties of the Health Indicator (HI) are put forward and their respective quantitative evaluation methods are developed. With the desirable properties, a nonlinear data fusion method based on Genetic Programming (GP) is proposed to construct a superior composite HI. In this way, the multiple CM signals are fused to provide a better prediction capability. Finally, the proposed integrated methodology is validated on the C-MAPSS data set of aircraft turbine engines.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Reliability Engineering \& System Safety},
	author = {Wen, Pengfei and Zhao, Shuai and Chen, Shaowei and Li, Yong},
	month = jan,
	year = {2021},
	keywords = {Data fusion, Degradation modeling, Multiple sensors, Prognostics, Remaining useful life},
	pages = {107241},
}

As one of the key techniques in Prognostics and Health Management (PHM), accurate Remaining Useful Life (RUL) prediction can effectively reduce the number of downtime maintenance and significantly improve economic benefits. In this paper, a generalized RUL prediction method is proposed for complex systems with multiple Condition Monitoring (CM) signals. A stochastic degradation model is proposed to characterize the system degradation behavior, based on which the respective reliability characteristics such as the RUL and its Confidence Interval (CI) are explicitly derived. Considering the degradation model, two desirable properties of the Health Indicator (HI) are put forward and their respective quantitative evaluation methods are developed. With the desirable properties, a nonlinear data fusion method based on Genetic Programming (GP) is proposed to construct a superior composite HI. In this way, the multiple CM signals are fused to provide a better prediction capability. Finally, the proposed integrated methodology is validated on the C-MAPSS data set of aircraft turbine engines.

@article{liu_complex_2020,
	title = {Complex engineered system health indexes extraction using low frequency raw time-series data based on deep learning methods},
	volume = {161},
	issn = {0263-2241},
	url = {https://www.sciencedirect.com/science/article/pii/S0263224120304280},
	doi = {10.1016/j.measurement.2020.107890},
	abstract = {Data analysis methods based on deep learning are attracting more and more attention in the field of health monitoring, fault diagnosis and failure prognostics of complex systems, such as aircraft airborne systems and engines. In this study, several health monitoring methods proposed from deep learning are demonstrated based on a real data set from an airborne system of commercial aircraft, where Health Indexes (HIs) are derived based on the raw sensor data to characterize the health state of the system in-service. Determining the optimal degradation evaluation index is the key to further failure prognostics. So, a set of metrics to characterize the suitability of different of HI deriving methods has been proposed. This metrics includes monotonicity, prognosability, and trendability. The better HI selected can effectively characterize the health state of aircraft air conditioning system, which is helpful for further failure prognostics and converting the scheduled maintenance into condition-based maintenance.},
	urldate = {2023-10-12},
	journal = {Measurement},
	author = {Liu, Cui and Sun, Jianzhong and Liu, He and Lei, Shiying and Hu, Xinhua},
	month = sep,
	year = {2020},
	keywords = {Air conditioning system, Condition-based maintenance, Deep learning, Long short term memory, System health monitoring},
	pages = {107890},
}

Data analysis methods based on deep learning are attracting more and more attention in the field of health monitoring, fault diagnosis and failure prognostics of complex systems, such as aircraft airborne systems and engines. In this study, several health monitoring methods proposed from deep learning are demonstrated based on a real data set from an airborne system of commercial aircraft, where Health Indexes (HIs) are derived based on the raw sensor data to characterize the health state of the system in-service. Determining the optimal degradation evaluation index is the key to further failure prognostics. So, a set of metrics to characterize the suitability of different of HI deriving methods has been proposed. This metrics includes monotonicity, prognosability, and trendability. The better HI selected can effectively characterize the health state of aircraft air conditioning system, which is helpful for further failure prognostics and converting the scheduled maintenance into condition-based maintenance.

@inproceedings{maschler_continual_2020,
	title = {Continual {Learning} of {Fault} {Prediction} for {Turbofan} {Engines} using {Deep} {Learning} with {Elastic} {Weight} {Consolidation}},
	volume = {1},
	doi = {10.1109/ETFA46521.2020.9211903},
	abstract = {Fault prediction based upon deep learning algorithms has great potential in industrial automation: By automatically adapting to different usage contexts, it would greatly expand the usefulness of current predictive maintenance solutions. However, restrictions regarding the centralized accumulation of data necessary for such automatic adaption call for a distributed approach to training these algorithms. Therefore, in this paper, a continual learning based algorithm for fault prediction is presented, allowing for distributed, cooperative learning by elastic weight consolidation. This algorithm is then evaluated on a large NASA turbofan engine dataset and shows promising results regarding the performant training on decentral sub-datasets for industrial automation scenarios.},
	booktitle = {2020 25th {IEEE} {International} {Conference} on {Emerging} {Technologies} and {Factory} {Automation} ({ETFA})},
	author = {Maschler, Benjamin and Vietz, Hannes and Jazdi, Nasser and Weyrich, Michael},
	month = sep,
	year = {2020},
	note = {ISSN: 1946-0759},
	keywords = {Continual learning, Deep learning, Elastic weight consolidation, Engines, Fault prognostics, Industrial Automation, Machine learning, NASA, Neural networks, Prediction algorithms, Prediction methods, Task analysis, Training, Training data, Transfer learning},
	pages = {959--966},
}

Fault prediction based upon deep learning algorithms has great potential in industrial automation: By automatically adapting to different usage contexts, it would greatly expand the usefulness of current predictive maintenance solutions. However, restrictions regarding the centralized accumulation of data necessary for such automatic adaption call for a distributed approach to training these algorithms. Therefore, in this paper, a continual learning based algorithm for fault prediction is presented, allowing for distributed, cooperative learning by elastic weight consolidation. This algorithm is then evaluated on a large NASA turbofan engine dataset and shows promising results regarding the performant training on decentral sub-datasets for industrial automation scenarios.

@inproceedings{fanaee-t_cyclefootprint_2020,
	address = {Cham},
	series = {Communications in {Computer} and {Information} {Science}},
	title = {{CycleFootprint}: {A} {Fully} {Automated} {Method} for {Extracting} {Operation} {Cycles} from {Historical} {Raw} {Data} of {Multiple} {Sensors}},
	isbn = {978-3-030-66770-2},
	shorttitle = {{CycleFootprint}},
	doi = {10.1007/978-3-030-66770-2_3},
	abstract = {Extracting operation cycles from the historical reading of sensors is an essential step in IoT data analytics. For instance, we can exploit the obtained cycles for learning the normal states to feed into semi-supervised models or dictionaries for efficient real-time anomaly detection on the sensors. However, this is a difficult problem due to this fact that we may have different types of cycles, each of which with varying lengths. Current approaches are highly dependent on manual efforts by the aid of visualization and knowledge of domain experts, which is not feasible on a large scale. We propose a fully automated method called CycleFootprint that can: 1) identify the most relevant signal that has the most obvious recurring patterns among multiple signals; and 2) automatically find the cycles from the selected signal. The main idea behind CycleFootprint is mining footprints in the cycles. We assume that there should be a unique pattern in each cycle that shows up repeatedly in each cycle. By mining those footprints, we can identify cycles. We evaluate our method with existing labeled ground truth data of a real separator in marine application equipped with multiple health monitoring sensors. 86\% of cycles extracted by our method match fully or with at least 99\% overlap with true cycles, which sounds promising given its unsupervised and fully automated nature.},
	language = {en},
	booktitle = {{IoT} {Streams} for {Data}-{Driven} {Predictive} {Maintenance} and {IoT}, {Edge}, and {Mobile} for {Embedded} {Machine} {Learning}},
	publisher = {Springer International Publishing},
	author = {Fanaee-T, Hadi and Bouguelia, Mohamed-Rafik and Rahat, Mahmoud and Blixt, Jonathan and Singh, Harpal},
	editor = {Gama, Joao and Pashami, Sepideh and Bifet, Albert and Sayed-Mouchawe, Moamar and Fröning, Holger and Pernkopf, Franz and Schiele, Gregor and Blott, Michaela},
	year = {2020},
	keywords = {Cycle detection, IoT, Sensors},
	pages = {30--44},
}

Extracting operation cycles from the historical reading of sensors is an essential step in IoT data analytics. For instance, we can exploit the obtained cycles for learning the normal states to feed into semi-supervised models or dictionaries for efficient real-time anomaly detection on the sensors. However, this is a difficult problem due to this fact that we may have different types of cycles, each of which with varying lengths. Current approaches are highly dependent on manual efforts by the aid of visualization and knowledge of domain experts, which is not feasible on a large scale. We propose a fully automated method called CycleFootprint that can: 1) identify the most relevant signal that has the most obvious recurring patterns among multiple signals; and 2) automatically find the cycles from the selected signal. The main idea behind CycleFootprint is mining footprints in the cycles. We assume that there should be a unique pattern in each cycle that shows up repeatedly in each cycle. By mining those footprints, we can identify cycles. We evaluate our method with existing labeled ground truth data of a real separator in marine application equipped with multiple health monitoring sensors. 86% of cycles extracted by our method match fully or with at least 99% overlap with true cycles, which sounds promising given its unsupervised and fully automated nature.

@article{calikus_no_2020,
	title = {No free lunch but a cheaper supper: {A} general framework for streaming anomaly detection},
	volume = {155},
	issn = {0957-4174},
	shorttitle = {No free lunch but a cheaper supper},
	url = {https://www.sciencedirect.com/science/article/pii/S0957417420302773},
	doi = {10.1016/j.eswa.2020.113453},
	abstract = {In recent years, research interest in detecting anomalies in temporal streaming data has increased significantly. A variety of algorithms are being developed in the data mining community. They can be broadly divided into two categories, namely general-purpose and ad hoc ones. In most cases, general approaches assume a one-size-fits-all solution model, and strive to design a single “optimal” anomaly detector which can detect all anomalies in any domain. To date, there exists no universal method that has been shown to outperform the others across different anomaly types, use cases and datasets. In this paper, we propose SAFARI, a framework created by abstracting and unifying the fundamental tasks within the streaming anomaly detection. SAFARI provides a flexible and extensible anomaly detection procedure to overcome the limitations of one-size-fits-all solutions. Such abstraction helps to facilitate more elaborate algorithm comparisons by allowing us to isolate the effects of shared and unique characteristics of diverse algorithms on the performance. Using the framework, we have identified a research gap that motivated us to propose a novel learning strategy. We implemented twenty different anomaly detectors and conducted an extensive evaluation study, comparing their performances using real-world benchmark datasets with different properties. The results indicate that there is no single superior detector which works perfectly for every case, proving our hypothesis that “there is no free lunch” in the streaming anomaly detection world. Finally, we discuss the benefits and drawbacks of each method in-depth, drawing a set of conclusions and guidelines to guide future users of SAFARI.},
	language = {en},
	urldate = {2023-05-21},
	journal = {Expert Systems with Applications},
	author = {Calikus, Ece and Nowaczyk, Sławomir and Sant’Anna, Anita and Dikmen, Onur},
	month = oct,
	year = {2020},
	keywords = {Anomaly detection, Online learning, Reservoir sampling, Stream mining},
	pages = {113453},
}

In recent years, research interest in detecting anomalies in temporal streaming data has increased significantly. A variety of algorithms are being developed in the data mining community. They can be broadly divided into two categories, namely general-purpose and ad hoc ones. In most cases, general approaches assume a one-size-fits-all solution model, and strive to design a single “optimal” anomaly detector which can detect all anomalies in any domain. To date, there exists no universal method that has been shown to outperform the others across different anomaly types, use cases and datasets. In this paper, we propose SAFARI, a framework created by abstracting and unifying the fundamental tasks within the streaming anomaly detection. SAFARI provides a flexible and extensible anomaly detection procedure to overcome the limitations of one-size-fits-all solutions. Such abstraction helps to facilitate more elaborate algorithm comparisons by allowing us to isolate the effects of shared and unique characteristics of diverse algorithms on the performance. Using the framework, we have identified a research gap that motivated us to propose a novel learning strategy. We implemented twenty different anomaly detectors and conducted an extensive evaluation study, comparing their performances using real-world benchmark datasets with different properties. The results indicate that there is no single superior detector which works perfectly for every case, proving our hypothesis that “there is no free lunch” in the streaming anomaly detection world. Finally, we discuss the benefits and drawbacks of each method in-depth, drawing a set of conclusions and guidelines to guide future users of SAFARI.

@article{fan_transfer_2020,
	title = {Transfer learning for remaining useful life prediction based on consensus self-organizing models},
	volume = {203},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832020305998},
	doi = {10.1016/j.ress.2020.107098},
	abstract = {The traditional paradigm for developing machine prognostics usually relies on generalization from data acquired in experiments under controlled conditions prior to deployment of the equipment. Detecting or predicting failures and estimating machine health in this way assumes that future field data will have a very similar distribution to the experiment data. However, many complex machines operate under dynamic environmental conditions and are used in many different ways. This makes collecting comprehensive data very challenging, and the assumption that pre-deployment data and post-deployment data follow very similar distributions is unlikely to hold. In this work, we present a feature-representation based transfer learning (TL) method for predicting Remaining Useful Life (RUL) of equipment, under scenarios that samples with previously unseen conditions are presented in the target domain and the labels are available only for the source domain, but not the target domain. This setting corresponds to generalizing from a limited number of run-to-failure experiments performed prior to deployment into making prognostics with data coming from deployed equipment that is being used under multiple new operating conditions and experiencing previously unseen faults. We employ a deviation detection method, Consensus Self-Organizing Models (COSMO), to create transferable features for building the RUL regression model. These features capture how different a particular equipment is in comparison to its peers. The efficiency of the proposed TL method is demonstrated using the NASA Turbofan Engine Degradation Simulation Data Set. Models using the COSMO transferable features show better performance than other methods on predicting RUL when the target domain is more complex than the source domain.},
	language = {en},
	urldate = {2023-05-21},
	journal = {Reliability Engineering \& System Safety},
	author = {Fan, Yuantao and Nowaczyk, Sławomir and Rögnvaldsson, Thorsteinn},
	month = nov,
	year = {2020},
	keywords = {Consensus self-organizing models, Domain adaptation, Feature-Representation transfer, Remaining useful life prediction, Transfer learning},
	pages = {107098},
}

The traditional paradigm for developing machine prognostics usually relies on generalization from data acquired in experiments under controlled conditions prior to deployment of the equipment. Detecting or predicting failures and estimating machine health in this way assumes that future field data will have a very similar distribution to the experiment data. However, many complex machines operate under dynamic environmental conditions and are used in many different ways. This makes collecting comprehensive data very challenging, and the assumption that pre-deployment data and post-deployment data follow very similar distributions is unlikely to hold. In this work, we present a feature-representation based transfer learning (TL) method for predicting Remaining Useful Life (RUL) of equipment, under scenarios that samples with previously unseen conditions are presented in the target domain and the labels are available only for the source domain, but not the target domain. This setting corresponds to generalizing from a limited number of run-to-failure experiments performed prior to deployment into making prognostics with data coming from deployed equipment that is being used under multiple new operating conditions and experiencing previously unseen faults. We employ a deviation detection method, Consensus Self-Organizing Models (COSMO), to create transferable features for building the RUL regression model. These features capture how different a particular equipment is in comparison to its peers. The efficiency of the proposed TL method is demonstrated using the NASA Turbofan Engine Degradation Simulation Data Set. Models using the COSMO transferable features show better performance than other methods on predicting RUL when the target domain is more complex than the source domain.

@article{schulz_computational_2020,
	title = {Computational {Psychiatry} for {Computers}},
	volume = {23},
	issn = {2589-0042},
	url = {https://www.sciencedirect.com/science/article/pii/S258900422030969X},
	doi = {10.1016/j.isci.2020.101772},
	abstract = {Computational psychiatry is a nascent field that attempts to use multi-level analyses of the underlying computational problems that we face in navigating a complex, uncertain and changing world to illuminate mental dysfunction and disease. Two particular foci of the field are the costs and benefits of environmental adaptivity and the danger and necessity of heuristics. Here, we examine the extent to which these foci and others can be used to study the actual and potential flaws of the artificial computational devices that we are increasingly inventing and empowering to navigate this very same environment on our behalf.},
	language = {en},
	number = {12},
	urldate = {2023-03-08},
	journal = {iScience},
	author = {Schulz, Eric and Dayan, Peter},
	month = dec,
	year = {2020},
	keywords = {Computer Science, Human-Computer Interaction, Psychology},
	pages = {101772},
}

Computational psychiatry is a nascent field that attempts to use multi-level analyses of the underlying computational problems that we face in navigating a complex, uncertain and changing world to illuminate mental dysfunction and disease. Two particular foci of the field are the costs and benefits of environmental adaptivity and the danger and necessity of heuristics. Here, we examine the extent to which these foci and others can be used to study the actual and potential flaws of the artificial computational devices that we are increasingly inventing and empowering to navigate this very same environment on our behalf.

@article{shanahan_artificial_2020,
	title = {Artificial {Intelligence} and the {Common} {Sense} of {Animals}},
	volume = {24},
	issn = {1364-6613},
	url = {https://www.sciencedirect.com/science/article/pii/S1364661320302163},
	doi = {10.1016/j.tics.2020.09.002},
	abstract = {The problem of common sense remains a major obstacle to progress in artificial intelligence. Here, we argue that common sense in humans is founded on a set of basic capacities that are possessed by many other animals, capacities pertaining to the understanding of objects, space, and causality. The field of animal cognition has developed numerous experimental protocols for studying these capacities and, thanks to progress in deep reinforcement learning (RL), it is now possible to apply these methods directly to evaluate RL agents in 3D environments. Besides evaluation, the animal cognition literature offers a rich source of behavioural data, which can serve as inspiration for RL tasks and curricula.},
	language = {en},
	number = {11},
	urldate = {2023-03-08},
	journal = {Trends in Cognitive Sciences},
	author = {Shanahan, Murray and Crosby, Matthew and Beyret, Benjamin and Cheke, Lucy},
	month = nov,
	year = {2020},
	pages = {862--872},
}

The problem of common sense remains a major obstacle to progress in artificial intelligence. Here, we argue that common sense in humans is founded on a set of basic capacities that are possessed by many other animals, capacities pertaining to the understanding of objects, space, and causality. The field of animal cognition has developed numerous experimental protocols for studying these capacities and, thanks to progress in deep reinforcement learning (RL), it is now possible to apply these methods directly to evaluate RL agents in 3D environments. Besides evaluation, the animal cognition literature offers a rich source of behavioural data, which can serve as inspiration for RL tasks and curricula.

@article{hadsell_embracing_2020,
	title = {Embracing {Change}: {Continual} {Learning} in {Deep} {Neural} {Networks}},
	volume = {24},
	issn = {1364-6613},
	shorttitle = {Embracing {Change}},
	url = {https://www.sciencedirect.com/science/article/pii/S1364661320302199},
	doi = {10.1016/j.tics.2020.09.004},
	abstract = {Artificial intelligence research has seen enormous progress over the past few decades, but it predominantly relies on fixed datasets and stationary environments. Continual learning is an increasingly relevant area of study that asks how artificial systems might learn sequentially, as biological systems do, from a continuous stream of correlated data. In the present review, we relate continual learning to the learning dynamics of neural networks, highlighting the potential it has to considerably improve data efficiency. We further consider the many new biologically inspired approaches that have emerged in recent years, focusing on those that utilize regularization, modularity, memory, and meta-learning, and highlight some of the most promising and impactful directions.},
	language = {en},
	number = {12},
	urldate = {2023-03-05},
	journal = {Trends in Cognitive Sciences},
	author = {Hadsell, Raia and Rao, Dushyant and Rusu, Andrei A. and Pascanu, Razvan},
	month = dec,
	year = {2020},
	keywords = {artificial intelligence, lifelong, memory, meta-learning, non-stationary},
	pages = {1028--1040},
}

Artificial intelligence research has seen enormous progress over the past few decades, but it predominantly relies on fixed datasets and stationary environments. Continual learning is an increasingly relevant area of study that asks how artificial systems might learn sequentially, as biological systems do, from a continuous stream of correlated data. In the present review, we relate continual learning to the learning dynamics of neural networks, highlighting the potential it has to considerably improve data efficiency. We further consider the many new biologically inspired approaches that have emerged in recent years, focusing on those that utilize regularization, modularity, memory, and meta-learning, and highlight some of the most promising and impactful directions.

@inproceedings{le_nguyen_challenges_2020,
	address = {Cham},
	series = {Communications in {Computer} and {Information} {Science}},
	title = {Challenges of {Stream} {Learning} for {Predictive} {Maintenance} in the {Railway} {Sector}},
	isbn = {978-3-030-66770-2},
	doi = {10.1007/978-3-030-66770-2_2},
	abstract = {Smart trains nowadays are equipped with sensors that generate an abundance of data during operation. Such data may, directly or indirectly, reflect the health state of the trains. Thus, it is of interest to analyze these data in a timely manner, preferably on-the-fly as they are being generated, to make maintenance operations more proactive and efficient. This paper provides a brief overview of predictive maintenance and stream learning, with the primary goal of leveraging stream learning in order to enhance maintenance operations in the railway sector. We justify the applicability and promising benefits of stream learning via the example of a real-world railway dataset of the train doors.},
	language = {en},
	booktitle = {{IoT} {Streams} for {Data}-{Driven} {Predictive} {Maintenance} and {IoT}, {Edge}, and {Mobile} for {Embedded} {Machine} {Learning}},
	publisher = {Springer International Publishing},
	author = {Le Nguyen, Minh Huong and Turgis, Fabien and Fayemi, Pierre-Emmanuel and Bifet, Albert},
	editor = {Gama, Joao and Pashami, Sepideh and Bifet, Albert and Sayed-Mouchawe, Moamar and Fröning, Holger and Pernkopf, Franz and Schiele, Gregor and Blott, Michaela},
	year = {2020},
	keywords = {Predictive maintenance, Railway, Stream learning},
	pages = {14--29},
}

Smart trains nowadays are equipped with sensors that generate an abundance of data during operation. Such data may, directly or indirectly, reflect the health state of the trains. Thus, it is of interest to analyze these data in a timely manner, preferably on-the-fly as they are being generated, to make maintenance operations more proactive and efficient. This paper provides a brief overview of predictive maintenance and stream learning, with the primary goal of leveraging stream learning in order to enhance maintenance operations in the railway sector. We justify the applicability and promising benefits of stream learning via the example of a real-world railway dataset of the train doors.

@inproceedings{miyata_concept_2020,
	address = {New York, NY, USA},
	series = {{WIMS} 2020},
	title = {Concept {Drift} {Detection} on {Data} {Stream} for {Revising} {DBSCAN} {Cluster}},
	isbn = {978-1-4503-7542-9},
	url = {https://doi.org/10.1145/3405962.3405990},
	doi = {10.1145/3405962.3405990},
	abstract = {Data stream mining of IoT data can help operators immediately isolate causes of equipment alarms. The challenge, however, is how to keep the classifiers high-purity (i.e., keep data of the same class in the right cluster) while dealing with the concept drifting ascribed to differences between alarm models and entities. We propose continuously revising the classification model in accordance with the data distribution and trend changes. Evaluations showed there was no purity deterioration for oscillation condition data with a drifting rate of 1\%. This result demonstrates that our approach can help operators improve their decision making.},
	urldate = {2023-02-16},
	booktitle = {Proceedings of the 10th {International} {Conference} on {Web} {Intelligence}, {Mining} and {Semantics}},
	publisher = {Association for Computing Machinery},
	author = {Miyata, Yasushi and Ishikawa, Hiroshi},
	month = aug,
	year = {2020},
	keywords = {Concept Drift, DBSCAN, clustering, data stream, power grid},
	pages = {104--110},
}

Data stream mining of IoT data can help operators immediately isolate causes of equipment alarms. The challenge, however, is how to keep the classifiers high-purity (i.e., keep data of the same class in the right cluster) while dealing with the concept drifting ascribed to differences between alarm models and entities. We propose continuously revising the classification model in accordance with the data distribution and trend changes. Evaluations showed there was no purity deterioration for oscillation condition data with a drifting rate of 1%. This result demonstrates that our approach can help operators improve their decision making.

@article{faouzi_pyts_2020,
	title = {pyts: {A} {Python} {Package} for {Time} {Series} {Classification}},
	volume = {21},
	issn = {1533-7928},
	shorttitle = {pyts},
	url = {http://jmlr.org/papers/v21/19-763.html},
	abstract = {pyts is an open-source Python package for time series classification. This versatile toolbox provides implementations of many algorithms published in the literature, preprocessing functionalities, and data set loading utilities. pyts relies on the standard scientific Python packages numpy, scipy, scikit-learn, joblib, and numba, and is distributed under the BSD-3-Clause license. Documentation contains installation instructions, a detailed user guide, a full API description, and concrete self-contained examples.},
	number = {46},
	urldate = {2023-02-12},
	journal = {Journal of Machine Learning Research},
	author = {Faouzi, Johann and Janati, Hicham},
	year = {2020},
	pages = {1--6},
}

pyts is an open-source Python package for time series classification. This versatile toolbox provides implementations of many algorithms published in the literature, preprocessing functionalities, and data set loading utilities. pyts relies on the standard scientific Python packages numpy, scipy, scikit-learn, joblib, and numba, and is distributed under the BSD-3-Clause license. Documentation contains installation instructions, a detailed user guide, a full API description, and concrete self-contained examples.

@article{benschoten_mpa_2020,
	title = {{MPA}: a novel cross-language {API} for time series analysis},
	volume = {5},
	issn = {2475-9066},
	shorttitle = {{MPA}},
	url = {https://joss.theoj.org/papers/10.21105/joss.02179},
	doi = {10.21105/joss.02179},
	abstract = {Van Benschoten et al., (2020). MPA: a novel cross-language API for time series analysis. Journal of Open Source Software, 5(49), 2179, https://doi.org/10.21105/joss.02179},
	language = {en},
	number = {49},
	urldate = {2023-02-12},
	journal = {Journal of Open Source Software},
	author = {Benschoten, Andrew H. Van and Ouyang, Austin and Bischoff, Francisco and Marrs, Tyler W.},
	month = may,
	year = {2020},
	pages = {2179},
}

Van Benschoten et al., (2020). MPA: a novel cross-language API for time series analysis. Journal of Open Source Software, 5(49), 2179, https://doi.org/10.21105/joss.02179

@article{charte_analysis_2020,
	title = {An analysis on the use of autoencoders for representation learning: {Fundamentals}, learning task case studies, explainability and challenges},
	volume = {404},
	issn = {0925-2312},
	shorttitle = {An analysis on the use of autoencoders for representation learning},
	url = {https://www.sciencedirect.com/science/article/pii/S092523122030624X},
	doi = {10.1016/j.neucom.2020.04.057},
	abstract = {In many machine learning tasks, learning a good representation of the data can be the key to building a well-performant solution. This is because most learning algorithms operate with the features in order to find models for the data. For instance, classification performance can improve if the data is mapped to a space where classes are easily separated, and regression can be facilitated by finding a manifold of data in the feature space. As a general rule, features are transformed by means of statistical methods such as principal component analysis, or manifold learning techniques such as Isomap or locally linear embedding. From a plethora of representation learning methods, one of the most versatile tools is the autoencoder. In this paper we aim to demonstrate how to influence its learned representations to achieve the desired learning behavior. To this end, we present a series of learning tasks: data embedding for visualization, image denoising, semantic hashing, detection of abnormal behaviors and instance generation. We model them from the representation learning perspective, following the state of the art methodologies in each field. A solution is proposed for each task employing autoencoders as the only learning method. The theoretical developments are put into practice using a selection of datasets for the different problems and implementing each solution, followed by a discussion of the results in each case study and a brief explanation of other six learning applications. We also explore the current challenges and approaches to explainability in the context of autoencoders. All of this helps conclude that, thanks to alterations in their structure as well as their objective function, autoencoders may be the core of a possible solution to many problems which can be modeled as a transformation of the feature space.},
	language = {en},
	urldate = {2022-12-07},
	journal = {Neurocomputing},
	author = {Charte, David and Charte, Francisco and del Jesus, María J. and Herrera, Francisco},
	month = sep,
	year = {2020},
	keywords = {Autoencoders, Deep learning, Feature extraction, Representation learning},
	pages = {93--107},
}

In many machine learning tasks, learning a good representation of the data can be the key to building a well-performant solution. This is because most learning algorithms operate with the features in order to find models for the data. For instance, classification performance can improve if the data is mapped to a space where classes are easily separated, and regression can be facilitated by finding a manifold of data in the feature space. As a general rule, features are transformed by means of statistical methods such as principal component analysis, or manifold learning techniques such as Isomap or locally linear embedding. From a plethora of representation learning methods, one of the most versatile tools is the autoencoder. In this paper we aim to demonstrate how to influence its learned representations to achieve the desired learning behavior. To this end, we present a series of learning tasks: data embedding for visualization, image denoising, semantic hashing, detection of abnormal behaviors and instance generation. We model them from the representation learning perspective, following the state of the art methodologies in each field. A solution is proposed for each task employing autoencoders as the only learning method. The theoretical developments are put into practice using a selection of datasets for the different problems and implementing each solution, followed by a discussion of the results in each case study and a brief explanation of other six learning applications. We also explore the current challenges and approaches to explainability in the context of autoencoders. All of this helps conclude that, thanks to alterations in their structure as well as their objective function, autoencoders may be the core of a possible solution to many problems which can be modeled as a transformation of the feature space.

@inproceedings{boniol_automated_2020,
	title = {Automated {Anomaly} {Detection} in {Large} {Sequences}},
	doi = {10.1109/ICDE48307.2020.00182},
	abstract = {Subsequence anomaly (or outlier) detection in long sequences is an important problem with applications in a wide range of domains. However, current approaches have severe limitations: they either require prior domain knowledge, or become cumbersome and expensive to use in situations with recurrent anomalies of the same type. In this work, we address these problems, and propose NorM, a novel approach, suitable for domain-agnostic anomaly detection. NorM is based on a new data series primitive, which permits to detect anomalies based on their (dis)similarity to a model that represents normal behavior. The experimental results on several real datasets demonstrate that the proposed approach outperforms by a large margin the current state-of-the art algorithms in terms of accuracy, while being orders of magnitude faster.},
	booktitle = {2020 {IEEE} 36th {International} {Conference} on {Data} {Engineering} ({ICDE})},
	author = {Boniol, Paul and Linardi, Michele and Roncallo, Federico and Palpanas, Themis},
	month = apr,
	year = {2020},
	note = {ISSN: 2375-026X},
	keywords = {Anomaly detection, Anomaly discovery, Computational modeling, Conferences, Data Series, Data engineering, Data models, Electrocardiography, Time Series, Time series analysis},
	pages = {1834--1837},
}

Subsequence anomaly (or outlier) detection in long sequences is an important problem with applications in a wide range of domains. However, current approaches have severe limitations: they either require prior domain knowledge, or become cumbersome and expensive to use in situations with recurrent anomalies of the same type. In this work, we address these problems, and propose NorM, a novel approach, suitable for domain-agnostic anomaly detection. NorM is based on a new data series primitive, which permits to detect anomalies based on their (dis)similarity to a model that represents normal behavior. The experimental results on several real datasets demonstrate that the proposed approach outperforms by a large margin the current state-of-the art algorithms in terms of accuracy, while being orders of magnitude faster.

@article{udugama_novel_2020,
	title = {A novel use for an old problem: {The} {Tennessee} {Eastman} challenge process as an activating teaching tool},
	volume = {30},
	issn = {1749-7728},
	shorttitle = {A novel use for an old problem},
	url = {https://www.sciencedirect.com/science/article/pii/S1749772819300545},
	doi = {10.1016/j.ece.2019.09.002},
	abstract = {The domains of process design, operations and control are highly interdependent and thus affect operational efficiency and robustness of industrial facilities. Despite this, they are mostly kept in isolation from each other even in commercial projects, which is probably due to the perceived increase in complexity and aligned with prior experience from university education. Process design covers mostly the steady state, whereas process control is generally taught with a strong focus on classical control theory, which can be difficult to apply in industry. A reason for not combining the three domains of process design, operations and control is the lack of simple, student friendly teaching aids that can be used within time constraints of (under-)graduate teaching. The Tennessee Eastman Process (TEP) challenge, introduced in the early 90′s, originally as a process control benchmark simulation, can be used as a teaching aid to impart practical understanding of process design, dynamics and control to students. To this end, we will discuss the unique properties of the TEP, which make it an excellent tool to introduce process dynamics and control while reinforcing understanding of fundamentals, unit operations and the complexities and consequences of combining unit operations. We will then identify key developments that are needed to make the TEP a useful tool for teaching and discuss developments that have been carried out by the authors in this respect. Subsequently, it is shown how the modified model creates learning opportunities with respect to elementary as well as advanced control techniques and design assessments. The paper concludes with an outline of how this tool is currently used in a process design course with a strong emphasis on dynamics and control and in a classical undergraduate course on process control.},
	language = {en},
	urldate = {2022-05-17},
	journal = {Education for Chemical Engineers},
	author = {Udugama, Isuru A. and Gernaey, Krist V. and Taube, Michael A. and Bayer, Christoph},
	month = jan,
	year = {2020},
	keywords = {Control education, Operator training, Peaching software, Process design, Process dynamics},
	pages = {20--31},
}

The domains of process design, operations and control are highly interdependent and thus affect operational efficiency and robustness of industrial facilities. Despite this, they are mostly kept in isolation from each other even in commercial projects, which is probably due to the perceived increase in complexity and aligned with prior experience from university education. Process design covers mostly the steady state, whereas process control is generally taught with a strong focus on classical control theory, which can be difficult to apply in industry. A reason for not combining the three domains of process design, operations and control is the lack of simple, student friendly teaching aids that can be used within time constraints of (under-)graduate teaching. The Tennessee Eastman Process (TEP) challenge, introduced in the early 90′s, originally as a process control benchmark simulation, can be used as a teaching aid to impart practical understanding of process design, dynamics and control to students. To this end, we will discuss the unique properties of the TEP, which make it an excellent tool to introduce process dynamics and control while reinforcing understanding of fundamentals, unit operations and the complexities and consequences of combining unit operations. We will then identify key developments that are needed to make the TEP a useful tool for teaching and discuss developments that have been carried out by the authors in this respect. Subsequently, it is shown how the modified model creates learning opportunities with respect to elementary as well as advanced control techniques and design assessments. The paper concludes with an outline of how this tool is currently used in a process design course with a strong emphasis on dynamics and control and in a classical undergraduate course on process control.

@article{mariscotti_data_2020,
	title = {Data sets of measured pantograph voltage and current of {European} {AC} railways},
	volume = {30},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340920303711},
	doi = {10.1016/j.dib.2020.105477},
	abstract = {AC railways are characterized by peculiar Power Quality phenomena, where moving loads (trains, locomotives, etc.) interact with the supply network that provides electrical energy through the overhead contact line. Distortion, resonances, transients overlap in a complex dynamic scenario, that sees several and various problems of Power Quality, network stability, power and energy metering and disturbance to systems and equipment. For all related studies and analysis raw experimental data are extremely important. The provided data consists of time-domain waveforms of sampled pantograph voltage and current: each recording is tagged with the specific train operating condition (traction, cruising/coasting, braking, standstill), the active power and the speed, to support correlation and clustering of data.},
	language = {en},
	urldate = {2022-05-16},
	journal = {Data in Brief},
	author = {Mariscotti, Andrea},
	month = jun,
	year = {2020},
	keywords = {Harmonics, Measurement of electrical quantities, Power quality, Railways, Rolling stock},
	pages = {105477},
}

AC railways are characterized by peculiar Power Quality phenomena, where moving loads (trains, locomotives, etc.) interact with the supply network that provides electrical energy through the overhead contact line. Distortion, resonances, transients overlap in a complex dynamic scenario, that sees several and various problems of Power Quality, network stability, power and energy metering and disturbance to systems and equipment. For all related studies and analysis raw experimental data are extremely important. The provided data consists of time-domain waveforms of sampled pantograph voltage and current: each recording is tagged with the specific train operating condition (traction, cruising/coasting, braking, standstill), the active power and the speed, to support correlation and clustering of data.

@article{signorino_dataset_2020,
	title = {Dataset of measured and commented pantograph electric arcs in {DC} railways},
	volume = {31},
	issn = {2352-3409},
	url = {https://www.sciencedirect.com/science/article/pii/S2352340920308726},
	doi = {10.1016/j.dib.2020.105978},
	abstract = {DC railways are characterized by particularly intense arcing caused by pantograph detachment, due to the large current intensity and the general implementation of onboard resonant filters, whose transient response is triggered by electric transients including electric arcs. Electric arc depends on the train speed (the relative speed between the sliding contact over the pantograph and the hot spot on the catenary system), the intensity of the collected pantograph current and the line voltage level. Electric arcs are broadband in nature and can trigger the system transient response dominated by the resonant filter, besides interfering with the operation of onboard equipment (such as for energy conversion and metering).},
	language = {en},
	urldate = {2022-05-16},
	journal = {Data in Brief},
	author = {Signorino, Davide and Giordano, Domenico and Mariscotti, Andrea and Gallo, Daniele and Femine, Antonio Delle and Balic, Fabio and Quintana, Jorge and Donadio, Lorenzo and Biancucci, Alfredo},
	month = aug,
	year = {2020},
	keywords = {Electric arc, Measurement of electrical quantities, Pantograph, Power quality, Railways, Rolling stock},
	pages = {105978},
}

DC railways are characterized by particularly intense arcing caused by pantograph detachment, due to the large current intensity and the general implementation of onboard resonant filters, whose transient response is triggered by electric transients including electric arcs. Electric arc depends on the train speed (the relative speed between the sliding contact over the pantograph and the hot spot on the catenary system), the intensity of the collected pantograph current and the line voltage level. Electric arcs are broadband in nature and can trigger the system transient response dominated by the resonant filter, besides interfering with the operation of onboard equipment (such as for energy conversion and metering).

@article{yuan_general_2020,
	title = {A general end-to-end diagnosis framework for manufacturing systems},
	volume = {7},
	issn = {2095-5138},
	url = {https://doi.org/10.1093/nsr/nwz190},
	doi = {10.1093/nsr/nwz190},
	abstract = {The manufacturing sector is envisioned to be heavily influenced by artificial-intelligence-based technologies with the extraordinary increases in computational power and data volumes. A central challenge in the manufacturing sector lies in the requirement of a general framework to ensure satisfied diagnosis and monitoring performances in different manufacturing applications. Here, we propose a general data-driven, end-to-end framework for the monitoring of manufacturing systems. This framework, derived from deep-learning techniques, evaluates fused sensory measurements to detect and even predict faults and wearing conditions. This work exploits the predictive power of deep learning to automatically extract hidden degradation features from noisy, time-course data. We have experimented the proposed framework on 10 representative data sets drawn from a wide variety of manufacturing applications. Results reveal that the framework performs well in examined benchmark applications and can be applied in diverse contexts, indicating its potential use as a critical cornerstone in smart manufacturing.},
	number = {2},
	urldate = {2022-05-14},
	journal = {National Science Review},
	author = {Yuan, Ye and Ma, Guijun and Cheng, Cheng and Zhou, Beitong and Zhao, Huan and Zhang, Hai-Tao and Ding, Han},
	month = feb,
	year = {2020},
	pages = {418--429},
}

The manufacturing sector is envisioned to be heavily influenced by artificial-intelligence-based technologies with the extraordinary increases in computational power and data volumes. A central challenge in the manufacturing sector lies in the requirement of a general framework to ensure satisfied diagnosis and monitoring performances in different manufacturing applications. Here, we propose a general data-driven, end-to-end framework for the monitoring of manufacturing systems. This framework, derived from deep-learning techniques, evaluates fused sensory measurements to detect and even predict faults and wearing conditions. This work exploits the predictive power of deep learning to automatically extract hidden degradation features from noisy, time-course data. We have experimented the proposed framework on 10 representative data sets drawn from a wide variety of manufacturing applications. Results reveal that the framework performs well in examined benchmark applications and can be applied in diverse contexts, indicating its potential use as a critical cornerstone in smart manufacturing.

@article{fawwaz_real-time_2020,
	title = {Real-{Time} and {Robust} {Hydraulic} {System} {Fault} {Detection} via {Edge} {Computing}},
	volume = {10},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	issn = {2076-3417},
	url = {https://www.mdpi.com/2076-3417/10/17/5933},
	doi = {10.3390/app10175933},
	abstract = {We consider fault detection in a hydraulic system that maintains multivariate time-series sensor data. Such a real-world industrial environment could suffer from noisy data resulting from inaccuracies in hardware sensing or external interference. Thus, we propose a real-time and robust fault detection method for hydraulic systems that leverages cooperation between cloud and edge servers. The cloud server employs a new approach that includes a genetic algorithm (GA)-based feature selection that identifies feature-to-label correlations and feature-to-feature redundancies. A GA can efficiently process large search spaces, such as solving a combinatorial optimization problem to identify the optimal feature subset. By using fewer important features that require transmission and processing, this approach reduces detection time and improves model performance. We propose a long short-term memory autoencoder for a robust fault detection model that leverages temporal information on time-series sensor data and effectively handles noisy data. This detection model is then deployed at edge servers that provide computing resources near the data source to reduce latency. Our experimental results suggest that this method outperforms prior approaches by demonstrating lower detection times, higher accuracy, and increased robustness to noisy data. While we have a 63\% reduction of features, our model obtains a high accuracy of approximately 98\% and is robust to noisy data with a signal-to-noise ratio near 0 dB. Our method also performs at an average detection time of only 9.42 ms with a reduced average packet size of 179.98 KB from the maximum of 343.78 KB.},
	language = {en},
	number = {17},
	urldate = {2022-05-14},
	journal = {Applied Sciences},
	author = {Fawwaz, Dzaky Zakiyal and Chung, Sang-Hwa},
	month = jan,
	year = {2020},
	note = {Number: 17
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {autoencoder, edge computing, fault detection, feature selection, genetic algorithm, hydraulic system, long short-term memory},
	pages = {5933},
}

We consider fault detection in a hydraulic system that maintains multivariate time-series sensor data. Such a real-world industrial environment could suffer from noisy data resulting from inaccuracies in hardware sensing or external interference. Thus, we propose a real-time and robust fault detection method for hydraulic systems that leverages cooperation between cloud and edge servers. The cloud server employs a new approach that includes a genetic algorithm (GA)-based feature selection that identifies feature-to-label correlations and feature-to-feature redundancies. A GA can efficiently process large search spaces, such as solving a combinatorial optimization problem to identify the optimal feature subset. By using fewer important features that require transmission and processing, this approach reduces detection time and improves model performance. We propose a long short-term memory autoencoder for a robust fault detection model that leverages temporal information on time-series sensor data and effectively handles noisy data. This detection model is then deployed at edge servers that provide computing resources near the data source to reduce latency. Our experimental results suggest that this method outperforms prior approaches by demonstrating lower detection times, higher accuracy, and increased robustness to noisy data. While we have a 63% reduction of features, our model obtains a high accuracy of approximately 98% and is robust to noisy data with a signal-to-noise ratio near 0 dB. Our method also performs at an average detection time of only 9.42 ms with a reduced average packet size of 179.98 KB from the maximum of 343.78 KB.

@article{wang_deep_2020,
	title = {Deep learning for fault-relevant feature extraction and fault classification with stacked supervised auto-encoder},
	volume = {92},
	issn = {0959-1524},
	url = {https://www.sciencedirect.com/science/article/pii/S0959152420302225},
	doi = {10.1016/j.jprocont.2020.05.015},
	abstract = {Stacked auto-encoder (SAE)-based deep learning has been introduced for fault classification in recent years, which has the potential to extract deep abstract features from the raw input data. However, SAE cannot ensure the relevance of deep features with the fault types due to its unsupervised self-reconstruction in the pretraining stage. To overcome this problem, a stacked supervised auto-encoder is proposed to pretrain the deep network and obtain deep fault-relevant features from raw input data. In each supervised auto-encoder, informative features are learned from the input data with the goal that they can largely distinguish different fault types. By stacking multiple supervised auto-encoders hierarchically, high-level fault-relevant features are gradually learned from raw input data, which can improve the classification accuracy of the classifiers. The proposed SSAE is tested on the Tennessee–Eastman (TE) benchmark process and a real industrial hydrocracking process. The results show the effectiveness and flexibility of SSAE.},
	language = {en},
	urldate = {2022-05-02},
	journal = {Journal of Process Control},
	author = {Wang, Yalin and Yang, Haibing and Yuan, Xiaofeng and Shardt, Yuri A. W. and Yang, Chunhua and Gui, Weihua},
	month = aug,
	year = {2020},
	keywords = {Deep learning, Fault classification, Process monitoring, Stacked auto-encoder (SAE), Tennessee–Eastman process},
	pages = {79--89},
}

Stacked auto-encoder (SAE)-based deep learning has been introduced for fault classification in recent years, which has the potential to extract deep abstract features from the raw input data. However, SAE cannot ensure the relevance of deep features with the fault types due to its unsupervised self-reconstruction in the pretraining stage. To overcome this problem, a stacked supervised auto-encoder is proposed to pretrain the deep network and obtain deep fault-relevant features from raw input data. In each supervised auto-encoder, informative features are learned from the input data with the goal that they can largely distinguish different fault types. By stacking multiple supervised auto-encoders hierarchically, high-level fault-relevant features are gradually learned from raw input data, which can improve the classification accuracy of the classifiers. The proposed SSAE is tested on the Tennessee–Eastman (TE) benchmark process and a real industrial hydrocracking process. The results show the effectiveness and flexibility of SSAE.

@article{zheng_new_2020,
	title = {A new unsupervised data mining method based on the stacked autoencoder for chemical process fault diagnosis},
	volume = {135},
	issn = {0098-1354},
	url = {https://www.sciencedirect.com/science/article/pii/S009813541930986X},
	doi = {10.1016/j.compchemeng.2020.106755},
	abstract = {Process monitoring plays an important role in chemical process safety management, and fault diagnosis is a vital step of process monitoring. Among fault diagnosis researches, supervised ones are inappropriate for industrial applications due to the lack of labeled historical data in real situations. Thereby, unsupervised methods which are capable of dealing with unlabeled data should be developed for fault diagnosis. In this work, a new unsupervised data mining method based on deep learning is proposed for isolating different conditions of chemical process, including normal operations and faults, and thus labeled database can be created efficiently for constructing fault diagnosis model. The proposed method mainly consists of three steps: feature extraction by the convolutional stacked autoencoder (SAE), feature visualization by the t-distributed stochastic neighbor embedding (t-SNE) algorithm, and clustering. The benchmark Tennessee Eastman process (TEP) and an industrial hydrocracking instance are utilized to illustrate the effectiveness of the proposed data mining method.},
	language = {en},
	urldate = {2022-05-02},
	journal = {Computers \& Chemical Engineering},
	author = {Zheng, Shaodong and Zhao, Jinsong},
	month = apr,
	year = {2020},
	keywords = {Clustering, Data mining, Fault diagnosis, The SAE, The TEP, Unsupervised},
	pages = {106755},
}

Process monitoring plays an important role in chemical process safety management, and fault diagnosis is a vital step of process monitoring. Among fault diagnosis researches, supervised ones are inappropriate for industrial applications due to the lack of labeled historical data in real situations. Thereby, unsupervised methods which are capable of dealing with unlabeled data should be developed for fault diagnosis. In this work, a new unsupervised data mining method based on deep learning is proposed for isolating different conditions of chemical process, including normal operations and faults, and thus labeled database can be created efficiently for constructing fault diagnosis model. The proposed method mainly consists of three steps: feature extraction by the convolutional stacked autoencoder (SAE), feature visualization by the t-distributed stochastic neighbor embedding (t-SNE) algorithm, and clustering. The benchmark Tennessee Eastman process (TEP) and an industrial hydrocracking instance are utilized to illustrate the effectiveness of the proposed data mining method.

@article{zhang_fault_2020,
	title = {Fault {Detection} in the {Tennessee} {Eastman} {Benchmark} {Process} {Using} {Principal} {Component} {Difference} {Based} on {K}-{Nearest} {Neighbors}},
	volume = {8},
	issn = {2169-3536},
	doi = {10.1109/ACCESS.2020.2977421},
	abstract = {Industrial data usually have nonlinear or multimodal characteristics which do not meet the data assumptions of statistics in principal component analysis (PCA). Therefore, PCA has a lower fault detection rate in industrial processes. Aiming at the above limitations of PCA, a fault detection method using principal component difference based on k-nearest neighbors (Diff-PCA) is proposed in this paper. First, find the k nearest neighbors set of each sample in the training data set and calculate its mean vector. Second, build an augmented vector using each sample and its corresponding mean vector. Third, calculate the loading matrix and score matrix using PCA. Next, calculate the estimated scores using the mean vector of each sample and missing data imputation technique for PCA. At last, build two new statistics using the difference between the real scores and estimated scores to detect faults. In addition, the fault diagnosis method based on contribution plots of monitored variables is also proposed in this paper. In Diff-PCA, the difference skill can eliminate the impact of the nonlinear and multimodal structure on fault detection. Meanwhile, the monitored subspaces by the two new statistics are different from that by T2 and SPE in PCA. The efficiency of the proposed strategy is implemented in two numerical cases (nonlinear and multimode) and the Tennessee Eastman (TE) processes. The fault detection results indicate that Diff-PCA outperforms the conventional PCA, Kernel PCA, dynamic PCA, principal component-based k nearest neighbor rule and k nearest neighbor rule.},
	journal = {IEEE Access},
	author = {Zhang, Cheng and Guo, Qingxiu and Li, Yuan},
	year = {2020},
	note = {Conference Name: IEEE Access},
	keywords = {Covariance matrices, Fault detection, Fault detection and diagnosis, Fault diagnosis, Loading, Monitoring, Principal component analysis, Tennessee Eastman processes, Training data, k nearest neighbors, principal component analysis, principal component difference},
	pages = {49999--50009},
}

Industrial data usually have nonlinear or multimodal characteristics which do not meet the data assumptions of statistics in principal component analysis (PCA). Therefore, PCA has a lower fault detection rate in industrial processes. Aiming at the above limitations of PCA, a fault detection method using principal component difference based on k-nearest neighbors (Diff-PCA) is proposed in this paper. First, find the k nearest neighbors set of each sample in the training data set and calculate its mean vector. Second, build an augmented vector using each sample and its corresponding mean vector. Third, calculate the loading matrix and score matrix using PCA. Next, calculate the estimated scores using the mean vector of each sample and missing data imputation technique for PCA. At last, build two new statistics using the difference between the real scores and estimated scores to detect faults. In addition, the fault diagnosis method based on contribution plots of monitored variables is also proposed in this paper. In Diff-PCA, the difference skill can eliminate the impact of the nonlinear and multimodal structure on fault detection. Meanwhile, the monitored subspaces by the two new statistics are different from that by T2 and SPE in PCA. The efficiency of the proposed strategy is implemented in two numerical cases (nonlinear and multimode) and the Tennessee Eastman (TE) processes. The fault detection results indicate that Diff-PCA outperforms the conventional PCA, Kernel PCA, dynamic PCA, principal component-based k nearest neighbor rule and k nearest neighbor rule.

@article{van_engelen_survey_2020,
	title = {A survey on semi-supervised learning},
	volume = {109},
	issn = {1573-0565},
	url = {https://doi.org/10.1007/s10994-019-05855-6},
	doi = {10.1007/s10994-019-05855-6},
	abstract = {Semi-supervised learning is the branch of machine learning concerned with using labelled as well as unlabelled data to perform certain learning tasks. Conceptually situated between supervised and unsupervised learning, it permits harnessing the large amounts of unlabelled data available in many use cases in combination with typically smaller sets of labelled data. In recent years, research in this area has followed the general trends observed in machine learning, with much attention directed at neural network-based models and generative learning. The literature on the topic has also expanded in volume and scope, now encompassing a broad spectrum of theory, algorithms and applications. However, no recent surveys exist to collect and organize this knowledge, impeding the ability of researchers and engineers alike to utilize it. Filling this void, we present an up-to-date overview of semi-supervised learning methods, covering earlier work as well as more recent advances. We focus primarily on semi-supervised classification, where the large majority of semi-supervised learning research takes place. Our survey aims to provide researchers and practitioners new to the field as well as more advanced readers with a solid understanding of the main approaches and algorithms developed over the past two decades, with an emphasis on the most prominent and currently relevant work. Furthermore, we propose a new taxonomy of semi-supervised classification algorithms, which sheds light on the different conceptual and methodological approaches for incorporating unlabelled data into the training process. Lastly, we show how the fundamental assumptions underlying most semi-supervised learning algorithms are closely connected to each other, and how they relate to the well-known semi-supervised clustering assumption.},
	language = {en},
	number = {2},
	urldate = {2022-04-25},
	journal = {Machine Learning},
	author = {van Engelen, Jesper E. and Hoos, Holger H.},
	month = feb,
	year = {2020},
	keywords = {Classification, Machine learning, Semi-supervised learning},
	pages = {373--440},
}

Semi-supervised learning is the branch of machine learning concerned with using labelled as well as unlabelled data to perform certain learning tasks. Conceptually situated between supervised and unsupervised learning, it permits harnessing the large amounts of unlabelled data available in many use cases in combination with typically smaller sets of labelled data. In recent years, research in this area has followed the general trends observed in machine learning, with much attention directed at neural network-based models and generative learning. The literature on the topic has also expanded in volume and scope, now encompassing a broad spectrum of theory, algorithms and applications. However, no recent surveys exist to collect and organize this knowledge, impeding the ability of researchers and engineers alike to utilize it. Filling this void, we present an up-to-date overview of semi-supervised learning methods, covering earlier work as well as more recent advances. We focus primarily on semi-supervised classification, where the large majority of semi-supervised learning research takes place. Our survey aims to provide researchers and practitioners new to the field as well as more advanced readers with a solid understanding of the main approaches and algorithms developed over the past two decades, with an emphasis on the most prominent and currently relevant work. Furthermore, we propose a new taxonomy of semi-supervised classification algorithms, which sheds light on the different conceptual and methodological approaches for incorporating unlabelled data into the training process. Lastly, we show how the fundamental assumptions underlying most semi-supervised learning algorithms are closely connected to each other, and how they relate to the well-known semi-supervised clustering assumption.

@article{wang_generalizing_2020,
	title = {Generalizing from a {Few} {Examples}: {A} {Survey} on {Few}-shot {Learning}},
	volume = {53},
	issn = {0360-0300},
	shorttitle = {Generalizing from a {Few} {Examples}},
	url = {https://doi.org/10.1145/3386252},
	doi = {10.1145/3386252},
	abstract = {Machine learning has been highly successful in data-intensive applications but is often hampered when the data set is small. Recently, Few-shot Learning (FSL) is proposed to tackle this problem. Using prior knowledge, FSL can rapidly generalize to new tasks containing only a few samples with supervised information. In this article, we conduct a thorough survey to fully understand FSL. Starting from a formal definition of FSL, we distinguish FSL from several relevant machine learning problems. We then point out that the core issue in FSL is that the empirical risk minimizer is unreliable. Based on how prior knowledge can be used to handle this core issue, we categorize FSL methods from three perspectives: (i) data, which uses prior knowledge to augment the supervised experience; (ii) model, which uses prior knowledge to reduce the size of the hypothesis space; and (iii) algorithm, which uses prior knowledge to alter the search for the best hypothesis in the given hypothesis space. With this taxonomy, we review and discuss the pros and cons of each category. Promising directions, in the aspects of the FSL problem setups, techniques, applications, and theories, are also proposed to provide insights for future research.1},
	number = {3},
	urldate = {2022-04-25},
	journal = {ACM Computing Surveys},
	author = {Wang, Yaqing and Yao, Quanming and Kwok, James T. and Ni, Lionel M.},
	month = jun,
	year = {2020},
	keywords = {Few-shot learning, low-shot learning, meta-learning, one-shot learning, prior knowledge, small sample learning},
	pages = {63:1--63:34},
}

Machine learning has been highly successful in data-intensive applications but is often hampered when the data set is small. Recently, Few-shot Learning (FSL) is proposed to tackle this problem. Using prior knowledge, FSL can rapidly generalize to new tasks containing only a few samples with supervised information. In this article, we conduct a thorough survey to fully understand FSL. Starting from a formal definition of FSL, we distinguish FSL from several relevant machine learning problems. We then point out that the core issue in FSL is that the empirical risk minimizer is unreliable. Based on how prior knowledge can be used to handle this core issue, we categorize FSL methods from three perspectives: (i) data, which uses prior knowledge to augment the supervised experience; (ii) model, which uses prior knowledge to reduce the size of the hypothesis space; and (iii) algorithm, which uses prior knowledge to alter the search for the best hypothesis in the given hypothesis space. With this taxonomy, we review and discuss the pros and cons of each category. Promising directions, in the aspects of the FSL problem setups, techniques, applications, and theories, are also proposed to provide insights for future research.1

@article{yu_continuous_2020,
	title = {Continuous {Support} {Vector} {Regression} for {Nonstationary} {Streaming} {Data}},
	issn = {2168-2275},
	doi = {10.1109/TCYB.2020.3015266},
	abstract = {Quadratic programming is the process of solving a special type of mathematical optimization problem. Recent advances in online solutions for quadratic programming problems (QPPs) have created opportunities to widen the scope of applications for support vector regression (SVR). In this vein, efforts to make SVR compatible with streaming data have been met with substantial success. However, streaming data with concept drift remain problematic because the trained prediction function in SVR tends to drift as the data distribution drifts. Aiming to contribute a solution to this aspect of SVR's advancement, we have developed continuous SVR (C-SVR) to solve regression problems with nonstationary streaming data, that is, data where the optimal input-output prediction function can drift over time. The basic idea of C-SVR is to continuously learn a series of input-output functions over a series of time windows to make predictions about different periods. However, strikingly, the learning process in different time windows is not independent. An additional similarity term in the QPP, which is solved incrementally, threads the various input-output functions together by conveying some learned knowledge through consecutive time windows. How much learned knowledge is transferred is determined by the extent of the concept drift. Experimental evaluations with both synthetic and real-world datasets indicate that C-SVR has better performance than most existing methods for nonstationary streaming data regression.},
	journal = {IEEE Transactions on Cybernetics},
	author = {Yu, Hang and Lu, Jie and Zhang, Guangquan},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Cybernetics},
	keywords = {Concept drift, Cybernetics, Microsoft Windows, Prediction algorithms, Quadratic programming, Support vector machines, Training, Vegetation, continuous learning, streaming data, support vector regression (SVR)},
	pages = {1--14},
}

Quadratic programming is the process of solving a special type of mathematical optimization problem. Recent advances in online solutions for quadratic programming problems (QPPs) have created opportunities to widen the scope of applications for support vector regression (SVR). In this vein, efforts to make SVR compatible with streaming data have been met with substantial success. However, streaming data with concept drift remain problematic because the trained prediction function in SVR tends to drift as the data distribution drifts. Aiming to contribute a solution to this aspect of SVR's advancement, we have developed continuous SVR (C-SVR) to solve regression problems with nonstationary streaming data, that is, data where the optimal input-output prediction function can drift over time. The basic idea of C-SVR is to continuously learn a series of input-output functions over a series of time windows to make predictions about different periods. However, strikingly, the learning process in different time windows is not independent. An additional similarity term in the QPP, which is solved incrementally, threads the various input-output functions together by conveying some learned knowledge through consecutive time windows. How much learned knowledge is transferred is determined by the extent of the concept drift. Experimental evaluations with both synthetic and real-world datasets indicate that C-SVR has better performance than most existing methods for nonstationary streaming data regression.

@article{mcinnes_umap_2020,
	title = {{UMAP}: {Uniform} {Manifold} {Approximation} and {Projection} for {Dimension} {Reduction}},
	shorttitle = {{UMAP}},
	url = {http://arxiv.org/abs/1802.03426},
	abstract = {UMAP (Uniform Manifold Approximation and Projection) is a novel manifold learning technique for dimension reduction. UMAP is constructed from a theoretical framework based in Riemannian geometry and algebraic topology. The result is a practical scalable algorithm that applies to real world data. The UMAP algorithm is competitive with t-SNE for visualization quality, and arguably preserves more of the global structure with superior run time performance. Furthermore, UMAP has no computational restrictions on embedding dimension, making it viable as a general purpose dimension reduction technique for machine learning.},
	urldate = {2022-03-15},
	journal = {arXiv:1802.03426 [cs, stat]},
	author = {McInnes, Leland and Healy, John and Melville, James},
	month = sep,
	year = {2020},
	note = {arXiv: 1802.03426},
	keywords = {Computer Science - Computational Geometry, Computer Science - Machine Learning, Statistics - Machine Learning},
}

UMAP (Uniform Manifold Approximation and Projection) is a novel manifold learning technique for dimension reduction. UMAP is constructed from a theoretical framework based in Riemannian geometry and algebraic topology. The result is a practical scalable algorithm that applies to real world data. The UMAP algorithm is competitive with t-SNE for visualization quality, and arguably preserves more of the global structure with superior run time performance. Furthermore, UMAP has no computational restrictions on embedding dimension, making it viable as a general purpose dimension reduction technique for machine learning.

@inproceedings{hoong_ong_predictive_2020,
	title = {Predictive {Maintenance} for {Edge}-{Based} {Sensor} {Networks}: {A} {Deep} {Reinforcement} {Learning} {Approach}},
	shorttitle = {Predictive {Maintenance} for {Edge}-{Based} {Sensor} {Networks}},
	doi = {10.1109/WF-IoT48130.2020.9221098},
	abstract = {Failure of mission-critical equipment interrupts production and results in monetary loss. The risk of unplanned equipment downtime can be minimized through Predictive Maintenance of revenue generating assets to ensure optimal performance and safe operation of equipment. However, the increased sensorization of the equipment generates a data deluge, and existing machine-learning based predictive model alone becomes inadequate for timely equipment condition predictions. In this paper, a model-free Deep Reinforcement Learning algorithm is proposed for predictive equipment maintenance from an equipment-based sensor network context. Within each equipment, a sensor device aggregates raw sensor data, and the equipment health status is analyzed for anomalous events. Unlike traditional black-box regression models, the proposed algorithm self-learns an optimal maintenance policy and provides actionable recommendation for each equipment. Our experimental results demonstrate the potential for broader range of equipment maintenance applications as an automatic learning framework.},
	booktitle = {2020 {IEEE} 6th {World} {Forum} on {Internet} of {Things} ({WF}-{IoT})},
	author = {Hoong Ong, Kevin Shen and Niyato, Dusit and Yuen, Chau},
	month = jun,
	year = {2020},
	keywords = {Industries, Mission critical systems, Prediction algorithms, Predictive models, Productivity, Reinforcement learning, Schedules},
	pages = {1--6},
}

Failure of mission-critical equipment interrupts production and results in monetary loss. The risk of unplanned equipment downtime can be minimized through Predictive Maintenance of revenue generating assets to ensure optimal performance and safe operation of equipment. However, the increased sensorization of the equipment generates a data deluge, and existing machine-learning based predictive model alone becomes inadequate for timely equipment condition predictions. In this paper, a model-free Deep Reinforcement Learning algorithm is proposed for predictive equipment maintenance from an equipment-based sensor network context. Within each equipment, a sensor device aggregates raw sensor data, and the equipment health status is analyzed for anomalous events. Unlike traditional black-box regression models, the proposed algorithm self-learns an optimal maintenance policy and provides actionable recommendation for each equipment. Our experimental results demonstrate the potential for broader range of equipment maintenance applications as an automatic learning framework.

@article{skordilis_deep_2020,
	title = {A deep reinforcement learning approach for real-time sensor-driven decision making and predictive analytics},
	volume = {147},
	issn = {0360-8352},
	url = {https://www.sciencedirect.com/science/article/pii/S036083522030334X},
	doi = {10.1016/j.cie.2020.106600},
	abstract = {The increased complexity of sensor-intensive systems with expensive subsystems and costly repairs and failures calls for efficient real-time control and decision making policies. Deep reinforcement learning has demonstrated great potential in addressing highly complex and challenging control and decision making problems. Despite its potential to derive real-time policies using real-time data for dynamic systems, it has been rarely used for sensor-driven maintenance related problems. In this paper, we propose two novel decision making methods in which reinforcement learning and particle filtering are utilized for (i) deriving real-time maintenance policies and (ii) estimating remaining useful life for sensor-monitored degrading systems. The proposed framework introduces a new direction with many potential opportunities for system monitoring. To demonstrate the effectiveness of the proposed methods, numerical experiments are provided from a set of simulated data and a turbofan engine dataset provided by NASA.},
	language = {en},
	urldate = {2022-03-03},
	journal = {Computers \& Industrial Engineering},
	author = {Skordilis, Erotokritos and Moghaddass, Ramin},
	month = sep,
	year = {2020},
	keywords = {Decision-making, Deep reinforcement learning, Particle filters, Real-time control, Remaining useful life estimation},
	pages = {106600},
}

The increased complexity of sensor-intensive systems with expensive subsystems and costly repairs and failures calls for efficient real-time control and decision making policies. Deep reinforcement learning has demonstrated great potential in addressing highly complex and challenging control and decision making problems. Despite its potential to derive real-time policies using real-time data for dynamic systems, it has been rarely used for sensor-driven maintenance related problems. In this paper, we propose two novel decision making methods in which reinforcement learning and particle filtering are utilized for (i) deriving real-time maintenance policies and (ii) estimating remaining useful life for sensor-monitored degrading systems. The proposed framework introduces a new direction with many potential opportunities for system monitoring. To demonstrate the effectiveness of the proposed methods, numerical experiments are provided from a set of simulated data and a turbofan engine dataset provided by NASA.

@article{sheikh_battery_2020,
	title = {A {Battery} {Health} {Monitoring} {Method} {Using} {Machine} {Learning}: {A} {Data}-{Driven} {Approach}},
	volume = {13},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	issn = {1996-1073},
	shorttitle = {A {Battery} {Health} {Monitoring} {Method} {Using} {Machine} {Learning}},
	url = {https://www.mdpi.com/1996-1073/13/14/3658},
	doi = {10.3390/en13143658},
	abstract = {Batteries are combinations of electrochemical cells that generate electricity to power electrical devices. Batteries are continuously converting chemical energy to electrical energy, and require appropriate maintenance to provide maximum efficiency. Management systems having specialized monitoring features; such as charge controlling mechanisms and temperature regulation are used to prevent health, safety, and property hazards that complement the use of batteries. These systems utilize measures of merit to regulate battery performances. Figures such as the state-of-health (SOH) and state-of-charge (SOC) are used to estimate the performance and state of the battery. In this paper, we propose an intelligent method to investigate the aforementioned parameters using a data-driven approach. We use a machine learning algorithm that extracts significant features from the discharge curves to estimate these parameters. Extensive simulations have been carried out to evaluate the performance of the proposed method under different currents and temperatures.},
	language = {en},
	number = {14},
	urldate = {2022-02-09},
	journal = {Energies},
	author = {Sheikh, Shehzar Shahzad and Anjum, Mahnoor and Khan, Muhammad Abdullah and Hassan, Syed Ali and Khalid, Hassan Abdullah and Gastli, Adel and Ben-Brahim, Lazhar},
	month = jan,
	year = {2020},
	note = {Number: 14
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {battery health monitoring, ecml, feature extraction, knee-point calculation, machine learning, state of health},
	pages = {3658},
}

Batteries are combinations of electrochemical cells that generate electricity to power electrical devices. Batteries are continuously converting chemical energy to electrical energy, and require appropriate maintenance to provide maximum efficiency. Management systems having specialized monitoring features; such as charge controlling mechanisms and temperature regulation are used to prevent health, safety, and property hazards that complement the use of batteries. These systems utilize measures of merit to regulate battery performances. Figures such as the state-of-health (SOH) and state-of-charge (SOC) are used to estimate the performance and state of the battery. In this paper, we propose an intelligent method to investigate the aforementioned parameters using a data-driven approach. We use a machine learning algorithm that extracts significant features from the discharge curves to estimate these parameters. Extensive simulations have been carried out to evaluate the performance of the proposed method under different currents and temperatures.

@inproceedings{farbiz_cognitive_2020,
	title = {A {Cognitive} {Analytics} based {Approach} for {Machine} {Health} {Monitoring}, {Anomaly} {Detection}, and {Predictive} {Maintenance}},
	doi = {10.1109/ICIEA48937.2020.9248409},
	abstract = {Traditionally, there are two major limitations for machine learning (ML)-assisted manufacturing applications. First, it would require a tremendous amount of manual data annotations for ML models. Second, ML models are often learned offline and unable to capture the machine dynamism and adapt to changes over the time. In this paper, we propose a framework based on the concept of cognitive analytics with unsupervised learning for machine health monitoring, anomaly detection and predictive maintenance. The experimental results on an industrial robot demonstrates the effectiveness of the proposed framework in the identified use case.},
	booktitle = {2020 15th {IEEE} {Conference} on {Industrial} {Electronics} and {Applications} ({ICIEA})},
	author = {Farbiz, Farzam and Miaolong, Yuan and Yu, Zhou},
	month = nov,
	year = {2020},
	note = {ISSN: 2158-2297},
	keywords = {Adaptation models, Anomaly detection, Machine learning, Manufacturing, Monitoring, Predictive maintenance, Service robots, Unsupervised learning, anomaly detection, cognitive analytics, ecml, machine health monitoring, predictive maintenance},
	pages = {1104--1109},
}

Traditionally, there are two major limitations for machine learning (ML)-assisted manufacturing applications. First, it would require a tremendous amount of manual data annotations for ML models. Second, ML models are often learned offline and unable to capture the machine dynamism and adapt to changes over the time. In this paper, we propose a framework based on the concept of cognitive analytics with unsupervised learning for machine health monitoring, anomaly detection and predictive maintenance. The experimental results on an industrial robot demonstrates the effectiveness of the proposed framework in the identified use case.

@article{dai_machinery_2020,
	title = {Machinery {Health} {Monitoring} {Based} on {Unsupervised} {Feature} {Learning} via {Generative} {Adversarial} {Networks}},
	volume = {25},
	issn = {1941-014X},
	doi = {10.1109/TMECH.2020.3012179},
	abstract = {It confronts great difficulty to apply traditional artificial intelligence (AI) techniques to machinery prognostics and health management in manufacturing systems due to the lack of abnormal samples corresponding to different fault conditions. This article explores an unsupervised feature learning method for machinery health monitoring by proposing a generative adversarial networks (GAN) model that exploits the merits of the autoencoder and the traditional GAN. The major contribution is that the data distribution of the normal samples is accurately learned by the GAN model within both the signal spectrum and latent representation spaces. Specifically, the discriminative feature for machinery health monitoring is learned in an unsupervised manner by the proposed method in three steps. First, the proposed GAN model is trained by the normal samples of the inspected machine with the aim to correctly reconstruct the signal spectrum and its latent representation. Then, the trained model is applied to test the online samples of the same machine with unknown health conditions. Finally, the dissimilarity between the tested samples and their reconstructed ones in the latent representation space is taken as the discriminative feature. The feature value will increase significantly if a fault occurs in the inspected machine because the abnormal samples are never trained in the proposed GAN model. Experimental studies on three different machines are conducted to validate the proposed method and its superiority over the traditional methods in detecting abnormal points and characterizing fault propagation.},
	number = {5},
	journal = {IEEE/ASME Transactions on Mechatronics},
	author = {Dai, Jun and Wang, Jun and Huang, Weiguo and Shi, Juanjuan and Zhu, Zhongkui},
	month = oct,
	year = {2020},
	note = {Conference Name: IEEE/ASME Transactions on Mechatronics},
	keywords = {Artificial intelligence (AI), Feature extraction, Gallium nitride, Generative adversarial networks, Generators, Monitoring, Training, deep learning, ecml, generative adversarial networks (GAN), machinery health monitoring, smart manufacturing, unsupervised learning},
	pages = {2252--2263},
}

It confronts great difficulty to apply traditional artificial intelligence (AI) techniques to machinery prognostics and health management in manufacturing systems due to the lack of abnormal samples corresponding to different fault conditions. This article explores an unsupervised feature learning method for machinery health monitoring by proposing a generative adversarial networks (GAN) model that exploits the merits of the autoencoder and the traditional GAN. The major contribution is that the data distribution of the normal samples is accurately learned by the GAN model within both the signal spectrum and latent representation spaces. Specifically, the discriminative feature for machinery health monitoring is learned in an unsupervised manner by the proposed method in three steps. First, the proposed GAN model is trained by the normal samples of the inspected machine with the aim to correctly reconstruct the signal spectrum and its latent representation. Then, the trained model is applied to test the online samples of the same machine with unknown health conditions. Finally, the dissimilarity between the tested samples and their reconstructed ones in the latent representation space is taken as the discriminative feature. The feature value will increase significantly if a fault occurs in the inspected machine because the abnormal samples are never trained in the proposed GAN model. Experimental studies on three different machines are conducted to validate the proposed method and its superiority over the traditional methods in detecting abnormal points and characterizing fault propagation.

@article{yu_multiscale_2020,
	title = {Multiscale intelligent fault detection system based on agglomerative hierarchical clustering using stacked denoising autoencoder with temporal information},
	volume = {95},
	issn = {1568-4946},
	url = {https://www.sciencedirect.com/science/article/pii/S1568494620304646},
	doi = {10.1016/j.asoc.2020.106525},
	abstract = {Deep learning-based process monitoring has achieved remarkable progress. Generally, a deep model is empirically selected before the data features are learned. In this study, the interpretability and suitability of stacked denoising autoencoder (SDAE) in process monitoring territory are theoretically analyzed and validated. Considering that the data will show different feature representations at different scales, such as overall outline, local information, and microscopic details, this study utilizes the concept of multiscale analysis to mine the feature information of raw data deeply in different scales. The multiscale analysis is performed on the basis of agglomerative hierarchical clustering and silhouette coefficient, which makes the analysis data characteristics-based and intelligently abandons the intervention of manual prior knowledge. Then, the SDAE models are established under each scale to learn the high-order and robust features from the data with noise and fluctuation, and all monitoring results of the different scales are integrated using the Bayesian inference. Finally, given the temporal information in sequence data, the state representation of previous events is embedded into the current decision through a sliding window. The numerical process, benchmark Tennessee Eastman and real steel plate process are used to analyze the superiority of the proposed method (MSDAE-TP) over other deep learning-based monitoring methods.},
	language = {en},
	urldate = {2022-01-14},
	journal = {Applied Soft Computing},
	author = {Yu, Jianbo and Yan, Xuefeng},
	month = oct,
	year = {2020},
	keywords = {Multiscale analysis, Process monitoring, Robust features, Stacked denoising autoencoder, Temporal information},
	pages = {106525},
}

Deep learning-based process monitoring has achieved remarkable progress. Generally, a deep model is empirically selected before the data features are learned. In this study, the interpretability and suitability of stacked denoising autoencoder (SDAE) in process monitoring territory are theoretically analyzed and validated. Considering that the data will show different feature representations at different scales, such as overall outline, local information, and microscopic details, this study utilizes the concept of multiscale analysis to mine the feature information of raw data deeply in different scales. The multiscale analysis is performed on the basis of agglomerative hierarchical clustering and silhouette coefficient, which makes the analysis data characteristics-based and intelligently abandons the intervention of manual prior knowledge. Then, the SDAE models are established under each scale to learn the high-order and robust features from the data with noise and fluctuation, and all monitoring results of the different scales are integrated using the Bayesian inference. Finally, given the temporal information in sequence data, the state representation of previous events is embedded into the current decision through a sliding window. The numerical process, benchmark Tennessee Eastman and real steel plate process are used to analyze the superiority of the proposed method (MSDAE-TP) over other deep learning-based monitoring methods.

@article{aydemir_anomaly_2020,
	title = {Anomaly monitoring improves remaining useful life estimation of industrial machinery},
	volume = {56},
	issn = {0278-6125},
	url = {https://www.sciencedirect.com/science/article/pii/S0278612520301060},
	doi = {10.1016/j.jmsy.2020.06.014},
	abstract = {Estimating remaining useful life (RUL) of industrial machinery based on their degradation data is very critical for various industries. Machine learning models are powerful and very popular tools for predicting time to failure of such industrial machinery. However, RUL is ill-defined during healthy operation. This paper proposes to use anomaly monitoring during both RUL estimator training and deployment to tackle with this problem. In this approach, raw sensor data is monitored and when a statistically significant change is detected, it is taken as the degradation onset point and a data-driven RUL estimation model is triggered. Initial results with a simple anomaly detector, suited for non-varying operating conditions, and multiple RUL estimation models showed that the anomaly triggered RUL estimation scheme enhances the estimation accuracy, on in-house simulation and benchmark C-MAPSS turbofan engine degradation data. The scheme can be employed to varying operating conditions with a suitable anomaly detector.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Journal of Manufacturing Systems},
	author = {Aydemir, Gurkan and Acar, Burak},
	month = jul,
	year = {2020},
	keywords = {Anomaly detection, Industrial prognostics and health management, Machine learning, Remaining Useful Life (RUL) estimation, sigkdd-rw},
	pages = {463--469},
}

Estimating remaining useful life (RUL) of industrial machinery based on their degradation data is very critical for various industries. Machine learning models are powerful and very popular tools for predicting time to failure of such industrial machinery. However, RUL is ill-defined during healthy operation. This paper proposes to use anomaly monitoring during both RUL estimator training and deployment to tackle with this problem. In this approach, raw sensor data is monitored and when a statistically significant change is detected, it is taken as the degradation onset point and a data-driven RUL estimation model is triggered. Initial results with a simple anomaly detector, suited for non-varying operating conditions, and multiple RUL estimation models showed that the anomaly triggered RUL estimation scheme enhances the estimation accuracy, on in-house simulation and benchmark C-MAPSS turbofan engine degradation data. The scheme can be employed to varying operating conditions with a suitable anomaly detector.

@article{scutari_bayesian_2020,
	title = {Bayesian network models for incomplete and dynamic data},
	volume = {74},
	issn = {1467-9574},
	url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/stan.12197},
	doi = {10.1111/stan.12197},
	abstract = {Bayesian networks are a versatile and powerful tool to model complex phenomena and the interplay of their components in a probabilistically principled way. Moving beyond the comparatively simple case of completely observed, static data, which has received the most attention in the literature, in this paper, we will review how Bayesian networks can model dynamic data and data with incomplete observations. Such data are the norm at the forefront of research and in practical applications, and Bayesian networks are uniquely positioned to model them due to their explainability and interpretability.},
	language = {en},
	number = {3},
	urldate = {2021-11-27},
	journal = {Statistica Neerlandica},
	author = {Scutari, Marco},
	year = {2020},
	note = {\_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/stan.12197},
	keywords = {Bayesian networks, dynamic data, incomplete data, inference, structure learning},
	pages = {397--419},
}

Bayesian networks are a versatile and powerful tool to model complex phenomena and the interplay of their components in a probabilistically principled way. Moving beyond the comparatively simple case of completely observed, static data, which has received the most attention in the literature, in this paper, we will review how Bayesian networks can model dynamic data and data with incomplete observations. Such data are the norm at the forefront of research and in practical applications, and Bayesian networks are uniquely positioned to model them due to their explainability and interpretability.

@article{huang_bayesian_2020,
	title = {A {Bayesian} network model to predict the effects of interruptions on train operations},
	volume = {114},
	issn = {0968-090X},
	url = {https://www.sciencedirect.com/science/article/pii/S0968090X19311118},
	doi = {10.1016/j.trc.2020.02.021},
	abstract = {Based on the Bayesian network (BN) paradigm, we propose a hybrid model to predict the three main consequences of disruptions and disturbances during train operations, namely, the primary delay (L), the number of affected trains (N), and the total delay times (T). To obtain an effective BN structure, we first analyze the dependencies of the involved factors on each station and among adjacent stations, given domain knowledge and expertise about operational characteristics. We then put forward four candidate BN structures, integrating expert knowledge, the interdependencies learned from real-world data, and real-time prediction and operational requirements. Next, we train the candidate structures based on a 5-fold cross-validation method, using the operational data from Wuhan-Guangzhou (W-G) and Xiamen-Shenzhen (X-S) high-speed railway (HSR) lines in China. The best performing structure is nominated to predict the consequences of disruptions and disturbances in the two HSR lines. Comparisons results show that the proposed model outperforms three other commonly used predictive models, reaching an average prediction accuracy of 96.6\%, 74.8\%, and 91.0\% on the W-G HSR line, and 94.8\%, 91.1\%, and 87.9\% on the X-S HSR line for variables L, N, and T, respectively.},
	language = {en},
	urldate = {2021-11-27},
	journal = {Transportation Research Part C: Emerging Technologies},
	author = {Huang, Ping and Lessan, Javad and Wen, Chao and Peng, Qiyuan and Fu, Liping and Li, Li and Xu, Xinyue},
	month = may,
	year = {2020},
	keywords = {Bayesian networks, Disturbances and disruptions, Real-time prediction, Train operation},
	pages = {338--358},
}

Based on the Bayesian network (BN) paradigm, we propose a hybrid model to predict the three main consequences of disruptions and disturbances during train operations, namely, the primary delay (L), the number of affected trains (N), and the total delay times (T). To obtain an effective BN structure, we first analyze the dependencies of the involved factors on each station and among adjacent stations, given domain knowledge and expertise about operational characteristics. We then put forward four candidate BN structures, integrating expert knowledge, the interdependencies learned from real-world data, and real-time prediction and operational requirements. Next, we train the candidate structures based on a 5-fold cross-validation method, using the operational data from Wuhan-Guangzhou (W-G) and Xiamen-Shenzhen (X-S) high-speed railway (HSR) lines in China. The best performing structure is nominated to predict the consequences of disruptions and disturbances in the two HSR lines. Comparisons results show that the proposed model outperforms three other commonly used predictive models, reaching an average prediction accuracy of 96.6%, 74.8%, and 91.0% on the W-G HSR line, and 94.8%, 91.1%, and 87.9% on the X-S HSR line for variables L, N, and T, respectively.

@article{fu_overview_2020,
	title = {An overview of recent multi-view clustering},
	volume = {402},
	issn = {0925-2312},
	url = {https://www.sciencedirect.com/science/article/pii/S0925231220303222},
	doi = {10.1016/j.neucom.2020.02.104},
	abstract = {With the widespread deployment of sensors and the Internet-of-Things, multi-view data has become more common and publicly available. Compared to traditional data that describes objects from single perspective, multi-view data is semantically richer, more useful, however more complex. Since traditional clustering algorithms cannot handle such data, multi-view clustering has become a research hotspot. In this paper, we review some of the latest multi-view clustering algorithms, which are reasonably divided into three categories. To evaluate their performance, we perform extensive experiments on seven real-world data sets. Three mainstream metrics are used, including clustering accuracy, normalized mutual information and purity. Based on the experimental results and a large number of literature reading, we also discuss existing problems in current multi-view clustering and point out possible research directions in the future. This research provides some insights for researchers in related fields and may further promote the development of multi-view clustering algorithms.},
	language = {en},
	urldate = {2021-11-07},
	journal = {Neurocomputing},
	author = {Fu, Lele and Lin, Pengfei and Vasilakos, Athanasios V. and Wang, Shiping},
	month = aug,
	year = {2020},
	keywords = {Graph-based clustering, Machine learning, Multi-view clustering, Space learning, Unsupervised learning, clustering, multiview, multiview clustering},
	pages = {148--161},
}

With the widespread deployment of sensors and the Internet-of-Things, multi-view data has become more common and publicly available. Compared to traditional data that describes objects from single perspective, multi-view data is semantically richer, more useful, however more complex. Since traditional clustering algorithms cannot handle such data, multi-view clustering has become a research hotspot. In this paper, we review some of the latest multi-view clustering algorithms, which are reasonably divided into three categories. To evaluate their performance, we perform extensive experiments on seven real-world data sets. Three mainstream metrics are used, including clustering accuracy, normalized mutual information and purity. Based on the experimental results and a large number of literature reading, we also discuss existing problems in current multi-view clustering and point out possible research directions in the future. This research provides some insights for researchers in related fields and may further promote the development of multi-view clustering algorithms.

@article{huang_mvstream_2020,
	title = {{MVStream}: {Multiview} {Data} {Stream} {Clustering}},
	volume = {31},
	issn = {2162-2388},
	shorttitle = {{MVStream}},
	doi = {10.1109/TNNLS.2019.2944851},
	abstract = {This article studies a new problem of data stream clustering, namely, multiview data stream (MVStream) clustering. Although many data stream clustering algorithms have been developed, they are restricted to the single-view streaming data, and clustering MVStreams still remains largely unsolved. In addition to the many issues encountered by the conventional single-view data stream clustering, such as capturing cluster evolution and discovering clusters of arbitrary shapes under the limited computational resources, the main challenge of MVStream clustering lies in integrating information from multiple views in a streaming manner and abstracting summary statistics from the integrated features simultaneously. In this article, we propose a novel MVStream clustering algorithm for the first time. The main idea is to design a multiview support vector domain description (MVSVDD) model, by which the information from multiple insufficient views can be integrated, and the outputting support vectors (SVs) are utilized to abstract the summary statistics of the historical multiview data objects. Based on the MVSVDD model, a new multiview cluster labeling method is designed, whereby clusters of arbitrary shapes can be discovered for each view. By tracking the cluster labels of SVs in each view, the cluster evolution associated with concept drift can be captured. Since the SVs occupy only a small portion of data objects, the proposed MVStream algorithm is quite efficient with the limited computational resources. Extensive experiments are conducted to demonstrate the effectiveness and efficiency of the proposed method.},
	number = {9},
	journal = {IEEE Transactions on Neural Networks and Learning Systems},
	author = {Huang, Ling and Wang, Chang-Dong and Chao, Hong-Yang and Yu, Philip S.},
	month = sep,
	year = {2020},
	note = {Conference Name: IEEE Transactions on Neural Networks and Learning Systems},
	keywords = {Clustering, Clustering algorithms, Computer science, Data models, Indexes, Shape, Support vector machines, Task analysis, clusters of arbitrary shapes, data stream, multiview, online, stream, stream learning, support vector (SV)},
	pages = {3482--3496},
}

This article studies a new problem of data stream clustering, namely, multiview data stream (MVStream) clustering. Although many data stream clustering algorithms have been developed, they are restricted to the single-view streaming data, and clustering MVStreams still remains largely unsolved. In addition to the many issues encountered by the conventional single-view data stream clustering, such as capturing cluster evolution and discovering clusters of arbitrary shapes under the limited computational resources, the main challenge of MVStream clustering lies in integrating information from multiple views in a streaming manner and abstracting summary statistics from the integrated features simultaneously. In this article, we propose a novel MVStream clustering algorithm for the first time. The main idea is to design a multiview support vector domain description (MVSVDD) model, by which the information from multiple insufficient views can be integrated, and the outputting support vectors (SVs) are utilized to abstract the summary statistics of the historical multiview data objects. Based on the MVSVDD model, a new multiview cluster labeling method is designed, whereby clusters of arbitrary shapes can be discovered for each view. By tracking the cluster labels of SVs in each view, the cluster evolution associated with concept drift can be captured. Since the SVs occupy only a small portion of data objects, the proposed MVStream algorithm is quite efficient with the limited computational resources. Extensive experiments are conducted to demonstrate the effectiveness and efficiency of the proposed method.

@article{chen_multi-view_2020,
	title = {Multi-{View} {Clustering} in {Latent} {Embedding} {Space}},
	volume = {34},
	copyright = {Copyright (c) 2020 Association for the Advancement of Artificial Intelligence},
	issn = {2374-3468},
	url = {https://ojs.aaai.org/index.php/AAAI/article/view/5756},
	doi = {10.1609/aaai.v34i04.5756},
	abstract = {Previous multi-view clustering algorithms mostly partition the multi-view data in their original feature space, the efficacy of which heavily and implicitly relies on the quality of the original feature presentation. In light of this, this paper proposes a novel approach termed Multi-view Clustering in Latent Embedding Space (MCLES), which is able to cluster the multi-view data in a learned latent embedding space while simultaneously learning the global structure and the cluster indicator matrix in a unified optimization framework. Specifically, in our framework, a latent embedding representation is firstly discovered which can effectively exploit the complementary information from different views. The global structure learning is then performed based on the learned latent embedding representation. Further, the cluster indicator matrix can be acquired directly with the learned global structure. An alternating optimization scheme is introduced to solve the optimization problem. Extensive experiments conducted on several real-world multi-view datasets have demonstrated the superiority of our approach.},
	language = {en},
	number = {04},
	urldate = {2021-11-07},
	journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
	author = {Chen, Man-Sheng and Huang, Ling and Wang, Chang-Dong and Huang, Dong},
	month = apr,
	year = {2020},
	note = {Number: 04},
	pages = {3513--3520},
}

Previous multi-view clustering algorithms mostly partition the multi-view data in their original feature space, the efficacy of which heavily and implicitly relies on the quality of the original feature presentation. In light of this, this paper proposes a novel approach termed Multi-view Clustering in Latent Embedding Space (MCLES), which is able to cluster the multi-view data in a learned latent embedding space while simultaneously learning the global structure and the cluster indicator matrix in a unified optimization framework. Specifically, in our framework, a latent embedding representation is firstly discovered which can effectively exploit the complementary information from different views. The global structure learning is then performed based on the learned latent embedding representation. Further, the cluster indicator matrix can be acquired directly with the learned global structure. An alternating optimization scheme is introduced to solve the optimization problem. Extensive experiments conducted on several real-world multi-view datasets have demonstrated the superiority of our approach.

@article{li_esa-stream_2020,
	title = {{ESA}-{Stream}: {Efficient} {Self}-{Adaptive} {Online} {Data} {Stream} {Clustering}},
	issn = {1558-2191},
	shorttitle = {{ESA}-{Stream}},
	doi = {10.1109/TKDE.2020.2990196},
	abstract = {Many big data applications produce a massive amount of high-dimensional, real-time, and evolving streaming data. Clustering such data streams with both effectiveness and efficiency are critical for these applications. Although there are well-known data stream clustering algorithms that are based on the popular online-offline framework, these algorithms still face some major challenges. Several critical questions are still not answer satisfactorily: How to perform dimensionality reduction effectively and efficiently in the online dynamic environment? How to enable the clustering algorithm to achieve complete real-time online processing? How to make algorithm parameters learn in a self-supervised or self-adaptive manner to cope with high-speed evolving streams? In this paper, we focus on tackling these challenges by proposing a fully online stream clustering algorithm (called ESA-Stream) that can learn parameters online dynamically in a self-adaptive manner, speedup dimensionality reduction, and cluster data streams effectively and efficiently in an online and dynamic environment Experiments on a wide range of synthetic and real-world data streams show that ESA-Stream outperforms state-of-the-art baselines considerably in both effectiveness and efficiency.},
	journal = {IEEE Transactions on Knowledge and Data Engineering},
	author = {Li, Yanni and Li, Hui and Wang, Zhi and Liu, Bing and Cui, Jiangtao and Fei, Hang},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Knowledge and Data Engineering},
	keywords = {Clustering algorithms, Clustering methods, Data Stream, Dimensionality reduction, Heuristic algorithms, Indexes, Online Clustering, Partitioning algorithms, Real-time systems, Self-Adaptive},
	pages = {1--1},
}

Many big data applications produce a massive amount of high-dimensional, real-time, and evolving streaming data. Clustering such data streams with both effectiveness and efficiency are critical for these applications. Although there are well-known data stream clustering algorithms that are based on the popular online-offline framework, these algorithms still face some major challenges. Several critical questions are still not answer satisfactorily: How to perform dimensionality reduction effectively and efficiently in the online dynamic environment? How to enable the clustering algorithm to achieve complete real-time online processing? How to make algorithm parameters learn in a self-supervised or self-adaptive manner to cope with high-speed evolving streams? In this paper, we focus on tackling these challenges by proposing a fully online stream clustering algorithm (called ESA-Stream) that can learn parameters online dynamically in a self-adaptive manner, speedup dimensionality reduction, and cluster data streams effectively and efficiently in an online and dynamic environment Experiments on a wide range of synthetic and real-world data streams show that ESA-Stream outperforms state-of-the-art baselines considerably in both effectiveness and efficiency.

@article{brito_da_silva_incremental_2020,
	title = {Incremental {Cluster} {Validity} {Indices} for {Online} {Learning} of {Hard} {Partitions}: {Extensions} and {Comparative} {Study}},
	volume = {8},
	issn = {2169-3536},
	shorttitle = {Incremental {Cluster} {Validity} {Indices} for {Online} {Learning} of {Hard} {Partitions}},
	doi = {10.1109/ACCESS.2020.2969849},
	abstract = {Validation is one of the most important aspects of clustering, particularly when the user is designing a trustworthy or explainable system. However, most clustering validation approaches require batch calculation. This is an important gap because of the value of clustering in real-time data streaming and other online learning applications. Therefore, interest has grown in providing online alternatives for validation. This paper extends the incremental cluster validity index (iCVI) family by presenting incremental versions of Calinski-Harabasz (iCH), Pakhira-Bandyopadhyay-Maulik (iPBM), WB index (iWB), Silhouette (iSIL), Negentropy Increment (iNI), Representative Cross Information Potential (irCIP), Representative Cross Entropy (irH), and Conn\_Index (iConn\_Index). This paper also provides a thorough comparative study of correct, under- and over-partitioning on the behavior of these iCVIs, the Partition Separation (PS) index as well as four recently introduced iCVIs: incremental Xie-Beni (iXB), incremental Davies-Bouldin (iDB), and incremental generalized Dunn's indices 43 and 53 (iGD43 and iGD53). Experiments were carried out using a framework that was designed to be as agnostic as possible to the clustering algorithms. The results on synthetic benchmark data sets showed that while evidence of most under-partitioning cases could be inferred from the behaviors of the majority of these iCVIs, over-partitioning was found to be a more challenging problem, detected by fewer of them. Interestingly, over-partitioning, rather then under-partitioning, was more prominently detected on the real-world data experiments within this study. The expansion of iCVIs provides significant novel opportunities for assessing and interpreting the results of unsupervised lifelong learning in real-time, wherein samples cannot be reprocessed due to memory and/or application constraints.},
	journal = {IEEE Access},
	author = {Brito Da Silva, Leonardo Enzo and Melton, Niklas Max and Wunsch, Donald C.},
	year = {2020},
	note = {Conference Name: IEEE Access},
	keywords = {Clustering, Clustering algorithms, Entropy, Government, Indexes, Partitioning algorithms, Prototypes, Subspace constraints, adaptive resonance theory (ART), cluster validation, cluster validity, data streams, incremental (online) clustering algorithms, incremental cluster validity index (iCVI), validation},
	pages = {22025--22047},
}

Validation is one of the most important aspects of clustering, particularly when the user is designing a trustworthy or explainable system. However, most clustering validation approaches require batch calculation. This is an important gap because of the value of clustering in real-time data streaming and other online learning applications. Therefore, interest has grown in providing online alternatives for validation. This paper extends the incremental cluster validity index (iCVI) family by presenting incremental versions of Calinski-Harabasz (iCH), Pakhira-Bandyopadhyay-Maulik (iPBM), WB index (iWB), Silhouette (iSIL), Negentropy Increment (iNI), Representative Cross Information Potential (irCIP), Representative Cross Entropy (irH), and Conn_Index (iConn_Index). This paper also provides a thorough comparative study of correct, under- and over-partitioning on the behavior of these iCVIs, the Partition Separation (PS) index as well as four recently introduced iCVIs: incremental Xie-Beni (iXB), incremental Davies-Bouldin (iDB), and incremental generalized Dunn's indices 43 and 53 (iGD43 and iGD53). Experiments were carried out using a framework that was designed to be as agnostic as possible to the clustering algorithms. The results on synthetic benchmark data sets showed that while evidence of most under-partitioning cases could be inferred from the behaviors of the majority of these iCVIs, over-partitioning was found to be a more challenging problem, detected by fewer of them. Interestingly, over-partitioning, rather then under-partitioning, was more prominently detected on the real-world data experiments within this study. The expansion of iCVIs provides significant novel opportunities for assessing and interpreting the results of unsupervised lifelong learning in real-time, wherein samples cannot be reprocessed due to memory and/or application constraints.

@article{yilmaz_pysad_2020,
	title = {{PySAD}: {A} {Streaming} {Anomaly} {Detection} {Framework} in {Python}},
	shorttitle = {{PySAD}},
	url = {http://arxiv.org/abs/2009.02572},
	abstract = {PySAD is an open-source python framework for anomaly detection on streaming data. PySAD serves various state-of-the-art methods for streaming anomaly detection. The framework provides a complete set of tools to design anomaly detection experiments ranging from projectors to probability calibrators. PySAD builds upon popular open-source frameworks such as PyOD and scikit-learn. We enforce software quality by enforcing compliance with PEP8 guidelines, functional testing and using continuous integration. The source code is publicly available on https://github.com/selimfirat/pysad.},
	urldate = {2021-10-15},
	journal = {arXiv:2009.02572 [cs, stat]},
	author = {Yilmaz, Selim F. and Kozat, Suleyman S.},
	month = sep,
	year = {2020},
	note = {arXiv: 2009.02572},
	keywords = {Computer Science - Machine Learning, Statistics - Machine Learning},
}

PySAD is an open-source python framework for anomaly detection on streaming data. PySAD serves various state-of-the-art methods for streaming anomaly detection. The framework provides a complete set of tools to design anomaly detection experiments ranging from projectors to probability calibrators. PySAD builds upon popular open-source frameworks such as PyOD and scikit-learn. We enforce software quality by enforcing compliance with PEP8 guidelines, functional testing and using continuous integration. The source code is publicly available on https://github.com/selimfirat/pysad.

@article{hammoud_multilayer_2020,
	title = {Multilayer networks: aspects, implementations, and application in biomedicine},
	volume = {5},
	issn = {2058-6345},
	shorttitle = {Multilayer networks},
	url = {https://doi.org/10.1186/s41044-020-00046-0},
	doi = {10.1186/s41044-020-00046-0},
	abstract = {Modeling and analyses of complex systems using network theory have been an object of study for a long time. They have caught attention in many disciplines such as sociology, epidemiology, ecology, psychology, biology, biomedicine, and other fields. Network theory is especially an efficient tool to model biological networks such as gene co-expression networks, protein-protein interaction networks, or pathways. Considering the enhanced resolutions of complex real-world systems, the interest has been directed to multilayered networks. However, despite this surge of recent attention, the use of the multilayer framework in the biological field is still in its youth. In this paper, we review the different aspects and terminologies of multilayered networks. We also briefly discuss the variant applications of the multilayer framework, and finally, we give an overview of various existing applications of the multilayer model in network biology.},
	number = {1},
	urldate = {2021-10-14},
	journal = {Big Data Analytics},
	author = {Hammoud, Zaynab and Kramer, Frank},
	month = jul,
	year = {2020},
	keywords = {Graph theory, Multilayered graphs, Network biology},
	pages = {2},
}

Modeling and analyses of complex systems using network theory have been an object of study for a long time. They have caught attention in many disciplines such as sociology, epidemiology, ecology, psychology, biology, biomedicine, and other fields. Network theory is especially an efficient tool to model biological networks such as gene co-expression networks, protein-protein interaction networks, or pathways. Considering the enhanced resolutions of complex real-world systems, the interest has been directed to multilayered networks. However, despite this surge of recent attention, the use of the multilayer framework in the biological field is still in its youth. In this paper, we review the different aspects and terminologies of multilayered networks. We also briefly discuss the variant applications of the multilayer framework, and finally, we give an overview of various existing applications of the multilayer model in network biology.

@inproceedings{wang_railway_2020,
	title = {Railway {Wagon} {Wheelset} {Fault} {Diagnosis} {Method} {Based} on {DBN}},
	doi = {10.1109/PHM-Shanghai49105.2020.9280980},
	abstract = {Wheelset is a crucial part of the operation and braking part of railway wagons, and its failure will affect the operation safety of the entire railway system. And fault diagnosis is an important basis for realizing railway scientific maintenance decisions. Learn the method of deep belief network (DBN) to process and diagnose fault data of freighters. The paper uses dynamic factors to optimize the update mechanism of particle velocity, proposes an active factor controlled particle swarm algorithm (APSO), and uses the APSO algorithm to optimize the selection of DBN weights to form a round-pair fault detection algorithm (APSO-DBN) This speeds up DBN training and improves the overall performance of the diagnostic model. The experimental steps and the effectiveness of the method are verified by experiments.},
	booktitle = {2020 {Global} {Reliability} and {Prognostics} and {Health} {Management} ({PHM}-{Shanghai})},
	author = {Wang, Hongkun and Li, Honghui and Li, Yusheng and Duan, Yuhang},
	month = oct,
	year = {2020},
	keywords = {Classification algorithms, Fault detection, Fault diagnosis, Heuristic algorithms, Particle swarm optimization, Rail transportation, Training, deep belief network, deep learning, fault detection, particle swarm algorithm, railway freighter wheelset},
	pages = {1--6},
}

Wheelset is a crucial part of the operation and braking part of railway wagons, and its failure will affect the operation safety of the entire railway system. And fault diagnosis is an important basis for realizing railway scientific maintenance decisions. Learn the method of deep belief network (DBN) to process and diagnose fault data of freighters. The paper uses dynamic factors to optimize the update mechanism of particle velocity, proposes an active factor controlled particle swarm algorithm (APSO), and uses the APSO algorithm to optimize the selection of DBN weights to form a round-pair fault detection algorithm (APSO-DBN) This speeds up DBN training and improves the overall performance of the diagnostic model. The experimental steps and the effectiveness of the method are verified by experiments.

@article{saucedo-dorantes_industrial_2020,
	title = {Industrial {Data}-{Driven} {Monitoring} {Based} on {Incremental} {Learning} {Applied} to the {Detection} of {Novel} {Faults}},
	volume = {16},
	issn = {1941-0050},
	doi = {10.1109/TII.2020.2973731},
	abstract = {The detection of uncharacterized events during electromechanical systems operation represents one of the most critical data challenges dealing with condition-based monitoring under the Industry 4.0 framework. Thus, the detection of novelty conditions and the learning of new patterns are considered as mandatory competencies in modern industrial applications. In this regard, this article proposes a novel multifault detection and identification scheme, based on machine learning, information data-fusion, novelty-detection, and incremental learning. First, statistical time-domain features estimated from multiple physical magnitudes acquired from the electrical motor under inspection are fused under a feature-fusion level scheme. Second, a self-organizing map structure is proposed to construct a data-based model of the available conditions of operation. Third, the incremental learning of the condition-based monitoring scheme is performed adding self-organizing structures and optimizing their projections through a linear discriminant analysis. The performance of the proposed scheme is validated under a complete set of experimental scenarios from two different cases of study, and the results compared with a classical approach.},
	number = {9},
	journal = {IEEE Transactions on Industrial Informatics},
	author = {Saucedo-Dorantes, Juan Jose and Delgado-Prieto, Miguel and Osornio-Rios, Roque Alfredo and Romero-Troncoso, Rene de Jesus},
	month = sep,
	year = {2020},
	note = {Conference Name: IEEE Transactions on Industrial Informatics},
	keywords = {Condition monitoring, Data models, Electromechanical systems, Fault detection, Industries, Monitoring, Self-organizing feature maps, Training, fault detection, feature extraction, incremental learning, machine learning, novelty detection},
	pages = {5985--5995},
}

The detection of uncharacterized events during electromechanical systems operation represents one of the most critical data challenges dealing with condition-based monitoring under the Industry 4.0 framework. Thus, the detection of novelty conditions and the learning of new patterns are considered as mandatory competencies in modern industrial applications. In this regard, this article proposes a novel multifault detection and identification scheme, based on machine learning, information data-fusion, novelty-detection, and incremental learning. First, statistical time-domain features estimated from multiple physical magnitudes acquired from the electrical motor under inspection are fused under a feature-fusion level scheme. Second, a self-organizing map structure is proposed to construct a data-based model of the available conditions of operation. Third, the incremental learning of the condition-based monitoring scheme is performed adding self-organizing structures and optimizing their projections through a linear discriminant analysis. The performance of the proposed scheme is validated under a complete set of experimental scenarios from two different cases of study, and the results compared with a classical approach.

@article{shen_deep_2020,
	title = {A {Deep} {Multi}-{Label} {Learning} {Framework} for the {Intelligent} {Fault} {Diagnosis} of {Machines}},
	volume = {8},
	issn = {2169-3536},
	doi = {10.1109/ACCESS.2020.3002826},
	abstract = {Deep learning has been applied in intelligent fault diagnosis of machines since it trains deep neural networks to simultaneously learn features and recognize faults. In the intelligent fault diagnosis methods based on deep learning, feature learning and fault recognition are achieved by solving a multi-class classification problem. The multi-class classification, however, has not considered the relationships of fault labels, leading to two weaknesses of these methods. One is that it cannot ensure to learn the correlated features for related faults and the other is that it cannot handle missing label problem. To overcome these weaknesses, we introduce a concept of multi-label classification into intelligent fault diagnosis and propose a deep multi-label learning framework called multi-label convolutional neural network (MLCNN). MLCNN builds the relationship between the labels, and thus it is able to learn the correlated features from mechanical vibration signals and be well trained using the samples with missing labels. A motor bearing diagnosis case and a compound fault diagnosis case are used to verify the proposed method, respectively. The results show that the relationships between features are learned by MLCNN, and the classification accuracies of MLCNN are higher than traditional methods when the missing label problem occurs.},
	journal = {IEEE Access},
	author = {Shen, Jianjun and Li, Shihao and Jia, Feng and Zuo, Hao and Ma, Junxing},
	year = {2020},
	note = {Conference Name: IEEE Access},
	keywords = {Compounds, Deep learning, Fault diagnosis, Neural networks, Task analysis, Training, Vibrations, bearing, compound fault, intelligent fault diagnosis, missing label problem, multi-label classification, multilabel},
	pages = {113557--113566},
}

Deep learning has been applied in intelligent fault diagnosis of machines since it trains deep neural networks to simultaneously learn features and recognize faults. In the intelligent fault diagnosis methods based on deep learning, feature learning and fault recognition are achieved by solving a multi-class classification problem. The multi-class classification, however, has not considered the relationships of fault labels, leading to two weaknesses of these methods. One is that it cannot ensure to learn the correlated features for related faults and the other is that it cannot handle missing label problem. To overcome these weaknesses, we introduce a concept of multi-label classification into intelligent fault diagnosis and propose a deep multi-label learning framework called multi-label convolutional neural network (MLCNN). MLCNN builds the relationship between the labels, and thus it is able to learn the correlated features from mechanical vibration signals and be well trained using the samples with missing labels. A motor bearing diagnosis case and a compound fault diagnosis case are used to verify the proposed method, respectively. The results show that the relationships between features are learned by MLCNN, and the classification accuracies of MLCNN are higher than traditional methods when the missing label problem occurs.

@article{bechini_tsf-dbscan_2020,
	title = {{TSF}-{DBSCAN}: a {Novel} {Fuzzy} {Density}-based {Approach} for {Clustering} {Unbounded} {Data} {Streams}},
	issn = {1941-0034},
	shorttitle = {{TSF}-{DBSCAN}},
	doi = {10.1109/TFUZZ.2020.3042645},
	abstract = {In recent years, several clustering algorithms have been proposed with the aim of mining knowledge from streams of data generated at a high speed by a variety of hardware platforms and software applications. Among these algorithms, density-based approaches have proved to be particularly attractive, thanks to their capability of handling outliers and capturing clusters with arbitrary shapes. The streaming setting poses additional challenges that need to be addressed as well: data streams are potentially unbounded and affected by concept drift, i.e. a modification over time in the underlying data generation process. In this paper, we propose Temporal Streaming Fuzzy DBSCAN (TSF-DBSCAN), a novel fuzzy clustering algorithm for streaming data. TSF-DBSCAN is an extension of the well-known DBSCAN algorithm, one of the most popular density-based clustering approaches. Fuzziness is introduced in TSF-DBSCAN to model the uncertainty about the distance threshold that defines the neighborhood of an object. As a consequence, TSF-DBSCAN identifies clusters with fuzzy overlapping borders. A fading model, which makes objects less relevant as they become more remote in time, endows TSF-DBSCAN with the capability of adapting to evolving data streams. The integration of the model in a two-stage approach ensures computational and memory efficiency: during the online stage continuously arriving objects are organized in proper data structures that are later exploited in the offline stage to determine a fine-grained partition. An extensive experimental analysis on synthetic and real world datasets shows that TSF-DBSCAN yields competitive performance when compared to other clustering algorithms recently proposed for streaming data.},
	journal = {IEEE Transactions on Fuzzy Systems},
	author = {Bechini, Alessio and Marcelloni, Francesco and Renda, Alessandro},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Fuzzy Systems},
	keywords = {Adaptation models, Clustering algorithms, DBSCAN, Data models, Data stream clustering, Fading channels, Partitioning algorithms, Proposals, Shape, density-based clustering, fuzzy clustering, streaming data},
	pages = {1--1},
}

In recent years, several clustering algorithms have been proposed with the aim of mining knowledge from streams of data generated at a high speed by a variety of hardware platforms and software applications. Among these algorithms, density-based approaches have proved to be particularly attractive, thanks to their capability of handling outliers and capturing clusters with arbitrary shapes. The streaming setting poses additional challenges that need to be addressed as well: data streams are potentially unbounded and affected by concept drift, i.e. a modification over time in the underlying data generation process. In this paper, we propose Temporal Streaming Fuzzy DBSCAN (TSF-DBSCAN), a novel fuzzy clustering algorithm for streaming data. TSF-DBSCAN is an extension of the well-known DBSCAN algorithm, one of the most popular density-based clustering approaches. Fuzziness is introduced in TSF-DBSCAN to model the uncertainty about the distance threshold that defines the neighborhood of an object. As a consequence, TSF-DBSCAN identifies clusters with fuzzy overlapping borders. A fading model, which makes objects less relevant as they become more remote in time, endows TSF-DBSCAN with the capability of adapting to evolving data streams. The integration of the model in a two-stage approach ensures computational and memory efficiency: during the online stage continuously arriving objects are organized in proper data structures that are later exploited in the offline stage to determine a fine-grained partition. An extensive experimental analysis on synthetic and real world datasets shows that TSF-DBSCAN yields competitive performance when compared to other clustering algorithms recently proposed for streaming data.

@article{lei_applications_2020,
	title = {Applications of machine learning to machine fault diagnosis: {A} review and roadmap},
	volume = {138},
	issn = {0888-3270},
	shorttitle = {Applications of machine learning to machine fault diagnosis},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327019308088},
	doi = {10.1016/j.ymssp.2019.106587},
	abstract = {Intelligent fault diagnosis (IFD) refers to applications of machine learning theories to machine fault diagnosis. This is a promising way to release the contribution from human labor and automatically recognize the health states of machines, thus it has attracted much attention in the last two or three decades. Although IFD has achieved a considerable number of successes, a review still leaves a blank space to systematically cover the development of IFD from the cradle to the bloom, and rarely provides potential guidelines for the future development. To bridge the gap, this article presents a review and roadmap to systematically cover the development of IFD following the progress of machine learning theories and offer a future perspective. In the past, traditional machine learning theories began to weak the contribution of human labor and brought the era of artificial intelligence to machine fault diagnosis. Over the recent years, the advent of deep learning theories has reformed IFD in further releasing the artificial assistance since the 2010s, which encourages to construct an end-to-end diagnosis procedure. It means to directly bridge the relationship between the increasingly-grown monitoring data and the health states of machines. In the future, transfer learning theories attempt to use the diagnosis knowledge from one or multiple diagnosis tasks to other related ones, which prospectively overcomes the obstacles in applications of IFD to engineering scenarios. Finally, the roadmap of IFD is pictured to show potential research trends when combined with the challenges in this field.},
	language = {en},
	urldate = {2021-09-30},
	journal = {Mechanical Systems and Signal Processing},
	author = {Lei, Yaguo and Yang, Bin and Jiang, Xinwei and Jia, Feng and Li, Naipeng and Nandi, Asoke K.},
	month = apr,
	year = {2020},
	keywords = {Deep learning, Intelligent fault diagnosis, Machine learning, Machines, Review and roadmap, Transfer learning},
	pages = {106587},
}

Intelligent fault diagnosis (IFD) refers to applications of machine learning theories to machine fault diagnosis. This is a promising way to release the contribution from human labor and automatically recognize the health states of machines, thus it has attracted much attention in the last two or three decades. Although IFD has achieved a considerable number of successes, a review still leaves a blank space to systematically cover the development of IFD from the cradle to the bloom, and rarely provides potential guidelines for the future development. To bridge the gap, this article presents a review and roadmap to systematically cover the development of IFD following the progress of machine learning theories and offer a future perspective. In the past, traditional machine learning theories began to weak the contribution of human labor and brought the era of artificial intelligence to machine fault diagnosis. Over the recent years, the advent of deep learning theories has reformed IFD in further releasing the artificial assistance since the 2010s, which encourages to construct an end-to-end diagnosis procedure. It means to directly bridge the relationship between the increasingly-grown monitoring data and the health states of machines. In the future, transfer learning theories attempt to use the diagnosis knowledge from one or multiple diagnosis tasks to other related ones, which prospectively overcomes the obstacles in applications of IFD to engineering scenarios. Finally, the roadmap of IFD is pictured to show potential research trends when combined with the challenges in this field.

@article{cheng_online_2020,
	title = {Online {Bearing} {Remaining} {Useful} {Life} {Prediction} {Based} on a {Novel} {Degradation} {Indicator} and {Convolutional} {Neural} {Networks}},
	volume = {25},
	issn = {1083-4435, 1941-014X},
	url = {http://arxiv.org/abs/1812.03315},
	doi = {10.1109/TMECH.2020.2971503},
	abstract = {In industrial applications, nearly half the failures of motors are caused by the degradation of rolling element bearings (REBs). Therefore, accurately estimating the remaining useful life (RUL) for REBs are of crucial importance to ensure the reliability and safety of mechanical systems. To tackle this challenge, model-based approaches are often limited by the complexity of mathematical modeling. Conventional data-driven approaches, on the other hand, require massive efforts to extract the degradation features and construct health index. In this paper, a novel online data-driven framework is proposed to exploit the adoption of deep convolutional neural networks (CNN) in predicting the RUL of bearings. More concretely, the raw vibrations of training bearings are first processed using the Hilbert-Huang transform (HHT) and a novel nonlinear degradation indicator is constructed as the label for learning. The CNN is then employed to identify the hidden pattern between the extracted degradation indicator and the vibration of training bearings, which makes it possible to estimate the degradation of the test bearings automatically. Finally, testing bearings' RULs are predicted by using a \${\textbackslash}epsilon\$-support vector regression model. The superior performance of the proposed RUL estimation framework, compared with the state-of-the-art approaches, is demonstrated through the experimental results. The generality of the proposed CNN model is also validated by transferring to bearings undergoing different operating conditions.},
	number = {3},
	urldate = {2021-09-30},
	journal = {IEEE/ASME Transactions on Mechatronics},
	author = {Cheng, Cheng and Ma, Guijun and Zhang, Yong and Sun, Mingyang and Teng, Fei and Ding, Han and Yuan, Ye},
	month = jun,
	year = {2020},
	note = {arXiv: 1812.03315},
	keywords = {Computer Science - Machine Learning, Electrical Engineering and Systems Science - Signal Processing, Statistics - Machine Learning},
	pages = {1243--1254},
}

In industrial applications, nearly half the failures of motors are caused by the degradation of rolling element bearings (REBs). Therefore, accurately estimating the remaining useful life (RUL) for REBs are of crucial importance to ensure the reliability and safety of mechanical systems. To tackle this challenge, model-based approaches are often limited by the complexity of mathematical modeling. Conventional data-driven approaches, on the other hand, require massive efforts to extract the degradation features and construct health index. In this paper, a novel online data-driven framework is proposed to exploit the adoption of deep convolutional neural networks (CNN) in predicting the RUL of bearings. More concretely, the raw vibrations of training bearings are first processed using the Hilbert-Huang transform (HHT) and a novel nonlinear degradation indicator is constructed as the label for learning. The CNN is then employed to identify the hidden pattern between the extracted degradation indicator and the vibration of training bearings, which makes it possible to estimate the degradation of the test bearings automatically. Finally, testing bearings' RULs are predicted by using a ${\}epsilon$-support vector regression model. The superior performance of the proposed RUL estimation framework, compared with the state-of-the-art approaches, is demonstrated through the experimental results. The generality of the proposed CNN model is also validated by transferring to bearings undergoing different operating conditions.

@article{wang_integrated_2020,
	title = {An integrated fault diagnosis and prognosis approach for predictive maintenance of wind turbine bearing with limited samples},
	volume = {145},
	issn = {0960-1481},
	url = {https://www.sciencedirect.com/science/article/pii/S0960148119309371},
	doi = {10.1016/j.renene.2019.06.103},
	abstract = {Predictive maintenance has raised much research interest to improve the system reliability of a wind turbine. This paper presents a new model based approach of integrated fault diagnosis and prognosis for wind turbine remaining useful life estimation, especially the cases with limited degradation data. Firstly, a wavelet transform based fault diagnosis method is investigated to analyze the bearing incipient defect signatures, and the extracted features are then fused by the Health Index algorithm to represent the bearing defect conditions. Taking the empirical physical knowledge and statistical model in a Bayesian framework, the bearing remaining useful life prediction with uncertainty quantification is achieved by particle filter in a recursive manner. The integrated fault diagnosis and prognosis approach is validated using bearing lifetime test data acquired from a wind turbine in field, and the performance comparison with typical data driven technique outlines the significance of the presented method.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Renewable Energy},
	author = {Wang, Jinjiang and Liang, Yuanyuan and Zheng, Yinghao and Gao, Robert X. and Zhang, Fengli},
	month = jan,
	year = {2020},
	keywords = {Defect diagnosis, Defect prognosis, Particle filter, Wind turbine bearing},
	pages = {642--650},
}

Predictive maintenance has raised much research interest to improve the system reliability of a wind turbine. This paper presents a new model based approach of integrated fault diagnosis and prognosis for wind turbine remaining useful life estimation, especially the cases with limited degradation data. Firstly, a wavelet transform based fault diagnosis method is investigated to analyze the bearing incipient defect signatures, and the extracted features are then fused by the Health Index algorithm to represent the bearing defect conditions. Taking the empirical physical knowledge and statistical model in a Bayesian framework, the bearing remaining useful life prediction with uncertainty quantification is achieved by particle filter in a recursive manner. The integrated fault diagnosis and prognosis approach is validated using bearing lifetime test data acquired from a wind turbine in field, and the performance comparison with typical data driven technique outlines the significance of the presented method.

@article{lee_integrated_2020,
	title = {An integrated assessment of safety and efficiency of aircraft maintenance strategies using agent-based modelling and stochastic {Petri} nets},
	volume = {202},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832020305536},
	doi = {10.1016/j.ress.2020.107052},
	abstract = {Aircraft maintenance is key for safe and efficient aircraft operations. While most studies propose cost-efficient maintenance strategies, the safety and efficiency of these strategies need to be quantified. This paper proposes a formal framework to assess the safety and efficiency of maintenance strategies by means of agent-based modelling, stochastically and dynamically coloured Petri nets, and Monte Carlo simulation. We model an end-to-end aircraft maintenance process, considering several maintenance stakeholders. We apply our framework for aircraft landing gear brakes, and use a Gamma process to model the degradation trends of the brakes. The numerical results show that applying data-driven strategies reduces the number of inspections by 36\%, while maintaining the same level of safety as in the case of traditional time-based maintenance strategies. Furthermore, in order to discuss the possibility to substitute all inspections by sensor monitoring, an advanced data-driven strategy using prognostics is considered. Overall, our proposed framework is generic and can readily be applied to assess the safety and efficiency of the maintenance of other aircraft components and maintenance strategies.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Lee, Juseong and Mitici, Mihaela},
	month = oct,
	year = {2020},
	keywords = {Aircraft maintenance, Gamma process, Landing gear brake, Safety, Simulation, Stochastically and dynamically coloured Petri nets},
	pages = {107052},
}

Aircraft maintenance is key for safe and efficient aircraft operations. While most studies propose cost-efficient maintenance strategies, the safety and efficiency of these strategies need to be quantified. This paper proposes a formal framework to assess the safety and efficiency of maintenance strategies by means of agent-based modelling, stochastically and dynamically coloured Petri nets, and Monte Carlo simulation. We model an end-to-end aircraft maintenance process, considering several maintenance stakeholders. We apply our framework for aircraft landing gear brakes, and use a Gamma process to model the degradation trends of the brakes. The numerical results show that applying data-driven strategies reduces the number of inspections by 36%, while maintaining the same level of safety as in the case of traditional time-based maintenance strategies. Furthermore, in order to discuss the possibility to substitute all inspections by sensor monitoring, an advanced data-driven strategy using prognostics is considered. Overall, our proposed framework is generic and can readily be applied to assess the safety and efficiency of the maintenance of other aircraft components and maintenance strategies.

@inproceedings{tegen_effects_2020,
	title = {The {Effects} of {Reluctant} and {Fallible} {Users} in {Interactive} {Online} {Machine} {Learning}},
	url = {http://urn.kb.se/resolve?urn=urn:nbn:se:mau:diva-17673},
	abstract = {DiVA portal is a finding tool for research publications and student theses written at the following 50 universities and research institutions.},
	language = {eng},
	urldate = {2021-09-05},
	publisher = {CEUR Workshops},
	author = {Tegen, Agnes and Davidsson, Paul and Persson, Jan A.},
	year = {2020},
	pages = {55--71},
}

DiVA portal is a finding tool for research publications and student theses written at the following 50 universities and research institutions.

@article{tavenard_tslearn_2020,
	title = {Tslearn, {A} {Machine} {Learning} {Toolkit} for {Time} {Series} {Data}},
	volume = {21},
	issn = {1533-7928},
	url = {http://jmlr.org/papers/v21/20-091.html},
	abstract = {tslearn is a general-purpose Python machine learning library for time series that offers tools for pre-processing and feature extraction as well as dedicated models for clustering, classification and regression. It follows scikit-learn's Application Programming Interface for transformers and estimators, allowing the use of standard pipelines and model selection tools on top of tslearn objects. It is distributed under the BSD-2-Clause license, and its source code is available at https://github.com/tslearn-team/tslearn.},
	number = {118},
	urldate = {2021-08-27},
	journal = {Journal of Machine Learning Research},
	author = {Tavenard, Romain and Faouzi, Johann and Vandewiele, Gilles and Divo, Felix and Androz, Guillaume and Holtz, Chester and Payne, Marie and Yurchak, Roman and Rußwurm, Marc and Kolar, Kushal and Woods, Eli},
	year = {2020},
	pages = {1--6},
}

tslearn is a general-purpose Python machine learning library for time series that offers tools for pre-processing and feature extraction as well as dedicated models for clustering, classification and regression. It follows scikit-learn's Application Programming Interface for transformers and estimators, allowing the use of standard pipelines and model selection tools on top of tslearn objects. It is distributed under the BSD-2-Clause license, and its source code is available at https://github.com/tslearn-team/tslearn.

@inproceedings{tamburri_sustainable_2020,
	title = {Sustainable {MLOps}: {Trends} and {Challenges}},
	shorttitle = {Sustainable {MLOps}},
	doi = {10.1109/SYNASC51798.2020.00015},
	abstract = {Even simply through a GoogleTrends search it becomes clear that Machine-Learning Operations-or MLOps, for short-are climbing in interest from both a scientific and practical perspective. On the one hand, software components and middleware are proliferating to support all manners of MLOps, from AutoML (i.e., software which enables developers with limited machine-learning expertise to train high-quality models specific to their domain or data) to feature-specific ML engineering, e.g., Explainability and Interpretability. On the other hand, the more these platforms penetrate the day-to-day activities of software operations, the more the risk for AI Software becoming unsustainable from a social, technical, or organisational perspective. This paper offers a concise definition of MLOps and AI Software Sustainability and outlines key challenges in its pursuit.},
	booktitle = {2020 22nd {International} {Symposium} on {Symbolic} and {Numeric} {Algorithms} for {Scientific} {Computing} ({SYNASC})},
	author = {Tamburri, Damian A.},
	month = sep,
	year = {2020},
	keywords = {DataOps, Decision making, MLOps, Machine learning, Machine-Learning Operations, Market research, Middleware, Scientific computing, Software Sustainability, Software systems, Sustainable development},
	pages = {17--23},
}

Even simply through a GoogleTrends search it becomes clear that Machine-Learning Operations-or MLOps, for short-are climbing in interest from both a scientific and practical perspective. On the one hand, software components and middleware are proliferating to support all manners of MLOps, from AutoML (i.e., software which enables developers with limited machine-learning expertise to train high-quality models specific to their domain or data) to feature-specific ML engineering, e.g., Explainability and Interpretability. On the other hand, the more these platforms penetrate the day-to-day activities of software operations, the more the risk for AI Software becoming unsustainable from a social, technical, or organisational perspective. This paper offers a concise definition of MLOps and AI Software Sustainability and outlines key challenges in its pursuit.

@article{li_survey_2020,
	title = {A {Survey} of {Data}-driven and {Knowledge}-aware {eXplainable} {AI}},
	issn = {1558-2191},
	doi = {10.1109/TKDE.2020.2983930},
	abstract = {We are witnessing a fast development of Artificial Intelligence (AI), but it becomes dramatically challenging to explain AI models in the past decade. “Explanation” has a flexible philosophical concept of “satisfying the subjective curiosity for causal information”, driving a wide spectrum of methods being invented and/or adapted from many aspects and communities, including machine learning, visual analytics, human-computer interaction and so on. Nevertheless, from the view-point of data and knowledge engineering (DKE), a best explaining practice that is cost-effective in terms of extra intelligence acquisition should exploit the causal information and scenarios which is hidden richly in the data itself. In the past several years, there are plenty of works contributing in this line but there is a lack of a clear taxonomy and systematic review of the current effort. To this end, we propose this survey, reviewing and taxonomizing existing efforts from the view-point of DKE, summarizing their contribution, technical essence and comparative characteristics. Specifically, we categorize methods into data-driven methods where explanation comes from the task-related data, and knowledge-aware methods where extraneous knowledge is incorporated. Furthermore, in the light of practice, we provide survey of state-of-art evaluation metrics and deployed explanation applications in industrial practice.},
	journal = {IEEE Transactions on Knowledge and Data Engineering},
	author = {Li, Xiao-Hui and Cao, Caleb Chen and Shi, Yuhan and Bai, Wei and Gao, Han and Qiu, Luyu and Wang, Cong and Gao, Yuanyuan and Zhang, Shenjia and Xue, Xun and Chen, Lei},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Knowledge and Data Engineering},
	keywords = {Algorithms, Data models, Data visualization, Deep Learning, Explainable AI (XAI), Feature extraction, Knowledge-base, Metrics, Predictive models, Task analysis, Taxonomy},
	pages = {1--1},
}

We are witnessing a fast development of Artificial Intelligence (AI), but it becomes dramatically challenging to explain AI models in the past decade. “Explanation” has a flexible philosophical concept of “satisfying the subjective curiosity for causal information”, driving a wide spectrum of methods being invented and/or adapted from many aspects and communities, including machine learning, visual analytics, human-computer interaction and so on. Nevertheless, from the view-point of data and knowledge engineering (DKE), a best explaining practice that is cost-effective in terms of extra intelligence acquisition should exploit the causal information and scenarios which is hidden richly in the data itself. In the past several years, there are plenty of works contributing in this line but there is a lack of a clear taxonomy and systematic review of the current effort. To this end, we propose this survey, reviewing and taxonomizing existing efforts from the view-point of DKE, summarizing their contribution, technical essence and comparative characteristics. Specifically, we categorize methods into data-driven methods where explanation comes from the task-related data, and knowledge-aware methods where extraneous knowledge is incorporated. Furthermore, in the light of practice, we provide survey of state-of-art evaluation metrics and deployed explanation applications in industrial practice.

@inproceedings{salierno_architecture_2020,
	address = {Cham},
	series = {Lecture {Notes} in {Business} {Information} {Processing}},
	title = {An {Architecture} for {Predictive} {Maintenance} of {Railway} {Points} {Based} on {Big} {Data} {Analytics}},
	isbn = {978-3-030-49165-9},
	doi = {10.1007/978-3-030-49165-9_3},
	abstract = {Massive amounts of data produced by railway systems are a valuable resource to enable Big Data analytics. Despite its richness, several challenges arise when dealing with the deployment of a big data architecture into a railway system. In this paper, we propose a four-layers big data architecture with the goal of establishing a data management policy to manage massive amounts of data produced by railway switch points and perform analytical tasks efficiently. An implementation of the architecture is given along with the realization of a Long Short-Term Memory prediction model for detecting failures on the Italian Railway Line of Milano - Monza - Chiasso.},
	language = {en},
	booktitle = {Advanced {Information} {Systems} {Engineering} {Workshops}},
	publisher = {Springer International Publishing},
	author = {Salierno, Giulio and Morvillo, Sabatino and Leonardi, Letizia and Cabri, Giacomo},
	editor = {Dupuy-Chessa, Sophie and Proper, Henderik A.},
	year = {2020},
	keywords = {Big data architecture, Predictive maintenance, Railway data},
	pages = {29--40},
}

Massive amounts of data produced by railway systems are a valuable resource to enable Big Data analytics. Despite its richness, several challenges arise when dealing with the deployment of a big data architecture into a railway system. In this paper, we propose a four-layers big data architecture with the goal of establishing a data management policy to manage massive amounts of data produced by railway switch points and perform analytical tasks efficiently. An implementation of the architecture is given along with the realization of a Long Short-Term Memory prediction model for detecting failures on the Italian Railway Line of Milano - Monza - Chiasso.

@inproceedings{guderlei_evaluating_2020,
	address = {Barcelona, Spain (Online)},
	title = {Evaluating {Unsupervised} {Representation} {Learning} for {Detecting} {Stances} of {Fake} {News}},
	url = {https://www.aclweb.org/anthology/2020.coling-main.558},
	doi = {10.18653/v1/2020.coling-main.558},
	abstract = {Our goal is to evaluate the usefulness of unsupervised representation learning techniques for detecting stances of Fake News. Therefore we examine several pre-trained language models with respect to their performance on two Fake News related data sets, both consisting of instances with a headline, an associated news article and the stance of the article towards the respective headline. Specifically, the aim is to understand how much hyperparameter tuning is necessary when fine-tuning the pre-trained architectures, how well transfer learning works in this specific case of stance detection and how sensitive the models are to changes in hyperparameters like batch size, learning rate (schedule), sequence length as well as the freezing technique. The results indicate that the computationally more expensive autoregression approach of XLNet (Yanget al., 2019) is outperformed by BERT-based models, notably by RoBERTa (Liu et al., 2019).While the learning rate seems to be the most important hyperparameter, experiments with different freezing techniques indicate that all evaluated architectures had already learned powerful language representations that pose a good starting point for fine-tuning them.},
	urldate = {2021-01-29},
	booktitle = {Proceedings of the 28th {International} {Conference} on {Computational} {Linguistics}},
	publisher = {International Committee on Computational Linguistics},
	author = {Guderlei, Maike and Aßenmacher, Matthias},
	month = dec,
	year = {2020},
	pages = {6339--6349},
}

Our goal is to evaluate the usefulness of unsupervised representation learning techniques for detecting stances of Fake News. Therefore we examine several pre-trained language models with respect to their performance on two Fake News related data sets, both consisting of instances with a headline, an associated news article and the stance of the article towards the respective headline. Specifically, the aim is to understand how much hyperparameter tuning is necessary when fine-tuning the pre-trained architectures, how well transfer learning works in this specific case of stance detection and how sensitive the models are to changes in hyperparameters like batch size, learning rate (schedule), sequence length as well as the freezing technique. The results indicate that the computationally more expensive autoregression approach of XLNet (Yanget al., 2019) is outperformed by BERT-based models, notably by RoBERTa (Liu et al., 2019).While the learning rate seems to be the most important hyperparameter, experiments with different freezing techniques indicate that all evaluated architectures had already learned powerful language representations that pose a good starting point for fine-tuning them.

@article{verma_edge-cloud_2020,
	title = {Edge-cloud computing performance benchmarking for {IoT} based machinery vibration monitoring},
	issn = {2213-8463},
	url = {http://www.sciencedirect.com/science/article/pii/S2213846320301759},
	doi = {10.1016/j.mfglet.2020.12.004},
	abstract = {Advances in low cost and reliable sensing, connectivity (Internet of Things), computational power, and advanced analytics, are leading to a new wave of innovation in machinery status sensing and condition monitoring. Significant research efforts are directed towards cloud computing architectures. However, given the latency, bandwidth, cost, security, and privacy concerns, further supported by the ever-increasing capabilities of edge computing devices, there is a need to consider both edge and cloud computing together to make informed decisions based upon context and performance. We present an edge-cloud performance evaluation for IoT based machinery vibration monitoring, to foster deployment for the contexts considered},
	language = {en},
	urldate = {2020-12-15},
	journal = {Manufacturing Letters},
	author = {Verma, Ankur and Goyal, Ayush and Kumara, Soundar and Kurfess, Thomas},
	month = dec,
	year = {2020},
	keywords = {Edge computing, Internet of Things, machinery vibration monitoring, smart manufacturing},
}

Advances in low cost and reliable sensing, connectivity (Internet of Things), computational power, and advanced analytics, are leading to a new wave of innovation in machinery status sensing and condition monitoring. Significant research efforts are directed towards cloud computing architectures. However, given the latency, bandwidth, cost, security, and privacy concerns, further supported by the ever-increasing capabilities of edge computing devices, there is a need to consider both edge and cloud computing together to make informed decisions based upon context and performance. We present an edge-cloud performance evaluation for IoT based machinery vibration monitoring, to foster deployment for the contexts considered

@article{zhang_bibliometric_2020,
	title = {A bibliometric review of a decade of research: {Big} data in business research – {Setting} a research agenda},
	issn = {0148-2963},
	shorttitle = {A bibliometric review of a decade of research},
	url = {http://www.sciencedirect.com/science/article/pii/S0148296320307475},
	doi = {10.1016/j.jbusres.2020.11.004},
	abstract = {The last several years have witnessed a surge of interest in artificial intelligence (AI). As the foundation of AI technologies, big data has attracted attention of researchers. Big data and data science have been recognized as new tools and methodologies for developing theories in business research (George, 2014). While several qualitative reviews have been conducted, there is still a lack of a quantitative and systematic review of big data in business research. Our review study fills this gap by depicting the development of big data in business research using bibliometric methods, such as publication counts and trends analysis, co-citation analysis, co-authorship analysis and keywords co-occurrence analysis. Based on the sample of 1366 primary focal articles and 55,718 secondary references, we visualize the landscape and evolution of big-data business research and capture the developmental trajectory and trends over time (between 2008 and 2018). Furthermore, based on our analyses, we provide several promising directions for future research. In doing so, we provide scholars with a systematic understanding of the development and panoramic roadmap of big data research in business.},
	language = {en},
	urldate = {2020-12-12},
	journal = {Journal of Business Research},
	author = {Zhang, Yucheng and Zhang, Meng and Li, Jing and Liu, Guangjian and Yang, Miles M. and Liu, Siqi},
	month = dec,
	year = {2020},
	keywords = {Bibliometric review, Big Data, Management and business, Scientific visualization},
}

The last several years have witnessed a surge of interest in artificial intelligence (AI). As the foundation of AI technologies, big data has attracted attention of researchers. Big data and data science have been recognized as new tools and methodologies for developing theories in business research (George, 2014). While several qualitative reviews have been conducted, there is still a lack of a quantitative and systematic review of big data in business research. Our review study fills this gap by depicting the development of big data in business research using bibliometric methods, such as publication counts and trends analysis, co-citation analysis, co-authorship analysis and keywords co-occurrence analysis. Based on the sample of 1366 primary focal articles and 55,718 secondary references, we visualize the landscape and evolution of big-data business research and capture the developmental trajectory and trends over time (between 2008 and 2018). Furthermore, based on our analyses, we provide several promising directions for future research. In doing so, we provide scholars with a systematic understanding of the development and panoramic roadmap of big data research in business.

@article{pearson_boldly_2020,
	title = {To {Boldly} {Go} {Where} {No} {Data} {Stream} {Has} {Gone} {Before}},
	volume = {1},
	issn = {2666-3899},
	url = {http://www.sciencedirect.com/science/article/pii/S266638992030235X},
	doi = {10.1016/j.patter.2020.100171},
	abstract = {As humanity explores the Solar System, the further our spacecraft get from Earth the further their data signals have to travel. We look at some of the biggest obstacles that come up when attempting to transfer data billions of kilometers across space using a power- and weight-limited spacecraft.},
	language = {en},
	number = {9},
	urldate = {2020-12-12},
	journal = {Patterns},
	author = {Pearson, Ezzy},
	month = dec,
	year = {2020},
	pages = {100171},
}

As humanity explores the Solar System, the further our spacecraft get from Earth the further their data signals have to travel. We look at some of the biggest obstacles that come up when attempting to transfer data billions of kilometers across space using a power- and weight-limited spacecraft.

@article{korycki_adaptive_2020,
	title = {Adaptive {Deep} {Forest} for {Online} {Learning} from {Drifting} {Data} {Streams}},
	url = {http://arxiv.org/abs/2010.07340},
	abstract = {Learning from data streams is among the most vital fields of contemporary data mining. The online analysis of information coming from those potentially unbounded data sources allows for designing reactive up-to-date models capable of adjusting themselves to continuous flows of data. While a plethora of shallow methods have been proposed for simpler low-dimensional streaming problems, almost none of them addressed the issue of learning from complex contextual data, such as images or texts. The former is represented mainly by adaptive decision trees that have been proven to be very efficient in streaming scenarios. The latter has been predominantly addressed by offline deep learning. In this work, we attempt to bridge the gap between these two worlds and propose Adaptive Deep Forest (ADF) - a natural combination of the successful tree-based streaming classifiers with deep forest, which represents an interesting alternative idea for learning from contextual data. The conducted experiments show that the deep forest approach can be effectively transformed into an online algorithm, forming a model that outperforms all state-of-the-art shallow adaptive classifiers, especially for high-dimensional complex streams.},
	urldate = {2020-12-12},
	journal = {arXiv:2010.07340 [cs]},
	author = {Korycki, Łukasz and Krawczyk, Bartosz},
	month = oct,
	year = {2020},
	note = {arXiv: 2010.07340},
	keywords = {Computer Science - Machine Learning, I.2.0, I.5.0},
}

Learning from data streams is among the most vital fields of contemporary data mining. The online analysis of information coming from those potentially unbounded data sources allows for designing reactive up-to-date models capable of adjusting themselves to continuous flows of data. While a plethora of shallow methods have been proposed for simpler low-dimensional streaming problems, almost none of them addressed the issue of learning from complex contextual data, such as images or texts. The former is represented mainly by adaptive decision trees that have been proven to be very efficient in streaming scenarios. The latter has been predominantly addressed by offline deep learning. In this work, we attempt to bridge the gap between these two worlds and propose Adaptive Deep Forest (ADF) - a natural combination of the successful tree-based streaming classifiers with deep forest, which represents an interesting alternative idea for learning from contextual data. The conducted experiments show that the deep forest approach can be effectively transformed into an online algorithm, forming a model that outperforms all state-of-the-art shallow adaptive classifiers, especially for high-dimensional complex streams.

@article{pang_bayesian_2020,
	title = {A {Bayesian} {Inference} for {Remaining} {Useful} {Life} {Estimation} by {Fusing} {Accelerated} {Degradation} {Data} and {Condition} {Monitoring} {Data}},
	issn = {0951-8320},
	url = {http://www.sciencedirect.com/science/article/pii/S0951832020308334},
	doi = {10.1016/j.ress.2020.107341},
	abstract = {This article addresses the problem of estimating the remaining useful life (RUL) of degrading products by fusing the accelerated degradation data and condition monitoring (CM) data. The proposed model differs from the existing models in adopting the non-conjugate prior distributions for random-effect parameters. First, a nonlinear diffusion process model is developed to characterize the degradation process of a product. Next, the relationship between the model parameters and accelerated stress level is established, and the accelerated degradation data are used to determine the prior distribution types and estimate the hyperparameters of the model parameters. Then, to fuse the accelerated degradation data and CM data, the Bayesian inference is used to update the posterior distributions of model parameters once the new degradation observations are available. In addition, the Markov Chain Monte Carlo (MCMC) method based on Gibbs sampling is used to obtain the Bayesian solution numerically. Finally, the approximate RUL distribution considering the randomness of model parameters is obtained by the MCMC method based on the concept of the first hitting time. The proposed method is verified by the practical case study of accelerometers. Comparison results demonstrate that the proposed method can obtain higher RUL estimation accuracy and less uncertainty.},
	language = {en},
	urldate = {2020-12-08},
	journal = {Reliability Engineering \& System Safety},
	author = {Pang, ZHENAN and Si, XIAOSHENG and Hu, CHANGHUA and Du, DANGBO and Pei, HONG},
	month = dec,
	year = {2020},
	keywords = {Bayesian inference, accelerated degradation data, diffusion model, non-conjugate prior distribution, remaining useful life},
	pages = {107341},
}

This article addresses the problem of estimating the remaining useful life (RUL) of degrading products by fusing the accelerated degradation data and condition monitoring (CM) data. The proposed model differs from the existing models in adopting the non-conjugate prior distributions for random-effect parameters. First, a nonlinear diffusion process model is developed to characterize the degradation process of a product. Next, the relationship between the model parameters and accelerated stress level is established, and the accelerated degradation data are used to determine the prior distribution types and estimate the hyperparameters of the model parameters. Then, to fuse the accelerated degradation data and CM data, the Bayesian inference is used to update the posterior distributions of model parameters once the new degradation observations are available. In addition, the Markov Chain Monte Carlo (MCMC) method based on Gibbs sampling is used to obtain the Bayesian solution numerically. Finally, the approximate RUL distribution considering the randomness of model parameters is obtained by the MCMC method based on the concept of the first hitting time. The proposed method is verified by the practical case study of accelerometers. Comparison results demonstrate that the proposed method can obtain higher RUL estimation accuracy and less uncertainty.

@article{georgievskaia_predictive_2020,
	series = {1st {Virtual} {European} {Conference} on {Fracture} - {VECF1}},
	title = {Predictive analytics as a way to smart maintenance of hydraulic turbines},
	volume = {28},
	issn = {2452-3216},
	url = {http://www.sciencedirect.com/science/article/pii/S2452321620306004},
	doi = {10.1016/j.prostr.2020.10.098},
	abstract = {Today, most energy companies face serious problems with the reliability and safety of large power equipment due to the long-term operation at off-design modes. This is especially true for hydraulic units that are traditionally used to ensure the required level of energy output and maintain the stability of a power grid due to their maneuverability. Off-design operational modes cause increased loads and stresses in the unit’s components, accelerate the growth of defects, stimulate premature failures, and can lead to a grave accident with large losses. Standard diagnostic systems for hydraulic units usually do not allow tracking hazardous defects such as fatigue cracks in the runner blades, guide vines, shaft, etc. The individuality of hydraulic units also excludes the use of any statistical methods to determine the time when the equipment will go to the limit state and its operation will become unreasonably dangerous. Therefore, the predictive analytics system is proposed as an additional approach to forecasting the appearance and growth of dangerous operational defects. The system realizes an analytical algorithm based on evaluating the fatigue strength of hydraulic unit elements under variable operating conditions and allows summing up damage from various external loads and in different time ranges. Input data for this predictive system is information about the actual operating time at every working mode and expected regime parameters for the upcoming period. The output data is information about actual and residual lifetime. All individual features are taken into account by the digital model which is a multidimensional matrix of equipment’s response to external influences. Data is generated in the cells for each unit’s elements with reference to the operating parameters and time scale. The proposed predictive analytics system allows not only to reduce the risk of accidents and unplanned shutdowns but it also is a way to smart maintenance of hydraulic turbines due to the development of the most effective, most reasonable, most lifetime-saving strategy of using the equipment.},
	language = {en},
	urldate = {2020-12-08},
	journal = {Procedia Structural Integrity},
	author = {Georgievskaia, Evgeniia},
	month = jan,
	year = {2020},
	keywords = {crack, failure, hydraulic turbines, lifetime, predictive analytics, reliability, smart maintenance},
	pages = {836--842},
}

Today, most energy companies face serious problems with the reliability and safety of large power equipment due to the long-term operation at off-design modes. This is especially true for hydraulic units that are traditionally used to ensure the required level of energy output and maintain the stability of a power grid due to their maneuverability. Off-design operational modes cause increased loads and stresses in the unit’s components, accelerate the growth of defects, stimulate premature failures, and can lead to a grave accident with large losses. Standard diagnostic systems for hydraulic units usually do not allow tracking hazardous defects such as fatigue cracks in the runner blades, guide vines, shaft, etc. The individuality of hydraulic units also excludes the use of any statistical methods to determine the time when the equipment will go to the limit state and its operation will become unreasonably dangerous. Therefore, the predictive analytics system is proposed as an additional approach to forecasting the appearance and growth of dangerous operational defects. The system realizes an analytical algorithm based on evaluating the fatigue strength of hydraulic unit elements under variable operating conditions and allows summing up damage from various external loads and in different time ranges. Input data for this predictive system is information about the actual operating time at every working mode and expected regime parameters for the upcoming period. The output data is information about actual and residual lifetime. All individual features are taken into account by the digital model which is a multidimensional matrix of equipment’s response to external influences. Data is generated in the cells for each unit’s elements with reference to the operating parameters and time scale. The proposed predictive analytics system allows not only to reduce the risk of accidents and unplanned shutdowns but it also is a way to smart maintenance of hydraulic turbines due to the development of the most effective, most reasonable, most lifetime-saving strategy of using the equipment.

@article{subramanian_white-box_2020,
	title = {White-box {Machine} learning approaches to identify governing equations for overall dynamics of manufacturing systems: {A} case study on distillation column},
	issn = {2666-8270},
	shorttitle = {White-box {Machine} learning approaches to identify governing equations for overall dynamics of manufacturing systems},
	url = {http://www.sciencedirect.com/science/article/pii/S2666827020300141},
	doi = {10.1016/j.mlwa.2020.100014},
	abstract = {Dynamical equations form the basis of design for manufacturing processes and control systems, however, identifying governing equations using mechanistic approach is tedious. Recently, Machine learning (ML) has shown promise to identify the governing dynamical equations for physical systems faster. This possibility of rapid identification of governing equations provides an exciting opportunity for advancing dynamical systems modeling. However, applicability of ML approach identifying governing mechanisms for dynamics of complex systems relevant to manufacturins systems has not been tested. We test and compare the efficacy of two white-box ML (SINDy and SymReg) approaches for predicting dynamics and structure of dynamical equations for overall dynamics in distillation column. Results demonstrate that a combination of ML approach should be used to identify full range of equations. In terms of physical law, few terms were interpretable as related to Fick’s law of diffusion and Henry’s law in SINDy whereas SymReg identified energy balance as driving dynamics.},
	language = {en},
	urldate = {2020-12-04},
	journal = {Machine Learning with Applications},
	author = {Subramanian, Renganathan and Moar, Raghav Rajesh and Singh, Shweta},
	month = dec,
	year = {2020},
	keywords = {ASPEN dynamics, Distillation column, Dynamic equation, Genetic programming, Machine learning, SINDy, Symbolic regression},
	pages = {100014},
}

Dynamical equations form the basis of design for manufacturing processes and control systems, however, identifying governing equations using mechanistic approach is tedious. Recently, Machine learning (ML) has shown promise to identify the governing dynamical equations for physical systems faster. This possibility of rapid identification of governing equations provides an exciting opportunity for advancing dynamical systems modeling. However, applicability of ML approach identifying governing mechanisms for dynamics of complex systems relevant to manufacturins systems has not been tested. We test and compare the efficacy of two white-box ML (SINDy and SymReg) approaches for predicting dynamics and structure of dynamical equations for overall dynamics in distillation column. Results demonstrate that a combination of ML approach should be used to identify full range of equations. In terms of physical law, few terms were interpretable as related to Fick’s law of diffusion and Henry’s law in SINDy whereas SymReg identified energy balance as driving dynamics.

@article{quintanar-gago_assessment_2020,
	title = {Assessment of steam turbine blade failure and damage mechanisms using a {Bayesian} network},
	issn = {0951-8320},
	url = {http://www.sciencedirect.com/science/article/pii/S095183202030822X},
	doi = {10.1016/j.ress.2020.107329},
	abstract = {Damage mechanisms that affect components within complex machines are often hard to detect and identify, especially if they are difficult to access, inspect and/or that are under continuous duty, compromising the reliability and performance of systems. In this paper, a Bayesian network model is developed to handle the interactions among common damage mechanisms and failure modes in nuclear steam turbine rotating blades. This model enables maintenance and inspection planning to better predict which portions(s) of the turbine will need repair. To compute the conditional probability tables, the model's unique quantification method combines expert judgement, the Recursive Noisy OR, and a damage mechanism susceptibility ranking that takes into account the synergistic interactions of the damage mechanisms. The approach can be suited to different turbine designs and purposes. The Bayesian network model development is described in detail, validated, and several examples of its application are presented.},
	language = {en},
	urldate = {2020-11-30},
	journal = {Reliability Engineering \& System Safety},
	author = {Quintanar-Gago, David A. and Nelson, Pamela F. and Díaz-Sánchez, Ángeles and Boldrick, Michael S.},
	month = nov,
	year = {2020},
	keywords = {Bayesian Network, Damage Mechanism, Maintenance, Recursive Noisy OR, Steam Turbine Blade},
	pages = {107329},
}

Damage mechanisms that affect components within complex machines are often hard to detect and identify, especially if they are difficult to access, inspect and/or that are under continuous duty, compromising the reliability and performance of systems. In this paper, a Bayesian network model is developed to handle the interactions among common damage mechanisms and failure modes in nuclear steam turbine rotating blades. This model enables maintenance and inspection planning to better predict which portions(s) of the turbine will need repair. To compute the conditional probability tables, the model's unique quantification method combines expert judgement, the Recursive Noisy OR, and a damage mechanism susceptibility ranking that takes into account the synergistic interactions of the damage mechanisms. The approach can be suited to different turbine designs and purposes. The Bayesian network model development is described in detail, validated, and several examples of its application are presented.

@article{teixeira_condition-based_2020,
	series = {30th {International} {Conference} on {Flexible} {Automation} and {Intelligent} {Manufacturing} ({FAIM2021})},
	title = {Condition-based maintenance implementation: a literature review},
	volume = {51},
	issn = {2351-9789},
	shorttitle = {Condition-based maintenance implementation},
	url = {http://www.sciencedirect.com/science/article/pii/S2351978920318886},
	doi = {10.1016/j.promfg.2020.10.033},
	abstract = {Industrial companies are increasingly dependent on the availability and performance of their equipment to remain competitive. This circumstance demands accurate and timely maintenance actions in alignment with the organizational objectives. Condition-Based Maintenance (CBM) is a strategy that considers information about the equipment condition to recommend appropriate maintenance actions. The main purpose of CBM is to prevent functional failures or a significant performance decrease of the monitored equipment. CBM relies on a wide range of resources and techniques required to detect deviations from the normal operating conditions, diagnose incipient failures or predict the future condition of an asset. To obtain meaningful information for maintenance decision making, relevant data must be collected and properly analyzed. Recent advances in Big Data analytics and Internet of Things (IoT) enable real-time decision making based on abundant data acquired from several different sources. However, each appliance must be designed according to the equipment configuration and considering the nature of specific failure modes. CBM implementation is a complex matter, regardless of the equipment characteristics. Therefore, to ensure cost-effectiveness, it must be addressed in a systematic and organized manner, considering the technical and financial issues involved. This paper presents a literature review on approaches to support CBM implementation. Published studies and standards that provide guidelines to implement CBM are analyzed and compared. For each existing approach, the steps recommended to implement CBM are listed and the main gaps are identified. Based on the literature, factors that can affect the effective implementation of CBM are also highlighted and discussed.},
	language = {en},
	urldate = {2020-11-23},
	journal = {Procedia Manufacturing},
	author = {Teixeira, Humberto Nuno and Lopes, Isabel and Braga, Ana Cristina},
	month = jan,
	year = {2020},
	keywords = {Condition monitoring (CM), Condition-Based Maintenance (CBM), Health Management (PHM), Prognostics},
	pages = {228--235},
}

Industrial companies are increasingly dependent on the availability and performance of their equipment to remain competitive. This circumstance demands accurate and timely maintenance actions in alignment with the organizational objectives. Condition-Based Maintenance (CBM) is a strategy that considers information about the equipment condition to recommend appropriate maintenance actions. The main purpose of CBM is to prevent functional failures or a significant performance decrease of the monitored equipment. CBM relies on a wide range of resources and techniques required to detect deviations from the normal operating conditions, diagnose incipient failures or predict the future condition of an asset. To obtain meaningful information for maintenance decision making, relevant data must be collected and properly analyzed. Recent advances in Big Data analytics and Internet of Things (IoT) enable real-time decision making based on abundant data acquired from several different sources. However, each appliance must be designed according to the equipment configuration and considering the nature of specific failure modes. CBM implementation is a complex matter, regardless of the equipment characteristics. Therefore, to ensure cost-effectiveness, it must be addressed in a systematic and organized manner, considering the technical and financial issues involved. This paper presents a literature review on approaches to support CBM implementation. Published studies and standards that provide guidelines to implement CBM are analyzed and compared. For each existing approach, the steps recommended to implement CBM are listed and the main gaps are identified. Based on the literature, factors that can affect the effective implementation of CBM are also highlighted and discussed.

@article{nikolakis_microservice_2020,
	series = {30th {International} {Conference} on {Flexible} {Automation} and {Intelligent} {Manufacturing} ({FAIM2021})},
	title = {A microservice architecture for predictive analytics in manufacturing},
	volume = {51},
	issn = {2351-9789},
	url = {http://www.sciencedirect.com/science/article/pii/S2351978920320102},
	doi = {10.1016/j.promfg.2020.10.153},
	abstract = {This paper discusses on the design, development and deployment of a flexible and modular platform supporting smart predictive maintenance operations, enabled by microservices architecture and virtualization technologies. Virtualization allows the platform to be deployed in a multi-tenant environment, while facilitating resource isolation and independency from specific technologies or services. Moreover, the proposed platform supports scalable data storage supporting an effective and efficient management of large volume of Industry 4.0 data. Methodologies of data-driven predictive maintenance are provided to the user as-a-service, facilitating offline training and online execution of pre-trained analytics models, while the connection of the raw data to contextual information support their understanding and interpretation, while guaranteeing interoperability across heterogeneous systems. A use case related to the predictive maintenance operations of a robotic manipulator is examined to demonstrate the effectiveness and the efficiency of the proposed platform.},
	language = {en},
	urldate = {2020-11-23},
	journal = {Procedia Manufacturing},
	author = {Nikolakis, N. and Marguglio, A. and Veneziano, G. and Greco, P. and Panicucci, S. and Cerquitelli, T. and Macii, E. and Andolina, S. and Alexopoulos, K.},
	month = jan,
	year = {2020},
	keywords = {Machine learning, Microservice architecture, Play approach, Plug, Robotics industry, Service-oriented platform},
	pages = {1091--1097},
}

This paper discusses on the design, development and deployment of a flexible and modular platform supporting smart predictive maintenance operations, enabled by microservices architecture and virtualization technologies. Virtualization allows the platform to be deployed in a multi-tenant environment, while facilitating resource isolation and independency from specific technologies or services. Moreover, the proposed platform supports scalable data storage supporting an effective and efficient management of large volume of Industry 4.0 data. Methodologies of data-driven predictive maintenance are provided to the user as-a-service, facilitating offline training and online execution of pre-trained analytics models, while the connection of the raw data to contextual information support their understanding and interpretation, while guaranteeing interoperability across heterogeneous systems. A use case related to the predictive maintenance operations of a robotic manipulator is examined to demonstrate the effectiveness and the efficiency of the proposed platform.

@article{ferreira_novel_2020,
	series = {30th {International} {Conference} on {Flexible} {Automation} and {Intelligent} {Manufacturing} ({FAIM2021})},
	title = {A novel approach to improve maintenance operations},
	volume = {51},
	issn = {2351-9789},
	url = {http://www.sciencedirect.com/science/article/pii/S2351978920320849},
	doi = {10.1016/j.promfg.2020.10.213},
	abstract = {In a market constantly changing and more demanding, the need to have any production cycle working without flaws is more important than ever. Thus, maintenance operations have taken an increasingly important role. However, different organizations may take different approaches depending on their area of business, legal aspects, as well as the company policy. Even though some concepts such as Total Productive Maintenance (TPM) and quality tools are normally used in most of the companies, there are no standard procedures or defined ways for using them across industries. Thus, it is usual many organizations implement statistical tests in order to see whether or not those improvements were successful. This paper intends to establish a procedure to be used across organizations, improving their maintenance procedures. This procedure establishes the correct sequence of tools to be used in the improvement process, conducting the maintenance operations to a higher level of efficiency and effectiveness, allowing to determine if the implemented improvements were successful, or if a different approach needs to be implemented.},
	language = {en},
	urldate = {2020-11-23},
	journal = {Procedia Manufacturing},
	author = {Ferreira, S. and Martins, L. and Silva, F. J. G. and Casais, R. B. and Campilho, R. D. S. G. and Sá, J. C.},
	month = jan,
	year = {2020},
	keywords = {Maintenance, Maintenance efficiency, Maintenance improvement, Maintenance tools, TPM, Total Productive Maintenance},
	pages = {1531--1537},
}

In a market constantly changing and more demanding, the need to have any production cycle working without flaws is more important than ever. Thus, maintenance operations have taken an increasingly important role. However, different organizations may take different approaches depending on their area of business, legal aspects, as well as the company policy. Even though some concepts such as Total Productive Maintenance (TPM) and quality tools are normally used in most of the companies, there are no standard procedures or defined ways for using them across industries. Thus, it is usual many organizations implement statistical tests in order to see whether or not those improvements were successful. This paper intends to establish a procedure to be used across organizations, improving their maintenance procedures. This procedure establishes the correct sequence of tools to be used in the improvement process, conducting the maintenance operations to a higher level of efficiency and effectiveness, allowing to determine if the implemented improvements were successful, or if a different approach needs to be implemented.

@article{liu_enhanced_2020,
	title = {An enhanced encoder–decoder framework for bearing remaining useful life prediction},
	issn = {0263-2241},
	url = {http://www.sciencedirect.com/science/article/pii/S0263224120312537},
	doi = {10.1016/j.measurement.2020.108753},
	abstract = {In recent years, data-driven approaches for remaining useful life (RUL) prognostics have aroused widespread concern. Bearings act as the fundamental component of machinery and their conditioning status is closely associated with the normal operation of equipment. Hence, it is crucial to accurately predict the remaining useful life of bearings. This paper explores the degradation process of bearings and proposes an enhanced encoder–decoder framework. The framework attempts to construct a decoder with the ability to look back and selectively mine underlying information in the encoder. Additionally, trigonometric functions and cumulative operation are employed to enhance the quality of health indicators. To verify the effectiveness of the proposed method, vibration data from PRONOSTIA platform are utilized for RUL prognostics. Compared with several state-of-the-art methods, the experimental results demonstrate the superiority and feasibility of the proposed method.},
	language = {en},
	urldate = {2020-11-23},
	journal = {Measurement},
	author = {Liu, Lu and Song, Xiao and Chen, Kai and Hou, Baocun and Chai, Xudong and Ning, Huansheng},
	month = nov,
	year = {2020},
	keywords = {Cumulative operation, Encoder–decoder, Fitness analysis, Trigonometric functions},
	pages = {108753},
}

In recent years, data-driven approaches for remaining useful life (RUL) prognostics have aroused widespread concern. Bearings act as the fundamental component of machinery and their conditioning status is closely associated with the normal operation of equipment. Hence, it is crucial to accurately predict the remaining useful life of bearings. This paper explores the degradation process of bearings and proposes an enhanced encoder–decoder framework. The framework attempts to construct a decoder with the ability to look back and selectively mine underlying information in the encoder. Additionally, trigonometric functions and cumulative operation are employed to enhance the quality of health indicators. To verify the effectiveness of the proposed method, vibration data from PRONOSTIA platform are utilized for RUL prognostics. Compared with several state-of-the-art methods, the experimental results demonstrate the superiority and feasibility of the proposed method.

@article{altendeitering_scalable_2020,
	series = {30th {International} {Conference} on {Flexible} {Automation} and {Intelligent} {Manufacturing} ({FAIM2021})},
	title = {Scalable {Detection} of {Concept} {Drift}: {A} {Learning} {Technique} {Based} on {Support} {Vector} {Machines}},
	volume = {51},
	issn = {2351-9789},
	shorttitle = {Scalable {Detection} of {Concept} {Drift}},
	url = {http://www.sciencedirect.com/science/article/pii/S2351978920319120},
	doi = {10.1016/j.promfg.2020.10.057},
	abstract = {The issue of concept drift describes how static machine-learning models build on historical data can become unreliable over time and pose a significant challenge to many applications. Although, there is a growing body of literature investigating concept drift existing solutions are often limited to a small number of samples or features and do not work well in Industry 4.0 scenarios. We are proposing a novel algorithm that extends the existing concept drift algorithm FLORA3 by utilizing support vector machines for the classification process. Through this combination of dynamic and static approaches the algorithm is capable of effectively analyzing data streams of high volume. For evaluation, we tested our algorithm on the publicly available data set ‘elec2’, which is based on the energy market in Australia. Our results show that the proposed algorithm needs less computational resources compared to other algorithms while maintaining a high level of accuracy.},
	language = {en},
	urldate = {2020-11-23},
	journal = {Procedia Manufacturing},
	author = {Altendeitering, Marcel and Dübler, Stephan},
	month = jan,
	year = {2020},
	keywords = {FLORA3, SVM, concept drift, energy data, machine learning},
	pages = {400--407},
}

The issue of concept drift describes how static machine-learning models build on historical data can become unreliable over time and pose a significant challenge to many applications. Although, there is a growing body of literature investigating concept drift existing solutions are often limited to a small number of samples or features and do not work well in Industry 4.0 scenarios. We are proposing a novel algorithm that extends the existing concept drift algorithm FLORA3 by utilizing support vector machines for the classification process. Through this combination of dynamic and static approaches the algorithm is capable of effectively analyzing data streams of high volume. For evaluation, we tested our algorithm on the publicly available data set ‘elec2’, which is based on the energy market in Australia. Our results show that the proposed algorithm needs less computational resources compared to other algorithms while maintaining a high level of accuracy.

@article{castellanos_accordant_2020,
	title = {{ACCORDANT}: {A} domain specific model and {DevOps} approach for big data analytics architectures},
	issn = {0164-1212},
	shorttitle = {{ACCORDANT}},
	url = {http://www.sciencedirect.com/science/article/pii/S0164121220302594},
	doi = {10.1016/j.jss.2020.110869},
	abstract = {Big data analytics (BDA) applications use machine learning algorithms to extract valuable insights from large, fast, and heterogeneous data sources. New software engineering challenges for BDA applications include ensuring performance levels of data-driven algorithms even in the presence of large data volume, velocity, and variety (3Vs). BDA software complexity frequently leads to delayed deployments, longer development cycles and challenging performance assessment. This paper proposes a Domain-Specific Model (DSM), and DevOps practices to design, deploy, and monitor performance metrics in BDA applications. Our proposal includes a design process, and a framework to define architectural inputs, software components, and deployment strategies through integrated high-level abstractions to enable QS monitoring. We evaluate our approach with four use cases from different domains to demonstrate a high level of generalization. Our results show a shorter deployment and monitoring times, and a higher gain factor per iteration compared to similar approaches.},
	language = {en},
	urldate = {2020-11-21},
	journal = {Journal of Systems and Software},
	author = {Castellanos, Camilo and Varela, Carlos A. and Correal, Dario},
	month = nov,
	year = {2020},
	keywords = {Big data analytics deployment, DevOps, Domain specific model, Performance monitoring, Quality scenarios, Software architecture},
	pages = {110869},
}

Big data analytics (BDA) applications use machine learning algorithms to extract valuable insights from large, fast, and heterogeneous data sources. New software engineering challenges for BDA applications include ensuring performance levels of data-driven algorithms even in the presence of large data volume, velocity, and variety (3Vs). BDA software complexity frequently leads to delayed deployments, longer development cycles and challenging performance assessment. This paper proposes a Domain-Specific Model (DSM), and DevOps practices to design, deploy, and monitor performance metrics in BDA applications. Our proposal includes a design process, and a framework to define architectural inputs, software components, and deployment strategies through integrated high-level abstractions to enable QS monitoring. We evaluate our approach with four use cases from different domains to demonstrate a high level of generalization. Our results show a shorter deployment and monitoring times, and a higher gain factor per iteration compared to similar approaches.

@article{fila_cloud_2020,
	series = {The 11th {International} {Conference} on {Emerging} {Ubiquitous} {Systems} and {Pervasive} {Networks} ({EUSPN} 2020) / {The} 10th {International} {Conference} on {Current} and {Future} {Trends} of {Information} and {Communication} {Technologies} in {Healthcare} ({ICTH} 2020) / {Affiliated} {Workshops}},
	title = {Cloud {Computing} for {Industrial} {Predictive} {Maintenance} {Based} on {Prognostics} and {Health} {Management}},
	volume = {177},
	issn = {1877-0509},
	url = {http://www.sciencedirect.com/science/article/pii/S1877050920323619},
	doi = {10.1016/j.procs.2020.10.090},
	abstract = {Predictive maintenance is based primarily on Prognostics and Health Management (PHM). The prognosis is a process for learning about the health status of a system and estimating its residual time before failure. A good maintenance decision is the result of a better estimate of the latter. Recently, the emergence of IT systems in the industrial field and in particular connected objects and cloud computing have contributed strongly to the improvement of the prognosis process. In this paper, we propose a new prognosis approach based on the Cloud Computing model and the principle of multitenancy in order to present the Prognosis as a Service. This approach provides an effective prognosis solution at the request of a client while ensuring a better quality of service. The effectiveness of our solution depends on the criteria for the performance of the prognosis system based on accuracy, accuracy, mean squared error and a Quality of Service (Qos).},
	language = {en},
	urldate = {2020-11-16},
	journal = {Procedia Computer Science},
	author = {Fila, Redouane and Khaili, Mohamed El and Mestari, Mohamed},
	month = jan,
	year = {2020},
	keywords = {Cloud Computing, Health Management (PHM), Performance Measurement, Predictive Maintenance, Prognosis as a Service, Prognostics, Quality of Service (Qos), Residual Life (RUL)},
	pages = {631--638},
}

Predictive maintenance is based primarily on Prognostics and Health Management (PHM). The prognosis is a process for learning about the health status of a system and estimating its residual time before failure. A good maintenance decision is the result of a better estimate of the latter. Recently, the emergence of IT systems in the industrial field and in particular connected objects and cloud computing have contributed strongly to the improvement of the prognosis process. In this paper, we propose a new prognosis approach based on the Cloud Computing model and the principle of multitenancy in order to present the Prognosis as a Service. This approach provides an effective prognosis solution at the request of a client while ensuring a better quality of service. The effectiveness of our solution depends on the criteria for the performance of the prognosis system based on accuracy, accuracy, mean squared error and a Quality of Service (Qos).

@article{kimera_improving_2020,
	title = {Improving ship yard ballast pumps’ operations: {A} {PCA} approach to predictive maintenance},
	volume = {1},
	issn = {2666-822X},
	shorttitle = {Improving ship yard ballast pumps’ operations},
	url = {http://www.sciencedirect.com/science/article/pii/S2666822X20300034},
	doi = {10.1016/j.martra.2020.100003},
	abstract = {This paper investigates a predictive maintenance approach for marine mechanical systems via an early warning system. A machine learning methodology was used to process and analyze the dock pump back pressure, flow rate, amperage and suction pressure data. Operating parameters for a dock pump were monitored for 40 weeks and the values were manually input into the tool. Unsupervised machine learning was used in order to draw inferences from data via MATLAB. A principal component analysis (PCA) algorithm was used to improve on the selection of the key operating parameters of the dock pumps. The dock pump flow rate and suction pressure, were the principal components that were 99.707\% sufficient to explain the variation in the data. Using the dataset explained by the PCA, two data classes were later used in the SVM algorithm for a binary classification approach. The developed tool predicted that the dock pump may fail/requires maintenance between seventh and eighth weeks. This prediction deviated from the actual ten weeks that it took the dock pump to fail. A prediction deviation from the actual failure time to failure could be attributed to the quality of the historical failure and maintenance data. Nevertheless, with less ambiguity of the data, the maintenance prediction tool can be used as a basis before sensor technology on the dock pumps is implemented.},
	language = {en},
	urldate = {2020-11-16},
	journal = {Maritime Transport Research},
	author = {Kimera, David and Nangolo, Filemon N.},
	month = jan,
	year = {2020},
	keywords = {Ballast pumps, Floating docks, Machine learning, Predictive maintenance, Principle component analysis},
	pages = {100003},
}

This paper investigates a predictive maintenance approach for marine mechanical systems via an early warning system. A machine learning methodology was used to process and analyze the dock pump back pressure, flow rate, amperage and suction pressure data. Operating parameters for a dock pump were monitored for 40 weeks and the values were manually input into the tool. Unsupervised machine learning was used in order to draw inferences from data via MATLAB. A principal component analysis (PCA) algorithm was used to improve on the selection of the key operating parameters of the dock pumps. The dock pump flow rate and suction pressure, were the principal components that were 99.707% sufficient to explain the variation in the data. Using the dataset explained by the PCA, two data classes were later used in the SVM algorithm for a binary classification approach. The developed tool predicted that the dock pump may fail/requires maintenance between seventh and eighth weeks. This prediction deviated from the actual ten weeks that it took the dock pump to fail. A prediction deviation from the actual failure time to failure could be attributed to the quality of the historical failure and maintenance data. Nevertheless, with less ambiguity of the data, the maintenance prediction tool can be used as a basis before sensor technology on the dock pumps is implemented.

@article{geetha_overview_2020,
	title = {Overview of machine learning and its adaptability in mechanical engineering},
	issn = {2214-7853},
	url = {http://www.sciencedirect.com/science/article/pii/S2214785320373363},
	doi = {10.1016/j.matpr.2020.09.611},
	abstract = {Artificial Intelligence unto a mechanical engineer is to take the system to a next level to develop a better device. To better understand the physical phenomenon, engineers design the products that we all interact with. Machine learning (ML) plays a central role in translating that technology to make the world a better place. Many of the tools in ML are embedded with these methods and techniques to design machines which are interactive. The present study is an overview of applications of ML in Mechanical Engineering.},
	language = {en},
	urldate = {2020-11-13},
	journal = {Materials Today: Proceedings},
	author = {Geetha, N. K. and Bridjesh, P.},
	month = nov,
	year = {2020},
	keywords = {Algorithm, Applications, Artificial intelligence, Machine learning, Mechanical engineering},
}

Artificial Intelligence unto a mechanical engineer is to take the system to a next level to develop a better device. To better understand the physical phenomenon, engineers design the products that we all interact with. Machine learning (ML) plays a central role in translating that technology to make the world a better place. Many of the tools in ML are embedded with these methods and techniques to design machines which are interactive. The present study is an overview of applications of ML in Mechanical Engineering.

@article{luong_heterogeneous_2020,
	title = {Heterogeneous ensemble selection for evolving data streams},
	issn = {0031-3203},
	url = {http://www.sciencedirect.com/science/article/pii/S003132032030546X},
	doi = {10.1016/j.patcog.2020.107743},
	abstract = {Ensemble learning has been widely applied to both batch data classification and streaming data classification. For the latter setting, most existing ensemble systems are homogenous, which means they are generated from only one type of learning model. In contrast, by combining several types of different learning models, a heterogeneous ensemble system can achieve greater diversity among its members, which helps to improve its performance. Although heterogeneous ensemble systems have achieved many successes in the batch classification setting, it is not trivial to extend them directly to the data stream setting. In this study, we propose a novel HEterogeneous Ensemble Selection (HEES) method, which dynamically selects an appropriate subset of base classifiers to predict data under the stream setting. We are inspired by the observation that a well-chosen subset of good base classifiers may outperform the whole ensemble system. Here, we define a good candidate as one that expresses not only high predictive performance but also high confidence in its prediction. Our selection process is thus divided into two sub-processes: accurate-candidate selection and confident-candidate selection. We define an accurate candidate in the stream context as a base classifier with high accuracy over the current concept, while a confident candidate as one with a confidence score higher than a certain threshold. In the first sub-process, we employ the prequential accuracy to estimate the performance of a base classifier at a specific time, while in the latter sub-process, we propose a new measure to quantify the predictive confidence and provide a method to learn the threshold incrementally. The final ensemble is formed by taking the intersection of the sets of confident classifiers and accurate classifiers. Experiments on a wide range of data streams show that the proposed method achieves competitive performance with lower running time in comparison to the state-of-the-art online ensemble methods.},
	language = {en},
	urldate = {2020-11-13},
	journal = {Pattern Recognition},
	author = {Luong, Anh Vu and Nguyen, Tien Thanh and Liew, Alan Wee-Chung and Wang, Shilin},
	month = nov,
	year = {2020},
	keywords = {Data streams, Ensemble selection, Heterogeneous ensembles},
	pages = {107743},
}

Ensemble learning has been widely applied to both batch data classification and streaming data classification. For the latter setting, most existing ensemble systems are homogenous, which means they are generated from only one type of learning model. In contrast, by combining several types of different learning models, a heterogeneous ensemble system can achieve greater diversity among its members, which helps to improve its performance. Although heterogeneous ensemble systems have achieved many successes in the batch classification setting, it is not trivial to extend them directly to the data stream setting. In this study, we propose a novel HEterogeneous Ensemble Selection (HEES) method, which dynamically selects an appropriate subset of base classifiers to predict data under the stream setting. We are inspired by the observation that a well-chosen subset of good base classifiers may outperform the whole ensemble system. Here, we define a good candidate as one that expresses not only high predictive performance but also high confidence in its prediction. Our selection process is thus divided into two sub-processes: accurate-candidate selection and confident-candidate selection. We define an accurate candidate in the stream context as a base classifier with high accuracy over the current concept, while a confident candidate as one with a confidence score higher than a certain threshold. In the first sub-process, we employ the prequential accuracy to estimate the performance of a base classifier at a specific time, while in the latter sub-process, we propose a new measure to quantify the predictive confidence and provide a method to learn the threshold incrementally. The final ensemble is formed by taking the intersection of the sets of confident classifiers and accurate classifiers. Experiments on a wide range of data streams show that the proposed method achieves competitive performance with lower running time in comparison to the state-of-the-art online ensemble methods.

@article{cakir_experimental_2020,
	title = {The experimental application of popular machine learning algorithms on predictive maintenance and the design of {IIoT} based condition monitoring system},
	issn = {0360-8352},
	url = {http://www.sciencedirect.com/science/article/pii/S0360835220306252},
	doi = {10.1016/j.cie.2020.106948},
	abstract = {With the fourth industrial revolution, which has become increasingly widespread in the manufacturing industry, traditional maintenance has been replaced by the industrial internet of things (IIoT) based on condition monitoring system (CMS). The IIoT concept provides easier and reliable maintenance. Unlike traditional maintenance, IIoT systems that perform real-time monitoring can provide great advantages to the company by notifying the related maintenance team members of the factory before a serious failure occurs. It is very important to detect faulty bearings before they reach the critical level during the rotation. In this study, an industry 4.0 compatible, IIoT based and low-cost CMS was created and it consists of three main parts. Firstly experimental setup, secondly IIoT based condition monitoring application (CMA) and finally machine learning (ML) models and their evaluation. The experimental setup contains mechanical and electronic materials. Although the most common method used in the classification of bearing damage is vibration data, it observed that characteristics such as sound level, current, rotational speed, and temperature should be included in the data set in order to increase the success of the classification. All these data were collected from the setup, which is 6203 type bearing connected to the universal motor shaft. The designed CMA provides real-time monitoring and recording of the data, which comes wirelessly from the setup, on a mobile device that has an Android operating system. The CMA can also send SMS and e-mail notifications to maintenance team supervisors over mobile devices in case critical thresholds are exceeded. Lastly, the data collected from the experimental setup was modeled for classification with popular ML algorithms such as support vector machine (SVM), linear discrimination analysis (LDA), random forest (RF), decision tree (DT), and k-nearest neighbor (kNN). The models were evaluated with accuracy, precision, TPR, TNR, FPR, FNR, F1 score and, Kappa metrics. During the evaluation of all models, it was observed that with the increase in the number of features in the data set, the accuracy, sensitivity, TPR, TNR, F1 score and Kappa metrics increased above 99\% at 95\% confidence interval, and FPR and FNR metrics fell below 1\%. Although ML models gave successful results, LDA and DT models gave results much faster than others did. On the other hand, the classification success of the LDA model is relatively low. However, DT model is the optimum choice for CMS due to its convenience in determining threshold values, and its ability to give fast and acceptable classification rates.},
	language = {en},
	urldate = {2020-11-03},
	journal = {Computers \& Industrial Engineering},
	author = {Cakir, Mustafa and Guvenc, Mehmet Ali and Mistikoglu, Selcuk},
	month = oct,
	year = {2020},
	keywords = {Condition monitoring, Industry 4.0, Internet of things, Machine learning, Predictive maintenance},
	pages = {106948},
}

With the fourth industrial revolution, which has become increasingly widespread in the manufacturing industry, traditional maintenance has been replaced by the industrial internet of things (IIoT) based on condition monitoring system (CMS). The IIoT concept provides easier and reliable maintenance. Unlike traditional maintenance, IIoT systems that perform real-time monitoring can provide great advantages to the company by notifying the related maintenance team members of the factory before a serious failure occurs. It is very important to detect faulty bearings before they reach the critical level during the rotation. In this study, an industry 4.0 compatible, IIoT based and low-cost CMS was created and it consists of three main parts. Firstly experimental setup, secondly IIoT based condition monitoring application (CMA) and finally machine learning (ML) models and their evaluation. The experimental setup contains mechanical and electronic materials. Although the most common method used in the classification of bearing damage is vibration data, it observed that characteristics such as sound level, current, rotational speed, and temperature should be included in the data set in order to increase the success of the classification. All these data were collected from the setup, which is 6203 type bearing connected to the universal motor shaft. The designed CMA provides real-time monitoring and recording of the data, which comes wirelessly from the setup, on a mobile device that has an Android operating system. The CMA can also send SMS and e-mail notifications to maintenance team supervisors over mobile devices in case critical thresholds are exceeded. Lastly, the data collected from the experimental setup was modeled for classification with popular ML algorithms such as support vector machine (SVM), linear discrimination analysis (LDA), random forest (RF), decision tree (DT), and k-nearest neighbor (kNN). The models were evaluated with accuracy, precision, TPR, TNR, FPR, FNR, F1 score and, Kappa metrics. During the evaluation of all models, it was observed that with the increase in the number of features in the data set, the accuracy, sensitivity, TPR, TNR, F1 score and Kappa metrics increased above 99% at 95% confidence interval, and FPR and FNR metrics fell below 1%. Although ML models gave successful results, LDA and DT models gave results much faster than others did. On the other hand, the classification success of the LDA model is relatively low. However, DT model is the optimum choice for CMS due to its convenience in determining threshold values, and its ability to give fast and acceptable classification rates.

@article{li_remaining_2020,
	title = {Remaining {Useful} {Life} {Prediction} based on a {Multi}-{Sensor} {Data} {Fusion} {Model}},
	issn = {0951-8320},
	url = {http://www.sciencedirect.com/science/article/pii/S0951832020307493},
	doi = {10.1016/j.ress.2020.107249},
	abstract = {With the rapid development of Industrial Internet of Things, more and more sensors have been used for condition monitoring and prognostics of industrial systems. Big data collected from sensor networks bring abundant information resources as well as technical challenges for remaining useful life (RUL) prediction. The major technical challenges include how to select informative sensors and fuse multi-sensor data to improve the prediction performance. To deal with the challenges, this paper proposes a RUL prediction method based on a multi-sensor data fusion model. In this method, the inherent degradation process of the system state is expressed using a state transition function following a Wiener process. Multi-sensor signals are explicated as various proxies of the inherent system degradation process using a multivariate measurement function. The system state is estimated by fusing multi-sensor signals using particle filtering. Informative sensors are selected by a prioritized sensor group selection algorithm. This algorithm first prioritizes sensors according to their individual performances in RUL prediction, and then selects an optimal sensor group based on their combined performances. The effectiveness of the proposed method is demonstrated using a simulation study and aircraft engine degradation data from NASA repository.},
	language = {en},
	urldate = {2020-10-26},
	journal = {Reliability Engineering \& System Safety},
	author = {Li, Naipeng and Gebraeel, Nagi and Lei, Yaguo and Fang, Xiaolei and Cai, Xiao and Yan, Tao},
	month = oct,
	year = {2020},
	keywords = {Prognostic degradation modeling, big data, multi-sensor fusion, remaining useful life prediction, state-space model},
	pages = {107249},
}

With the rapid development of Industrial Internet of Things, more and more sensors have been used for condition monitoring and prognostics of industrial systems. Big data collected from sensor networks bring abundant information resources as well as technical challenges for remaining useful life (RUL) prediction. The major technical challenges include how to select informative sensors and fuse multi-sensor data to improve the prediction performance. To deal with the challenges, this paper proposes a RUL prediction method based on a multi-sensor data fusion model. In this method, the inherent degradation process of the system state is expressed using a state transition function following a Wiener process. Multi-sensor signals are explicated as various proxies of the inherent system degradation process using a multivariate measurement function. The system state is estimated by fusing multi-sensor signals using particle filtering. Informative sensors are selected by a prioritized sensor group selection algorithm. This algorithm first prioritizes sensors according to their individual performances in RUL prediction, and then selects an optimal sensor group based on their combined performances. The effectiveness of the proposed method is demonstrated using a simulation study and aircraft engine degradation data from NASA repository.

@article{luo_multiple_2020,
	title = {Multiple degradation mode analysis via gated recurrent unit mode recognizer and life predictors for complex equipment},
	volume = {123},
	issn = {0166-3615},
	url = {http://www.sciencedirect.com/science/article/pii/S0166361520305662},
	doi = {10.1016/j.compind.2020.103332},
	abstract = {In order to ensure the reliability and safety of industrial complex equipment, it is necessary to predict and manage the health status of the equipment. Remaining useful life (RUL) prediction is the decision basis of condition-based maintenance (CBM) and one of the main tasks in prognostics and health management (PHM). Complex systems tend to have multiple degradation modes, while similar degradation features may have significantly different RUL labels in different degradation modes, which can be called feature multi-label problem. To solve the problem, a novel RUL prediction method was proposed, which first analyzed the degradation mode and then utilized the predictor for RUL prediction under the specific mode. In particular, a modified de-noising auto-encoder (DAE) was proposed for nonlinear feature extraction and noise reduction. Mode recognizer and life predictors based on gated recurrent unit (GRU) and fuzzy k-means were proposed as the core modules. Case studies of commercial modular aero-propulsion system simulation data and the life cycle data of bearing were conducted to verify the effectiveness of the proposed method. Results show that the proposed method achieved much higher prediction accuracy than other methods.},
	language = {en},
	urldate = {2020-10-26},
	journal = {Computers in Industry},
	author = {Luo, Qinyuan and Chang, Yuanhong and Chen, Jinglong and Jing, Hongjie and Lv, Haixin and Pan, Tongyang},
	month = dec,
	year = {2020},
	keywords = {Complex equipment, Multiple degradation mode, Recurrent neural network, Remaining useful life},
	pages = {103332},
}

In order to ensure the reliability and safety of industrial complex equipment, it is necessary to predict and manage the health status of the equipment. Remaining useful life (RUL) prediction is the decision basis of condition-based maintenance (CBM) and one of the main tasks in prognostics and health management (PHM). Complex systems tend to have multiple degradation modes, while similar degradation features may have significantly different RUL labels in different degradation modes, which can be called feature multi-label problem. To solve the problem, a novel RUL prediction method was proposed, which first analyzed the degradation mode and then utilized the predictor for RUL prediction under the specific mode. In particular, a modified de-noising auto-encoder (DAE) was proposed for nonlinear feature extraction and noise reduction. Mode recognizer and life predictors based on gated recurrent unit (GRU) and fuzzy k-means were proposed as the core modules. Case studies of commercial modular aero-propulsion system simulation data and the life cycle data of bearing were conducted to verify the effectiveness of the proposed method. Results show that the proposed method achieved much higher prediction accuracy than other methods.

@article{erhan_smart_2020,
	title = {Smart anomaly detection in sensor systems: {A} multi-perspective review},
	issn = {1566-2535},
	shorttitle = {Smart anomaly detection in sensor systems},
	url = {http://www.sciencedirect.com/science/article/pii/S1566253520303717},
	doi = {10.1016/j.inffus.2020.10.001},
	abstract = {Anomaly detection is concerned with identifying data patterns that deviate remarkably from the expected behaviour. This is an important research problem, due to its broad set of application domains, from data analysis to e-health, cybersecurity, predictive maintenance, fault prevention, and industrial automation. Herein, we review state-of-the-art methods that may be employed to detect anomalies in the specific area of sensor systems, which poses hard challenges in terms of information fusion, data volumes, data speed, and network/energy efficiency, to mention but the most pressing ones. In this context, anomaly detection is a particularly hard problem, given the need to find computing-energy-accuracy trade-offs in a constrained environment. We taxonomize methods ranging from conventional techniques (statistical methods, time-series analysis, signal processing, etc.) to data-driven techniques (supervised learning, reinforcement learning, deep learning, etc.). We also look at the impact that different architectural environments (Cloud, Fog, Edge) can have on the sensors ecosystem. The review points to the most promising intelligent-sensing methods, and pinpoints a set of interesting open issues and challenges.},
	language = {en},
	urldate = {2020-10-19},
	journal = {Information Fusion},
	author = {Erhan, L. and Ndubuaku, M. and Di Mauro, M. and Song, W. and Chen, M. and Fortino, G. and Bagdasar, O. and Liotta, A.},
	month = oct,
	year = {2020},
	keywords = {Anomaly detection, Intelligent sensing, Internet of things, Machine learning, Sensor systems},
}

Anomaly detection is concerned with identifying data patterns that deviate remarkably from the expected behaviour. This is an important research problem, due to its broad set of application domains, from data analysis to e-health, cybersecurity, predictive maintenance, fault prevention, and industrial automation. Herein, we review state-of-the-art methods that may be employed to detect anomalies in the specific area of sensor systems, which poses hard challenges in terms of information fusion, data volumes, data speed, and network/energy efficiency, to mention but the most pressing ones. In this context, anomaly detection is a particularly hard problem, given the need to find computing-energy-accuracy trade-offs in a constrained environment. We taxonomize methods ranging from conventional techniques (statistical methods, time-series analysis, signal processing, etc.) to data-driven techniques (supervised learning, reinforcement learning, deep learning, etc.). We also look at the impact that different architectural environments (Cloud, Fog, Edge) can have on the sensors ecosystem. The review points to the most promising intelligent-sensing methods, and pinpoints a set of interesting open issues and challenges.

@article{zonta_predictive_2020,
	title = {Predictive maintenance in the {Industry} 4.0: {A} systematic literature review},
	volume = {150},
	issn = {0360-8352},
	shorttitle = {Predictive maintenance in the {Industry} 4.0},
	url = {http://www.sciencedirect.com/science/article/pii/S0360835220305787},
	doi = {10.1016/j.cie.2020.106889},
	abstract = {Industry 4.0 is collaborating directly for the technological revolution. Both machines and managers are daily confronted with decision making involving a massive input of data and customization in the manufacturing process. The ability to predict the need for maintenance of assets at a specific future moment is one of the main challenges in this scope. The possibility of performing predictive maintenance contributes to enhancing machine downtime, costs, control, and quality of production. We observed that surveys and tutorials about Industry 4.0 focus mainly on addressing data analytics and machine learning methods to change production procedures, so not comprising predictive maintenance methods and their organization. In this context, this article presents a systematic literature review of initiatives of predictive maintenance in Industry 4.0, identifying and cataloging methods, standards, and applications. As the main contributions, this survey discusses the current challenges and limitations in predictive maintenance, in addition to proposing a novel taxonomy to classify this research area considering the needs of the Industry 4.0. We concluded that computer science, including artificial intelligence and distributed computing fields, is more and more present in an area where engineering was the dominant expertise, so detaching the importance of a multidisciplinary approach to address Industry 4.0 effectively.},
	language = {en},
	urldate = {2020-10-12},
	journal = {Computers \& Industrial Engineering},
	author = {Zonta, Tiago and da Costa, Cristiano André and da Rosa Righi, Rodrigo and de Lima, Miromar José and da Trindade, Eduardo Silveira and Li, Guann Pyng},
	month = dec,
	year = {2020},
	keywords = {Artificial intelligence, Conditional-based maintenance, Industry 4.0, Predictive Maintenance, Remaining Useful Life},
	pages = {106889},
}

Industry 4.0 is collaborating directly for the technological revolution. Both machines and managers are daily confronted with decision making involving a massive input of data and customization in the manufacturing process. The ability to predict the need for maintenance of assets at a specific future moment is one of the main challenges in this scope. The possibility of performing predictive maintenance contributes to enhancing machine downtime, costs, control, and quality of production. We observed that surveys and tutorials about Industry 4.0 focus mainly on addressing data analytics and machine learning methods to change production procedures, so not comprising predictive maintenance methods and their organization. In this context, this article presents a systematic literature review of initiatives of predictive maintenance in Industry 4.0, identifying and cataloging methods, standards, and applications. As the main contributions, this survey discusses the current challenges and limitations in predictive maintenance, in addition to proposing a novel taxonomy to classify this research area considering the needs of the Industry 4.0. We concluded that computer science, including artificial intelligence and distributed computing fields, is more and more present in an area where engineering was the dominant expertise, so detaching the importance of a multidisciplinary approach to address Industry 4.0 effectively.

@article{ahmer_unified_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {A unified approach towards performance monitoring and condition-based maintenance in grinding machines},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307289},
	doi = {10.1016/j.procir.2020.04.094},
	abstract = {The process controller in a precision grinder for bearing rings puts high performance demands on the machine to achieve desired quality in production. This paper presents a unique approach of adding additional sensors for machine condition monitoring for the purpose of learning and using high fidelity condition indicators. The consolidation of real-time sensor data and the process control signals yields high-dimensional dataset. Automatic segmentation helps optimize the amount of data for processing and data mining ahead of fault diagnosis. The proposed setup is state of the art for prognostics as part of condition-based maintenance in a production machine.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Procedia CIRP},
	author = {Ahmer, Muhammad and Marklund, Pär and Gustafsson, Martin and Berglund, Kim},
	month = jan,
	year = {2020},
	keywords = {Analytics, Automation, Condition monitoring, Grinding, Machining, Maintenance, Manufacturing, Measurement, Process Monitoring, Sensor},
	pages = {1388--1393},
}

The process controller in a precision grinder for bearing rings puts high performance demands on the machine to achieve desired quality in production. This paper presents a unique approach of adding additional sensors for machine condition monitoring for the purpose of learning and using high fidelity condition indicators. The consolidation of real-time sensor data and the process control signals yields high-dimensional dataset. Automatic segmentation helps optimize the amount of data for processing and data mining ahead of fault diagnosis. The proposed setup is state of the art for prognostics as part of condition-based maintenance in a production machine.

@article{dogan_machine_2020,
	title = {Machine {Learning} and {Data} {Mining} in {Manufacturing}},
	issn = {0957-4174},
	url = {http://www.sciencedirect.com/science/article/pii/S095741742030823X},
	doi = {10.1016/j.eswa.2020.114060},
	abstract = {Manufacturing organizations need to use different kinds of techniques and tools in order to fulfill their foundation goals. In this aspect, using machine learning (ML) and data mining (DM) techniques and tools could be very helpful for dealing with challenges in manufacturing. Therefore, in this paper, a comprehensive literature review is presented to provide an overview of how machine learning techniques can be applied to realize manufacturing mechanisms with intelligent actions. Furthermore, it points to several significant research questions that are unanswered in the recent literature having the same target. Our survey aims to provide researchers with a solid understanding of the main approaches and algorithms used to improve manufacturing processes over the past two decades. It presents the previous ML studies and recent advances in manufacturing by grouping them under four main subjects: scheduling, monitoring, quality, and failure. It comprehensively discusses existing solutions in manufacturing according to various aspects, including tasks (i.e., clustering, classification, regression), algorithms (i.e., support vector machine, neural network), learning types (i.e., ensemble learning, deep learning), and performance metrics (i.e., accuracy, mean absolute error). Furthermore, the main steps of knowledge discovery in databases (KDD) process to be followed in manufacturing applications are explained in detail. In addition, some statistics about the current state are also given from different perspectives. Besides, it explains the advantages of using machine learning techniques in manufacturing, expresses the ways to overcome certain challenges, and offers some possible further research directions.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Expert Systems with Applications},
	author = {Dogan, Alican and Birant, Derya},
	month = sep,
	year = {2020},
	keywords = {Machine learning, classification, clustering, data mining, manufacturing},
	pages = {114060},
}

Manufacturing organizations need to use different kinds of techniques and tools in order to fulfill their foundation goals. In this aspect, using machine learning (ML) and data mining (DM) techniques and tools could be very helpful for dealing with challenges in manufacturing. Therefore, in this paper, a comprehensive literature review is presented to provide an overview of how machine learning techniques can be applied to realize manufacturing mechanisms with intelligent actions. Furthermore, it points to several significant research questions that are unanswered in the recent literature having the same target. Our survey aims to provide researchers with a solid understanding of the main approaches and algorithms used to improve manufacturing processes over the past two decades. It presents the previous ML studies and recent advances in manufacturing by grouping them under four main subjects: scheduling, monitoring, quality, and failure. It comprehensively discusses existing solutions in manufacturing according to various aspects, including tasks (i.e., clustering, classification, regression), algorithms (i.e., support vector machine, neural network), learning types (i.e., ensemble learning, deep learning), and performance metrics (i.e., accuracy, mean absolute error). Furthermore, the main steps of knowledge discovery in databases (KDD) process to be followed in manufacturing applications are explained in detail. In addition, some statistics about the current state are also given from different perspectives. Besides, it explains the advantages of using machine learning techniques in manufacturing, expresses the ways to overcome certain challenges, and offers some possible further research directions.

@article{langone_interpretable_2020,
	title = {Interpretable {Anomaly} {Prediction}: {Predicting} anomalous behavior in industry 4.0 settings via regularized logistic regression tools},
	issn = {0169-023X},
	shorttitle = {Interpretable {Anomaly} {Prediction}},
	url = {http://www.sciencedirect.com/science/article/pii/S0169023X1830644X},
	doi = {10.1016/j.datak.2020.101850},
	abstract = {Prediction of anomalous behavior in industrial assets based on sensor reading represents a key focus in modern business practice. As a matter of fact, forecast of forthcoming faults is crucial to implement predictive maintenance, i.e. maintenance decision making based on real time information from components and systems, which allows, among other benefits, to reduce maintenance cost, minimize downtime, increase safety, enhance product quality and productivity. However, building a model able to predict the future occurrence of a failure is challenging for various reasons. First, data are usually highly imbalanced, meaning that patterns describing a faulty regime are much less numerous than normal behavior instances, which makes model design difficult. Second, model predictions should be not only accurate (to avoid false alarms and missed detections) but also explainable to operators responsible for scheduling maintenance or control actions. In this paper we introduce a method called Interpretable Anomaly Prediction (IAP) allowing to handle these issues by using regularized logistic regression as core prediction model. In particular, in contrast to anomaly detection algorithms which permit to identify if the current data are anomalous or not, the proposed technique is able to predict the probability that future data will be abnormal. Furthermore, feature extraction and selection mechanisms give insights on the possible root causes leading to failures. The proposed strategy is validated with a large imbalanced multivariate time-series dataset consisting of measurements of several process variables surrounding an high pressure plunger pump situated in a complex chemical plant.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Data \& Knowledge Engineering},
	author = {Langone, Rocco and Cuzzocrea, Alfredo and Skantzos, Nikolaos},
	month = aug,
	year = {2020},
	pages = {101850},
}

Prediction of anomalous behavior in industrial assets based on sensor reading represents a key focus in modern business practice. As a matter of fact, forecast of forthcoming faults is crucial to implement predictive maintenance, i.e. maintenance decision making based on real time information from components and systems, which allows, among other benefits, to reduce maintenance cost, minimize downtime, increase safety, enhance product quality and productivity. However, building a model able to predict the future occurrence of a failure is challenging for various reasons. First, data are usually highly imbalanced, meaning that patterns describing a faulty regime are much less numerous than normal behavior instances, which makes model design difficult. Second, model predictions should be not only accurate (to avoid false alarms and missed detections) but also explainable to operators responsible for scheduling maintenance or control actions. In this paper we introduce a method called Interpretable Anomaly Prediction (IAP) allowing to handle these issues by using regularized logistic regression as core prediction model. In particular, in contrast to anomaly detection algorithms which permit to identify if the current data are anomalous or not, the proposed technique is able to predict the probability that future data will be abnormal. Furthermore, feature extraction and selection mechanisms give insights on the possible root causes leading to failures. The proposed strategy is validated with a large imbalanced multivariate time-series dataset consisting of measurements of several process variables surrounding an high pressure plunger pump situated in a complex chemical plant.

@article{shi_dual-lstm_2020,
	title = {A {Dual}-{LSTM} {Framework} {Combining} {Change} {Point} {Detection} and {Remaining} {Useful} {Life} {Prediction}},
	issn = {0951-8320},
	url = {http://www.sciencedirect.com/science/article/pii/S0951832020307572},
	doi = {10.1016/j.ress.2020.107257},
	abstract = {Remaining Useful Life (RUL) prediction is a key task of Condition-based Maintenance (CBM). The massive data collected from multiple sensors enables monitoring the complex systems in near real-time. However, such multiple sensors data environments pose a challenging task of combining the sensor data to infer the quality and RUL of the system. To address this task, we propose a Dual-LSTM framework that leverages Long-Short Term Memory (LSTM) for degradation analysis and RUL prediction. The Dual-LSTM relaxes the strong assumption of the fixed change point and detects the uncertain change point unit by unit at first. Then, the Dual-LSTM predicts the health index beyond the change point which can be leveraged to calculate the RUL. The proposed Dual-LSTM (i) achieves real-time high-precision RUL prediction by connecting the change point detection and RUL prediction with the health index construction, (ii) introduces a novel one-dimension health index function, (iii) leverages historical information to achieve detection and prediction tasks by characterizing both long and short-term dependencies of sensor signals through LSTM network. The effectiveness of the proposed Dual-LSTM framework is validated and compared to state-of-art benchmark methods on two publicly available turbofan engine degradation datasets.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Reliability Engineering \& System Safety},
	author = {Shi, Zunya and Chehade, Abdallah},
	month = oct,
	year = {2020},
	keywords = {Change point detection, Long short-term memory, Neural networks, Prognosis, Remaining useful life, Sensor fusion},
	pages = {107257},
}

Remaining Useful Life (RUL) prediction is a key task of Condition-based Maintenance (CBM). The massive data collected from multiple sensors enables monitoring the complex systems in near real-time. However, such multiple sensors data environments pose a challenging task of combining the sensor data to infer the quality and RUL of the system. To address this task, we propose a Dual-LSTM framework that leverages Long-Short Term Memory (LSTM) for degradation analysis and RUL prediction. The Dual-LSTM relaxes the strong assumption of the fixed change point and detects the uncertain change point unit by unit at first. Then, the Dual-LSTM predicts the health index beyond the change point which can be leveraged to calculate the RUL. The proposed Dual-LSTM (i) achieves real-time high-precision RUL prediction by connecting the change point detection and RUL prediction with the health index construction, (ii) introduces a novel one-dimension health index function, (iii) leverages historical information to achieve detection and prediction tasks by characterizing both long and short-term dependencies of sensor signals through LSTM network. The effectiveness of the proposed Dual-LSTM framework is validated and compared to state-of-art benchmark methods on two publicly available turbofan engine degradation datasets.

@article{javed_benchmark_2020,
	title = {A benchmark study on time series clustering},
	volume = {1},
	issn = {2666-8270},
	url = {http://www.sciencedirect.com/science/article/pii/S2666827020300013},
	doi = {10.1016/j.mlwa.2020.100001},
	abstract = {This paper presents the first time series clustering benchmark utilizing all time series datasets currently available in the University of California Riverside (UCR) archive — the state of the art repository of time series data. Specifically, the benchmark examines eight popular clustering methods representing three categories of clustering algorithms (partitional, hierarchical and density-based) and three types of distance measures (Euclidean, dynamic time warping, and shape-based), while adhering to six restrictions on datasets and methods to make the comparison as unbiased as possible. A phased evaluation approach was then designed for summarizing dataset-level assessment metrics and discussing the results. The benchmark study presented can be a useful reference for the research community on its own; and the dataset-level assessment metrics reported may be used for designing evaluation frameworks to answer different research questions.},
	language = {en},
	urldate = {2020-10-05},
	journal = {Machine Learning with Applications},
	author = {Javed, Ali and Lee, Byung Suk and Rizzo, Donna M.},
	month = sep,
	year = {2020},
	keywords = {Benchmark, Clustering, Time series, UCR archive},
	pages = {100001},
}

This paper presents the first time series clustering benchmark utilizing all time series datasets currently available in the University of California Riverside (UCR) archive — the state of the art repository of time series data. Specifically, the benchmark examines eight popular clustering methods representing three categories of clustering algorithms (partitional, hierarchical and density-based) and three types of distance measures (Euclidean, dynamic time warping, and shape-based), while adhering to six restrictions on datasets and methods to make the comparison as unbiased as possible. A phased evaluation approach was then designed for summarizing dataset-level assessment metrics and discussing the results. The benchmark study presented can be a useful reference for the research community on its own; and the dataset-level assessment metrics reported may be used for designing evaluation frameworks to answer different research questions.

@article{burg_evaluation_2020,
	title = {An {Evaluation} of {Change} {Point} {Detection} {Algorithms}},
	url = {http://arxiv.org/abs/2003.06222},
	abstract = {Change point detection is an important part of time series analysis, as the presence of a change point indicates an abrupt and significant change in the data generating process. While many algorithms for change point detection exist, little attention has been paid to evaluating their performance on real-world time series. Algorithms are typically evaluated on simulated data and a small number of commonly-used series with unreliable ground truth. Clearly this does not provide sufficient insight into the comparative performance of these algorithms. Therefore, instead of developing yet another change point detection method, we consider it vastly more important to properly evaluate existing algorithms on real-world data. To achieve this, we present the first data set specifically designed for the evaluation of change point detection algorithms, consisting of 37 time series from various domains. Each time series was annotated by five expert human annotators to provide ground truth on the presence and location of change points. We analyze the consistency of the human annotators, and describe evaluation metrics that can be used to measure algorithm performance in the presence of multiple ground truth annotations. Subsequently, we present a benchmark study where 14 existing algorithms are evaluated on each of the time series in the data set. This study shows that binary segmentation (Scott and Knott, 1974) and Bayesian online change point detection (Adams and MacKay, 2007) are among the best performing methods. Our aim is that this data set will serve as a proving ground in the development of novel change point detection algorithms.},
	urldate = {2020-10-02},
	journal = {arXiv:2003.06222 [cs, stat]},
	author = {Burg, Gerrit J. J. van den and Williams, Christopher K. I.},
	month = may,
	year = {2020},
	note = {arXiv: 2003.06222},
	keywords = {62M10, Computer Science - Machine Learning, G.3, Statistics - Machine Learning, Statistics - Methodology},
}

Change point detection is an important part of time series analysis, as the presence of a change point indicates an abrupt and significant change in the data generating process. While many algorithms for change point detection exist, little attention has been paid to evaluating their performance on real-world time series. Algorithms are typically evaluated on simulated data and a small number of commonly-used series with unreliable ground truth. Clearly this does not provide sufficient insight into the comparative performance of these algorithms. Therefore, instead of developing yet another change point detection method, we consider it vastly more important to properly evaluate existing algorithms on real-world data. To achieve this, we present the first data set specifically designed for the evaluation of change point detection algorithms, consisting of 37 time series from various domains. Each time series was annotated by five expert human annotators to provide ground truth on the presence and location of change points. We analyze the consistency of the human annotators, and describe evaluation metrics that can be used to measure algorithm performance in the presence of multiple ground truth annotations. Subsequently, we present a benchmark study where 14 existing algorithms are evaluated on each of the time series in the data set. This study shows that binary segmentation (Scott and Knott, 1974) and Bayesian online change point detection (Adams and MacKay, 2007) are among the best performing methods. Our aim is that this data set will serve as a proving ground in the development of novel change point detection algorithms.

@article{truong_selective_2020,
	title = {Selective review of offline change point detection methods},
	volume = {167},
	issn = {0165-1684},
	url = {http://www.sciencedirect.com/science/article/pii/S0165168419303494},
	doi = {10.1016/j.sigpro.2019.107299},
	abstract = {This article presents a selective survey of algorithms for the offline detection of multiple change points in multivariate time series. A general yet structuring methodological strategy is adopted to organize this vast body of work. More precisely, detection algorithms considered in this review are characterized by three elements: a cost function, a search method and a constraint on the number of changes. Each of those elements is described, reviewed and discussed separately. Implementations of the main algorithms described in this article are provided within a Python package called ruptures.},
	language = {en},
	urldate = {2020-10-01},
	journal = {Signal Processing},
	author = {Truong, Charles and Oudre, Laurent and Vayatis, Nicolas},
	month = feb,
	year = {2020},
	keywords = {Change point detection, Segmentation, Statistical signal processing},
	pages = {107299},
}

This article presents a selective survey of algorithms for the offline detection of multiple change points in multivariate time series. A general yet structuring methodological strategy is adopted to organize this vast body of work. More precisely, detection algorithms considered in this review are characterized by three elements: a cost function, a search method and a constraint on the number of changes. Each of those elements is described, reviewed and discussed separately. Implementations of the main algorithms described in this article are provided within a Python package called ruptures.

@article{duan_variable-length_2020,
	title = {Variable-length {Subsequence} {Clustering} in {Time} {Series}},
	issn = {1558-2191},
	doi = {10.1109/TKDE.2020.2986965},
	abstract = {Subsequence clustering is an important issue in time series data mining. Observing that most time series consist of various patterns with different unknown lengths, we propose an optimization framework to adaptively estimate the lengths and representations for different patterns. Our framework minimizes the inner subsequence cluster errors with respect to subsequence clusters and segmentation under time series cover constraint where the subsequence cluster lengths can be variable. To optimize our framework, we first generate abundant initial subsequence clusters with different lengths. Then, three cluster operations, i.e., cluster splitting, combination and removing, are used to iteratively refine the cluster lengths and representations by respectively splitting clusters consisting of different patterns, joining neighboring clusters belonging to the same pattern and removing clusters to the predefined cluster number. During each cluster refinement, we employ an efficient algorithm to alternatively optimize subsequence clusters and segmentation based on dynamic programming. Our method can automatically and efficiently extract the unknown variable-length subsequence clusters in the time series. Comparative results with the state-of-the-art are conducted on various synthetic and real time series, and quantitative and qualitative performances demonstrate the effectiveness of our method.},
	journal = {IEEE Transactions on Knowledge and Data Engineering},
	author = {Duan, Jiangyong and Guo, Lili},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Knowledge and Data Engineering},
	keywords = {Adaptation models, Clustering algorithms, Clustering methods, Data mining, Feature extraction, Optimization, Time series analysis, subsequence clustering, time series data mining, time series segmentation, variable-length patterns},
	pages = {1--1},
}

Subsequence clustering is an important issue in time series data mining. Observing that most time series consist of various patterns with different unknown lengths, we propose an optimization framework to adaptively estimate the lengths and representations for different patterns. Our framework minimizes the inner subsequence cluster errors with respect to subsequence clusters and segmentation under time series cover constraint where the subsequence cluster lengths can be variable. To optimize our framework, we first generate abundant initial subsequence clusters with different lengths. Then, three cluster operations, i.e., cluster splitting, combination and removing, are used to iteratively refine the cluster lengths and representations by respectively splitting clusters consisting of different patterns, joining neighboring clusters belonging to the same pattern and removing clusters to the predefined cluster number. During each cluster refinement, we employ an efficient algorithm to alternatively optimize subsequence clusters and segmentation based on dynamic programming. Our method can automatically and efficiently extract the unknown variable-length subsequence clusters in the time series. Comparative results with the state-of-the-art are conducted on various synthetic and real time series, and quantitative and qualitative performances demonstrate the effectiveness of our method.

@article{keriven_newma_2020,
	title = {{NEWMA}: {A} {New} {Method} for {Scalable} {Model}-{Free} {Online} {Change}-{Point} {Detection}},
	volume = {68},
	issn = {1941-0476},
	shorttitle = {{NEWMA}},
	doi = {10.1109/TSP.2020.2990597},
	abstract = {We consider the problem of detecting abrupt changes in the distribution of a multi-dimensional time series, with limited computing power and memory. In this paper, we propose a new, simple method for model-free online change-point detection that relies only on fast and light recursive statistics, inspired by the classical Exponential Weighted Moving Average algorithm (EWMA). The proposed idea is to compute two EWMA statistics on the stream of data with different forgetting factors, and to compare them. By doing so, we show that we implicitly compare recent samples with older ones, without the need to explicitly store them. Additionally, we leverage Random Features (RFs) to efficiently use the Maximum Mean Discrepancy as a distance between distributions, furthermore exploiting recent optical hardware to compute high-dimensional RFs in near constant time. We show that our method is significantly faster than usual non-parametric methods for a given accuracy.},
	journal = {IEEE Transactions on Signal Processing},
	author = {Keriven, Nicolas and Garreau, Damien and Poli, Iacopo},
	year = {2020},
	note = {Conference Name: IEEE Transactions on Signal Processing},
	keywords = {Brain modeling, Change detection algorithms, Computational modeling, Hilbert space, Kernel, Microsoft Windows, Radio frequency, Signal processing algorithms, Streaming media, method of moments, optical computing},
	pages = {3515--3528},
}

We consider the problem of detecting abrupt changes in the distribution of a multi-dimensional time series, with limited computing power and memory. In this paper, we propose a new, simple method for model-free online change-point detection that relies only on fast and light recursive statistics, inspired by the classical Exponential Weighted Moving Average algorithm (EWMA). The proposed idea is to compute two EWMA statistics on the stream of data with different forgetting factors, and to compare them. By doing so, we show that we implicitly compare recent samples with older ones, without the need to explicitly store them. Additionally, we leverage Random Features (RFs) to efficiently use the Maximum Mean Discrepancy as a distance between distributions, furthermore exploiting recent optical hardware to compute high-dimensional RFs in near constant time. We show that our method is significantly faster than usual non-parametric methods for a given accuracy.

@article{foerster_dynamic_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Dynamic risk consideration of predicted maintenance needs regarding economic efficiency},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120306545},
	doi = {10.1016/j.procir.2020.04.067},
	abstract = {In the context of predictive maintenance, where machine failures are predicted, it is assumed that a failure prognosis has already taken place. To ensure that such a forecast is adequately implemented, the results must be appropriately integrated into the planning processes. However, this leads to a planning conflict if a machine is addressed by both production and maintenance planning. For this reason, a prioritization of alternative actions is derived based on the concept of risk management. In this way, the appropriate measures can be economically quantified and finally weighed against each other. The result is a quantitative evaluation basis for a economic consideration of impending machine malfunctions on the one hand and value-adding production orders on the other.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Foerster, Fabian and Nikelowski, Lukas},
	month = jan,
	year = {2020},
	keywords = {economic efficiency, maintenance planning, predictive maintenance, production scheduling, risk management},
	pages = {915--920},
}

In the context of predictive maintenance, where machine failures are predicted, it is assumed that a failure prognosis has already taken place. To ensure that such a forecast is adequately implemented, the results must be appropriately integrated into the planning processes. However, this leads to a planning conflict if a machine is addressed by both production and maintenance planning. For this reason, a prioritization of alternative actions is derived based on the concept of risk management. In this way, the appropriate measures can be economically quantified and finally weighed against each other. The result is a quantitative evaluation basis for a economic consideration of impending machine malfunctions on the one hand and value-adding production orders on the other.

@article{hennig_comparison_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Comparison of {Time} {Series} {Clustering} {Algorithms} for {Machine} {State} {Detection}},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307149},
	doi = {10.1016/j.procir.2020.03.084},
	abstract = {New developments in domains like mathematics and statistical learning and availability of easy-to-use, often freely accessible software tools offer great potential to transform the manufacturing domain and their grasp on the increased manufacturing data repositories sustainably. One of the most exciting developments is in the area of machine learning. Time series clustering could be utilized in machine state detection which can be used in predictive maintenance or online optimization. This paper presents a comparison of freely available time series clustering algorithms, by applying several combinations of different algorithms to a database of public benchmark technical data.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Hennig, Martin and Grafinger, Manfred and Gerhard, Detlef and Dumss, Stefan and Rosenberger, Patrick},
	month = jan,
	year = {2020},
	keywords = {Industry 4.0, Internet of Things, Machine Learning, Predictive Maintenance, Time Series Clustering, Unsupervised Learning},
	pages = {1352--1357},
}

New developments in domains like mathematics and statistical learning and availability of easy-to-use, often freely accessible software tools offer great potential to transform the manufacturing domain and their grasp on the increased manufacturing data repositories sustainably. One of the most exciting developments is in the area of machine learning. Time series clustering could be utilized in machine state detection which can be used in predictive maintenance or online optimization. This paper presents a comparison of freely available time series clustering algorithms, by applying several combinations of different algorithms to a database of public benchmark technical data.

@article{nentwich_data-driven_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Data-driven {Models} for {Fault} {Classification} and {Prediction} of {Industrial} {Robots}},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307642},
	doi = {10.1016/j.procir.2020.04.126},
	abstract = {Economic data acquisition and storage have been key enablers to pave the way for data-driven predictions of machine downtimes. Regarding industrial robots, such predictions can maximize the robot’s availability and effective life span. This paper focuses on the comparison of different data-driven models for robot fault prediction and classification by applying them to a data set derived from a robot test bed and illuminates the data transformation process from raw sensor data to domain knowledge motivated robot health indicators.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Nentwich, Corbinian and Junker, Sebastian and Reinhart, Gunther},
	month = jan,
	year = {2020},
	keywords = {Type your keywords here, separated by semicolons},
	pages = {1055--1060},
}

Economic data acquisition and storage have been key enablers to pave the way for data-driven predictions of machine downtimes. Regarding industrial robots, such predictions can maximize the robot’s availability and effective life span. This paper focuses on the comparison of different data-driven models for robot fault prediction and classification by applying them to a data set derived from a robot test bed and illuminates the data transformation process from raw sensor data to domain knowledge motivated robot health indicators.

@article{fahle_systematic_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Systematic review on machine learning ({ML}) methods for manufacturing processes – {Identifying} artificial intelligence ({AI}) methods for field application},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307435},
	doi = {10.1016/j.procir.2020.04.109},
	abstract = {Artificial Intelligence (AI) and especially machine learning (ML) become increasingly more frequently applicable in factory operations. This paper presents a systematic review of today’s applications of ML techniques in the factory environment. The utilization of ML methods related to manufacturing process planning and control, predictive maintenance, quality control, in situ process control and optimization, logistics, robotics, assistance and learning systems for shopfloor employees are being analyzed. Moreover, an overview of ML training concepts in learning factories is given. Furthermore, these concepts will be analyzed regarding the implemented ML method. Finally, research gaps are identified.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Fahle, Simon and Prinz, Christopher and Kuhlenkötter, Bernd},
	month = jan,
	year = {2020},
	keywords = {Artificial Intelligence, factory operation, machine learning, production systems},
	pages = {413--418},
}

Artificial Intelligence (AI) and especially machine learning (ML) become increasingly more frequently applicable in factory operations. This paper presents a systematic review of today’s applications of ML techniques in the factory environment. The utilization of ML methods related to manufacturing process planning and control, predictive maintenance, quality control, in situ process control and optimization, logistics, robotics, assistance and learning systems for shopfloor employees are being analyzed. Moreover, an overview of ML training concepts in learning factories is given. Furthermore, these concepts will be analyzed regarding the implemented ML method. Finally, research gaps are identified.

@article{neto_digital_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Digital twins in manufacturing: an assessment of drivers, enablers and barriers to implementation},
	volume = {93},
	issn = {2212-8271},
	shorttitle = {Digital twins in manufacturing},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307733},
	doi = {10.1016/j.procir.2020.04.131},
	abstract = {As we live through the fourth industrial revolution, cutting-edge technologies look to change the way manufacturing systems operate. In this context, an important technological framework gaining popularity is the digital twin, which enables a virtual mirror of a real subject, used in manufacturing to assess performance and predict behavior. In this study, we interview experts and review the literature to gain an overview of what exactly drives companies to look for digital twin solutions in the manufacturing environment, what factors enable these initiatives to be successful, and what are the barriers that compromise or slow down implementation efforts.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Neto, Anis Assad and Deschamps, Fernando and da Silva, Elias Ribeiro and de Lima, Edson Pinheiro},
	month = jan,
	year = {2020},
	keywords = {Industry 4.0, digital twin, manufacturing, smart factory},
	pages = {210--215},
}

As we live through the fourth industrial revolution, cutting-edge technologies look to change the way manufacturing systems operate. In this context, an important technological framework gaining popularity is the digital twin, which enables a virtual mirror of a real subject, used in manufacturing to assess performance and predict behavior. In this study, we interview experts and review the literature to gain an overview of what exactly drives companies to look for digital twin solutions in the manufacturing environment, what factors enable these initiatives to be successful, and what are the barriers that compromise or slow down implementation efforts.

@article{muhlbauer_automated_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Automated {Data} {Labeling} and {Anomaly} {Detection} {Using} {Airborne} {Sound} {Analysis}},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120307587},
	doi = {10.1016/j.procir.2020.04.121},
	abstract = {Anomaly detection is facing the challenge of generating a maximum of information with a limited number of sensors in production machines and minimal effort of data analysis. Machine operators are often able to perceive changes acoustically and based on experience. In order to imitate this intelligent human capability, a systematic methodology has been developed in this work. Firstly, an introduction to this topic will be given and the acoustic sensor set up as well as the data preprocessing will be described. Secondly, an approach for automatic data labeling as an input for the intelligent anomaly detection will be presented. Using a model-based approach, anomaly detection is performed in the next step. This allows not only the detection of gradual and sudden process changes but also systematic problem solving. Finally, the anomaly detection will be validated for cutting processes.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Mühlbauer, Matthias and Würschinger, Hubert and Polzer, Dominik and Ju, Shu and Hanenkamp, Nico},
	month = jan,
	year = {2020},
	keywords = {Airborne Sound, Anomaly Detection, Automated Data Labeling, Cutting Processes, Cyclic Production Processes, Process Monitoring},
	pages = {1247--1252},
}

Anomaly detection is facing the challenge of generating a maximum of information with a limited number of sensors in production machines and minimal effort of data analysis. Machine operators are often able to perceive changes acoustically and based on experience. In order to imitate this intelligent human capability, a systematic methodology has been developed in this work. Firstly, an introduction to this topic will be given and the acoustic sensor set up as well as the data preprocessing will be described. Secondly, an approach for automatic data labeling as an input for the intelligent anomaly detection will be presented. Using a model-based approach, anomaly detection is performed in the next step. This allows not only the detection of gradual and sudden process changes but also systematic problem solving. Finally, the anomaly detection will be validated for cutting processes.

@article{kozjek_multi-objective_2020,
	series = {53rd {CIRP} {Conference} on {Manufacturing} {Systems} 2020},
	title = {Multi-objective adjustment of remaining useful life predictions based on reinforcement learning},
	volume = {93},
	issn = {2212-8271},
	url = {http://www.sciencedirect.com/science/article/pii/S2212827120306582},
	doi = {10.1016/j.procir.2020.03.051},
	abstract = {Effective tracking of degradation in machine tools or vehicle, ship, and aircraft engines is key to ensure their high utilization, effective maintenance, and safety. Data from the built-in sensors can be used to build models that accurately predict the remaining useful life (RUL) of the observed system. However, existing approaches often lack the ability to incorporate domain-specific knowledge in form of degradation models. This paper proposes a reinforcement-learning based approach for encoding the degradation model used for multi-objective adjustment of RUL predictions. The approach is demonstrated with a case of RUL prediction for aircraft engines.},
	language = {en},
	urldate = {2020-09-28},
	journal = {Procedia CIRP},
	author = {Kozjek, Dominik and Malus, Andreja and Vrabič, Rok},
	month = jan,
	year = {2020},
	keywords = {predictive maintainance, reinforcement learning, remaining useful life},
	pages = {425--430},
}

Effective tracking of degradation in machine tools or vehicle, ship, and aircraft engines is key to ensure their high utilization, effective maintenance, and safety. Data from the built-in sensors can be used to build models that accurately predict the remaining useful life (RUL) of the observed system. However, existing approaches often lack the ability to incorporate domain-specific knowledge in form of degradation models. This paper proposes a reinforcement-learning based approach for encoding the degradation model used for multi-objective adjustment of RUL predictions. The approach is demonstrated with a case of RUL prediction for aircraft engines.

@article{sahal_big_2020,
	title = {Big data and stream processing platforms for {Industry} 4.0 requirements mapping for a predictive maintenance use case},
	volume = {54},
	issn = {0278-6125},
	url = {http://www.sciencedirect.com/science/article/pii/S0278612519300937},
	doi = {10.1016/j.jmsy.2019.11.004},
	language = {en},
	urldate = {2020-03-24},
	journal = {Journal of Manufacturing Systems},
	author = {Sahal, Radhya and Breslin, John G. and Ali, Muhammad Intizar},
	month = jan,
	year = {2020},
	keywords = {Big Data, Industry 4.0, Predictive maintenance, Railway, Stream processing, Wind turbines},
	pages = {138--151},
}

@article{montero_jimenez_towards_2020,
	title = {Towards multi-model approaches to predictive maintenance: {A} systematic literature survey on diagnostics and prognostics},
	volume = {56},
	issn = {0278-6125},
	shorttitle = {Towards multi-model approaches to predictive maintenance},
	url = {http://www.sciencedirect.com/science/article/pii/S0278612520301187},
	doi = {10.1016/j.jmsy.2020.07.008},
	abstract = {The use of a modern technological system requires a good engineering approach, optimized operations, and proper maintenance in order to keep the system in an optimal state. Predictive maintenance focuses on the organization of maintenance actions according to the actual health state of the system, aiming at giving a precise indication of when a maintenance intervention will be necessary. Predictive maintenance is normally implemented by means of specialized computational systems that incorporate one of several models to fulfil diagnostics and prognostics tasks. As complexity of technological systems increases over time, single-model approaches hardly fulfil all functions and objectives for predictive maintenance systems. It is increasingly common to find research studies that combine different models in multi-model approaches to overcome complexity of predictive maintenance tasks, considering the advantages and disadvantages of each single model and trying to combine the best of them. These multi-model approaches have not been extensively addressed by previous review studies on predictive maintenance. Besides, many of the possible combinations for multi-model approaches remain unexplored in predictive maintenance applications; this offers a vast field of opportunities when architecting new predictive maintenance systems. This systematic survey aims at presenting the current trends in diagnostics and prognostics giving special attention to multi-model approaches and summarizing the current challenges and research opportunities.},
	language = {en},
	urldate = {2020-09-19},
	journal = {Journal of Manufacturing Systems},
	author = {Montero Jimenez, Juan José and Schwartz, Sébastien and Vingerhoeds, Rob and Grabot, Bernard and Salaün, Michel},
	month = jul,
	year = {2020},
	keywords = {Diagnostics, Multi-model approaches, Predictive maintenance, Prognostics, Single-model approaches, Systematic literature review},
	pages = {539--557},
}

The use of a modern technological system requires a good engineering approach, optimized operations, and proper maintenance in order to keep the system in an optimal state. Predictive maintenance focuses on the organization of maintenance actions according to the actual health state of the system, aiming at giving a precise indication of when a maintenance intervention will be necessary. Predictive maintenance is normally implemented by means of specialized computational systems that incorporate one of several models to fulfil diagnostics and prognostics tasks. As complexity of technological systems increases over time, single-model approaches hardly fulfil all functions and objectives for predictive maintenance systems. It is increasingly common to find research studies that combine different models in multi-model approaches to overcome complexity of predictive maintenance tasks, considering the advantages and disadvantages of each single model and trying to combine the best of them. These multi-model approaches have not been extensively addressed by previous review studies on predictive maintenance. Besides, many of the possible combinations for multi-model approaches remain unexplored in predictive maintenance applications; this offers a vast field of opportunities when architecting new predictive maintenance systems. This systematic survey aims at presenting the current trends in diagnostics and prognostics giving special attention to multi-model approaches and summarizing the current challenges and research opportunities.

@article{franceschi_unsupervised_2020,
	title = {Unsupervised {Scalable} {Representation} {Learning} for {Multivariate} {Time} {Series}},
	url = {http://arxiv.org/abs/1901.10738},
	abstract = {Time series constitute a challenging data type for machine learning algorithms, due to their highly variable lengths and sparse labeling in practice. In this paper, we tackle this challenge by proposing an unsupervised method to learn universal embeddings of time series. Unlike previous works, it is scalable with respect to their length and we demonstrate the quality, transferability and practicability of the learned representations with thorough experiments and comparisons. To this end, we combine an encoder based on causal dilated convolutions with a novel triplet loss employing time-based negative sampling, obtaining general-purpose representations for variable length and multivariate time series.},
	urldate = {2020-07-20},
	journal = {arXiv:1901.10738 [cs, stat]},
	author = {Franceschi, Jean-Yves and Dieuleveut, Aymeric and Jaggi, Martin},
	month = jan,
	year = {2020},
	note = {arXiv: 1901.10738},
	keywords = {Computer Science - Machine Learning, Computer Science - Neural and Evolutionary Computing, Statistics - Machine Learning},
}

Time series constitute a challenging data type for machine learning algorithms, due to their highly variable lengths and sparse labeling in practice. In this paper, we tackle this challenge by proposing an unsupervised method to learn universal embeddings of time series. Unlike previous works, it is scalable with respect to their length and we demonstrate the quality, transferability and practicability of the learned representations with thorough experiments and comparisons. To this end, we combine an encoder based on causal dilated convolutions with a novel triplet loss employing time-based negative sampling, obtaining general-purpose representations for variable length and multivariate time series.

@inproceedings{pirasteh_interactive_2019,
	address = {New York, NY, USA},
	series = {{WIDM}'19},
	title = {Interactive feature extraction for diagnostic trouble codes in predictive maintenance: {A} case study from automotive domain},
	isbn = {978-1-4503-6296-2},
	shorttitle = {Interactive feature extraction for diagnostic trouble codes in predictive maintenance},
	url = {https://dl.acm.org/doi/10.1145/3304079.3310288},
	doi = {10.1145/3304079.3310288},
	abstract = {Predicting future maintenance needs of equipment can be addressed in a variety of ways. Methods based on machine learning approaches provide an interesting platform for mining large data sets to find patterns that might correlate with a given fault. In this paper, we approach predictive maintenance as a classification problem and use Random Forest to separate data readouts within a particular time window into those corresponding to faulty and non-faulty component categories. We utilize diagnostic trouble codes (DTCs) as an example of event-based data, and propose four categories of features that can be derived from DTCs as a predictive maintenance framework. We test the approach using large-scale data from a fleet of heavy duty trucks, and show that DTCs can be used within our framework as indicators of imminent failures in different components.},
	urldate = {2023-05-21},
	booktitle = {Proceedings of the {Workshop} on {Interactive} {Data} {Mining}},
	publisher = {Association for Computing Machinery},
	author = {Pirasteh, Parivash and Nowaczyk, Slawomir and Pashami, Sepideh and Löwenadler, Magnus and Thunberg, Klas and Ydreskog, Henrik and Berck, Peter},
	month = feb,
	year = {2019},
	keywords = {Predictive maintenance, diagnostic trouble codes, failure detection, feature extraction},
	pages = {1--10},
}

Predicting future maintenance needs of equipment can be addressed in a variety of ways. Methods based on machine learning approaches provide an interesting platform for mining large data sets to find patterns that might correlate with a given fault. In this paper, we approach predictive maintenance as a classification problem and use Random Forest to separate data readouts within a particular time window into those corresponding to faulty and non-faulty component categories. We utilize diagnostic trouble codes (DTCs) as an example of event-based data, and propose four categories of features that can be derived from DTCs as a predictive maintenance framework. We test the approach using large-scale data from a fleet of heavy duty trucks, and show that DTCs can be used within our framework as indicators of imminent failures in different components.

@inproceedings{le_nguyen_semi-supervised_2019,
	title = {Semi-supervised {Learning} over {Streaming} {Data} using {MOA}},
	doi = {10.1109/BigData47090.2019.9006217},
	abstract = {Machine learning algorithms for data streams usually suppose that all data examples available for learning are strictly labeled. Unfortunately, in real-world scenarios, data examples are not always labeled. Semi-supervised learning is a challenging task to learn using labeled and unlabeled data at the same time. It is especially relevant in the context of data streams, where the data is generated in real-time, and the labels may be missing due to various factors (e.g., network delay, errors during the communication between sensors, expensive labeling process, and others). In this paper, we present two novel approaches to handle missing labels for classification learning in data streams, namely cluster-and-label and self-training. We discuss the strengths and weaknesses of each solution to establish a baseline to evaluate semi-supervised learning techniques in data streams. These methods are implemented inside the MOA (Massive Online Analysis) open-source software as an internal benchmark component, to help researchers to run experimental comparisons on semi-supervised learning on data streams easily.},
	booktitle = {2019 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Le Nguyen, Minh Huong and Gomes, Heitor Murilo and Bifet, Albert},
	month = dec,
	year = {2019},
	keywords = {Clustering algorithms, Data models, Labeling, Prediction algorithms, Predictive models, Semisupervised learning, Supervised learning, data streams, semi-supervised learning},
	pages = {553--562},
}

Machine learning algorithms for data streams usually suppose that all data examples available for learning are strictly labeled. Unfortunately, in real-world scenarios, data examples are not always labeled. Semi-supervised learning is a challenging task to learn using labeled and unlabeled data at the same time. It is especially relevant in the context of data streams, where the data is generated in real-time, and the labels may be missing due to various factors (e.g., network delay, errors during the communication between sensors, expensive labeling process, and others). In this paper, we present two novel approaches to handle missing labels for classification learning in data streams, namely cluster-and-label and self-training. We discuss the strengths and weaknesses of each solution to establish a baseline to evaluate semi-supervised learning techniques in data streams. These methods are implemented inside the MOA (Massive Online Analysis) open-source software as an internal benchmark component, to help researchers to run experimental comparisons on semi-supervised learning on data streams easily.

@inproceedings{verdun_remote_2019,
	address = {Tokyo, Japan},
	title = {Remote {Diagnosis} and {Condition}-based {Maintenance} for {Rolling} {Stock} at {SNCF}},
	abstract = {For the rolling stock, the new generation of highly complex and communicating trains will fundamentally transform the current landscape and the industrial sector in the next few years. As the rolling stock is changing, its maintenance must evolve to meet new challenges. Nowadays, the maintenance system is mostly characterized by a combination of several maintenance types and strategies which goes from preventive systematic maintenance to corrective maintenance. The introduction of on-board/wayside diagnostic systems presents a significant opportunity for reducing maintenance costs, while also having a strong positive impact on reliability, availability and quality of service. Following this way, a new approach based on remote diagnosis and condition based maintenance has been incorporated into the SNCF maintenance process few years ago. This paper gives an overview of the work carried out over the last years, with the challenges encountered and the first results.},
	author = {Verdun, Cyril and Turgis, Fabien and Audier, Pierre},
	year = {2019},
}

For the rolling stock, the new generation of highly complex and communicating trains will fundamentally transform the current landscape and the industrial sector in the next few years. As the rolling stock is changing, its maintenance must evolve to meet new challenges. Nowadays, the maintenance system is mostly characterized by a combination of several maintenance types and strategies which goes from preventive systematic maintenance to corrective maintenance. The introduction of on-board/wayside diagnostic systems presents a significant opportunity for reducing maintenance costs, while also having a strong positive impact on reliability, availability and quality of service. Following this way, a new approach based on remote diagnosis and condition based maintenance has been incorporated into the SNCF maintenance process few years ago. This paper gives an overview of the work carried out over the last years, with the challenges encountered and the first results.

@inproceedings{bouchikhi_kernel_2019,
	title = {Kernel {Based} {Online} {Change} {Point} {Detection}},
	doi = {10.23919/EUSIPCO.2019.8902582},
	abstract = {Detecting change points in time series data is a challenging problem, in particular when no prior information on the data distribution and the nature of the change is available. In a former work, we introduced an online non-parametric change-point detection framework built upon direct density ratio estimation over two consecutive time segments, rather than modeling densities separately. This algorithm based on the theory of reproducing kernels showed positive and reliable detection results for a variety of problems. To further improve the detection performance of this approach, we propose in this paper to modify the original cost function in order to achieve unbiasedness of the density ratio estimation under the null hypothesis. Theoretical analysis and numerical simulations confirm the improved behavior of this method, as well as its efficiency compared to a state of the art one. Application to sentiment change detection in Twitter data streams is also presented.},
	booktitle = {2019 27th {European} {Signal} {Processing} {Conference} ({EUSIPCO})},
	author = {Bouchikhi, Ikram and Ferrari, André and Richard, Cédric and Bourrier, Anthony and Bernot, Marc},
	month = sep,
	year = {2019},
	note = {ISSN: 2076-1465},
	keywords = {Change detection algorithms, Dictionaries, Estimation, Europe, Kernel, Non-parametric change-point detection, Signal processing, Signal processing algorithms, convergence analysis, kernel least-mean-square algorithm, online learning, reproducing kernel Hilbert space},
	pages = {1--5},
}

Detecting change points in time series data is a challenging problem, in particular when no prior information on the data distribution and the nature of the change is available. In a former work, we introduced an online non-parametric change-point detection framework built upon direct density ratio estimation over two consecutive time segments, rather than modeling densities separately. This algorithm based on the theory of reproducing kernels showed positive and reliable detection results for a variety of problems. To further improve the detection performance of this approach, we propose in this paper to modify the original cost function in order to achieve unbiasedness of the density ratio estimation under the null hypothesis. Theoretical analysis and numerical simulations confirm the improved behavior of this method, as well as its efficiency compared to a state of the art one. Application to sentiment change detection in Twitter data streams is also presented.

@article{fahy_dynamic_2019,
	title = {Dynamic {Feature} {Selection} for {Clustering} {High} {Dimensional} {Data} {Streams}},
	volume = {7},
	issn = {2169-3536},
	doi = {10.1109/ACCESS.2019.2932308},
	abstract = {Change in a data stream can occur at the concept level and at the feature level. Change at the feature level can occur if new, additional features appear in the stream or if the importance and relevance of a feature changes as the stream progresses. This type of change has not received as much attention as concept-level change. Furthermore, a lot of the methods proposed for clustering streams (density-based, graph-based, and grid-based) rely on some form of distance as a similarity metric and this is problematic in high-dimensional data where the curse of dimensionality renders distance measurements and any concept of “density” difficult. To address these two challenges we propose combining them and framing the problem as a feature selection problem, specifically a dynamic feature selection problem. We propose a dynamic feature mask for clustering high dimensional data streams. Redundant features are masked and clustering is performed along unmasked, relevant features. If a feature's perceived importance changes, the mask is updated accordingly; previously unimportant features are unmasked and features which lose relevance become masked. The proposed method is algorithm-independent and can be used with any of the existing density-based clustering algorithms which typically do not have a mechanism for dealing with feature drift and struggle with high-dimensional data. We evaluate the proposed method on four density-based clustering algorithms across four high-dimensional streams; two text streams and two image streams. In each case, the proposed dynamic feature mask improves clustering performance and reduces the processing time required by the underlying algorithm. Furthermore, change at the feature level can be observed and tracked.},
	journal = {IEEE Access},
	author = {Fahy, Conor and Yang, Shengxiang},
	year = {2019},
	note = {Conference Name: IEEE Access},
	keywords = {Clustering algorithms, Correlation, Data stream clustering, Feature extraction, Heuristic algorithms, Measurement, Microsoft Windows, Streaming media, dynamic feature selection, feature drift, feature evolution, unsupervised feature selection},
	pages = {127128--127140},
}

Change in a data stream can occur at the concept level and at the feature level. Change at the feature level can occur if new, additional features appear in the stream or if the importance and relevance of a feature changes as the stream progresses. This type of change has not received as much attention as concept-level change. Furthermore, a lot of the methods proposed for clustering streams (density-based, graph-based, and grid-based) rely on some form of distance as a similarity metric and this is problematic in high-dimensional data where the curse of dimensionality renders distance measurements and any concept of “density” difficult. To address these two challenges we propose combining them and framing the problem as a feature selection problem, specifically a dynamic feature selection problem. We propose a dynamic feature mask for clustering high dimensional data streams. Redundant features are masked and clustering is performed along unmasked, relevant features. If a feature's perceived importance changes, the mask is updated accordingly; previously unimportant features are unmasked and features which lose relevance become masked. The proposed method is algorithm-independent and can be used with any of the existing density-based clustering algorithms which typically do not have a mechanism for dealing with feature drift and struggle with high-dimensional data. We evaluate the proposed method on four density-based clustering algorithms across four high-dimensional streams; two text streams and two image streams. In each case, the proposed dynamic feature mask improves clustering performance and reduces the processing time required by the underlying algorithm. Furthermore, change at the feature level can be observed and tracked.

@inproceedings{chawathe_condition_2019,
	title = {Condition {Monitoring} of {Hydraulic} {Systems} by {Classifying} {Sensor} {Data} {Streams}},
	doi = {10.1109/CCWC.2019.8666564},
	abstract = {Condition-based maintenance (CBM) of hydraulic systems requires methods for condition monitoring: Sensors installed in a hydraulic system for this purpose generate streams of real-time data that must be analyzed to accurately characterize the health of the system. Prior work has developed an experimental hydraulic system with such an installation and yielded a public dataset of sensor readings with associated values of condition variables that quantify the system's health. This paper presents classification-based methods for inferring these condition variables from the sensor data streams. These methods significantly improve on the classification accuracy reported in prior work on this data. Further, this accuracy is maintained even when the number of sensor-based attributes used as input is substantially reduced.},
	booktitle = {2019 {IEEE} 9th {Annual} {Computing} and {Communication} {Workshop} and {Conference} ({CCWC})},
	author = {Chawathe, Sudarshan S.},
	month = jan,
	year = {2019},
	keywords = {Bars, Condition monitoring, Feature extraction, Hydraulic systems, Monitoring, Training, classification, condition monitoring, condition-based maintenance, hydraulic systems, sensors},
	pages = {0898--0904},
}

Condition-based maintenance (CBM) of hydraulic systems requires methods for condition monitoring: Sensors installed in a hydraulic system for this purpose generate streams of real-time data that must be analyzed to accurately characterize the health of the system. Prior work has developed an experimental hydraulic system with such an installation and yielded a public dataset of sensor readings with associated values of condition variables that quantify the system's health. This paper presents classification-based methods for inferring these condition variables from the sensor data streams. These methods significantly improve on the classification accuracy reported in prior work on this data. Further, this accuracy is maintained even when the number of sensor-based attributes used as input is substantially reduced.

@article{peng_new_2019,
	title = {A {New} {Hierarchical} {Framework} for {Detection} and {Isolation} of {Multiple} {Faults} in {Complex} {Industrial} {Processes}},
	volume = {7},
	issn = {2169-3536},
	doi = {10.1109/ACCESS.2019.2892487},
	abstract = {In actual production practice, the occurrence probability of multiple faults is much higher than that of a single fault. Since the composition of multiple faults is uncertain, it is difficult to establish a single model for multifault diagnosis. In this paper, a new hierarchical framework is proposed for solving multifault detection and isolation problems. First, an adaptive dynamic kernel independent component analysis method is proposed for time-varying and unknown multifault detection. After that, a sparse local exponential discriminant analysis method is developed for the multimodal multifault isolation problem. Finally, the Tennessee Eastman process is used to validate the performance of the proposed methods, and the experimental results show that the proposed methods can efficiently detect and isolate multiple faults.},
	journal = {IEEE Access},
	author = {Peng, Kaixiang and Ren, Zhihao and Dong, Jie and Ma, Liang},
	year = {2019},
	note = {Conference Name: IEEE Access},
	keywords = {Covariance matrices, Eigenvalues and eigenfunctions, Independent component analysis, Kernel, Monitoring, Multiple faults, Principal component analysis, Production, accurate isolation, complex industrial processes, hierarchical framework, real-time detection},
	pages = {12006--12015},
}

In actual production practice, the occurrence probability of multiple faults is much higher than that of a single fault. Since the composition of multiple faults is uncertain, it is difficult to establish a single model for multifault diagnosis. In this paper, a new hierarchical framework is proposed for solving multifault detection and isolation problems. First, an adaptive dynamic kernel independent component analysis method is proposed for time-varying and unknown multifault detection. After that, a sparse local exponential discriminant analysis method is developed for the multimodal multifault isolation problem. Finally, the Tennessee Eastman process is used to validate the performance of the proposed methods, and the experimental results show that the proposed methods can efficiently detect and isolate multiple faults.

@article{luo_discriminant_2019,
	title = {Discriminant autoencoder for feature extraction in fault diagnosis},
	volume = {192},
	issn = {0169-7439},
	url = {https://www.sciencedirect.com/science/article/pii/S0169743918306257},
	doi = {10.1016/j.chemolab.2019.103814},
	abstract = {Nowadays, some traditional autoencoders and their extensions have been widely applied in data-driven fault diagnosis for feature extraction. However, because of the fact that traditional autoencoders could not make use of label information, the representations extracted by these traditional autoencoders may show disappointing results when handling ultimate discriminative task. In this paper, we propose a novel semi-supervised autoencoder, which is named as Discriminant Autoencoder. The training of proposed Discriminant Autoencoder includes a supervised process and an unsupervised process. And a distance penalty is added into the loss function, which enables the proposed Discriminant Autoencoder to extract more suitable representations from industrial data samples. In order to explain the effectiveness of this semi-supervised autoencoder, we carry out some experiments and give out a mathematical derivation. Here we use an industrial batch process dataset as the criterion dataset to test the performance of proposed Discriminant Autoencoder and other conventional autoencoders.},
	language = {en},
	urldate = {2022-05-02},
	journal = {Chemometrics and Intelligent Laboratory Systems},
	author = {Luo, Xiaoyi and Li, Xianmin and Wang, Ziyang and Liang, Jun},
	month = sep,
	year = {2019},
	keywords = {Autoencoder, Fault diagnosis, Feature extraction, Semi-supervised autoencoder},
	pages = {103814},
}

Nowadays, some traditional autoencoders and their extensions have been widely applied in data-driven fault diagnosis for feature extraction. However, because of the fact that traditional autoencoders could not make use of label information, the representations extracted by these traditional autoencoders may show disappointing results when handling ultimate discriminative task. In this paper, we propose a novel semi-supervised autoencoder, which is named as Discriminant Autoencoder. The training of proposed Discriminant Autoencoder includes a supervised process and an unsupervised process. And a distance penalty is added into the loss function, which enables the proposed Discriminant Autoencoder to extract more suitable representations from industrial data samples. In order to explain the effectiveness of this semi-supervised autoencoder, we carry out some experiments and give out a mathematical derivation. Here we use an industrial batch process dataset as the criterion dataset to test the performance of proposed Discriminant Autoencoder and other conventional autoencoders.

@article{kolesnikov_revisiting_2019,
	title = {Revisiting {Self}-{Supervised} {Visual} {Representation} {Learning}},
	url = {http://arxiv.org/abs/1901.09005},
	abstract = {Unsupervised visual representation learning remains a largely unsolved problem in computer vision research. Among a big body of recently proposed approaches for unsupervised learning of visual representations, a class of self-supervised techniques achieves superior performance on many challenging benchmarks. A large number of the pretext tasks for self-supervised learning have been studied, but other important aspects, such as the choice of convolutional neural networks (CNN), has not received equal attention. Therefore, we revisit numerous previously proposed self-supervised models, conduct a thorough large scale study and, as a result, uncover multiple crucial insights. We challenge a number of common practices in selfsupervised visual representation learning and observe that standard recipes for CNN design do not always translate to self-supervised representation learning. As part of our study, we drastically boost the performance of previously proposed techniques and outperform previously published state-of-the-art results by a large margin.},
	urldate = {2022-04-25},
	journal = {arXiv:1901.09005 [cs]},
	author = {Kolesnikov, Alexander and Zhai, Xiaohua and Beyer, Lucas},
	month = jan,
	year = {2019},
	note = {arXiv: 1901.09005},
	keywords = {Computer Science - Computer Vision and Pattern Recognition},
}

Unsupervised visual representation learning remains a largely unsolved problem in computer vision research. Among a big body of recently proposed approaches for unsupervised learning of visual representations, a class of self-supervised techniques achieves superior performance on many challenging benchmarks. A large number of the pretext tasks for self-supervised learning have been studied, but other important aspects, such as the choice of convolutional neural networks (CNN), has not received equal attention. Therefore, we revisit numerous previously proposed self-supervised models, conduct a thorough large scale study and, as a result, uncover multiple crucial insights. We challenge a number of common practices in selfsupervised visual representation learning and observe that standard recipes for CNN design do not always translate to self-supervised representation learning. As part of our study, we drastically boost the performance of previously proposed techniques and outperform previously published state-of-the-art results by a large margin.

@article{van_de_ven_three_2019,
	title = {Three scenarios for continual learning},
	url = {http://arxiv.org/abs/1904.07734},
	abstract = {Standard artificial neural networks suffer from the well-known issue of catastrophic forgetting, making continual or lifelong learning difficult for machine learning. In recent years, numerous methods have been proposed for continual learning, but due to differences in evaluation protocols it is difficult to directly compare their performance. To enable more structured comparisons, we describe three continual learning scenarios based on whether at test time task identity is provided and--in case it is not--whether it must be inferred. Any sequence of well-defined tasks can be performed according to each scenario. Using the split and permuted MNIST task protocols, for each scenario we carry out an extensive comparison of recently proposed continual learning methods. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of how efficient different methods are. In particular, when task identity must be inferred (i.e., class incremental learning), we find that regularization-based approaches (e.g., elastic weight consolidation) fail and that replaying representations of previous experiences seems required for solving this scenario.},
	urldate = {2022-03-19},
	journal = {arXiv:1904.07734 [cs, stat]},
	author = {van de Ven, Gido M. and Tolias, Andreas S.},
	month = apr,
	year = {2019},
	note = {arXiv: 1904.07734},
	keywords = {Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning, Statistics - Machine Learning},
}

Standard artificial neural networks suffer from the well-known issue of catastrophic forgetting, making continual or lifelong learning difficult for machine learning. In recent years, numerous methods have been proposed for continual learning, but due to differences in evaluation protocols it is difficult to directly compare their performance. To enable more structured comparisons, we describe three continual learning scenarios based on whether at test time task identity is provided and–in case it is not–whether it must be inferred. Any sequence of well-defined tasks can be performed according to each scenario. Using the split and permuted MNIST task protocols, for each scenario we carry out an extensive comparison of recently proposed continual learning methods. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of how efficient different methods are. In particular, when task identity must be inferred (i.e., class incremental learning), we find that regularization-based approaches (e.g., elastic weight consolidation) fail and that replaying representations of previous experiences seems required for solving this scenario.

@article{wang_deep_2019,
	series = {Deep {Learning} for {Pattern} {Recognition}},
	title = {Deep learning for sensor-based activity recognition: {A} survey},
	volume = {119},
	issn = {0167-8655},
	shorttitle = {Deep learning for sensor-based activity recognition},
	url = {https://www.sciencedirect.com/science/article/pii/S016786551830045X},
	doi = {10.1016/j.patrec.2018.02.010},
	abstract = {Sensor-based activity recognition seeks the profound high-level knowledge about human activities from multitudes of low-level sensor readings. Conventional pattern recognition approaches have made tremendous progress in the past years. However, those methods often heavily rely on heuristic hand-crafted feature extraction, which could hinder their generalization performance. Additionally, existing methods are undermined for unsupervised and incremental learning tasks. Recently, the recent advancement of deep learning makes it possible to perform automatic high-level feature extraction thus achieves promising performance in many areas. Since then, deep learning based methods have been widely adopted for the sensor-based activity recognition tasks. This paper surveys the recent advance of deep learning based sensor-based activity recognition. We summarize existing literature from three aspects: sensor modality, deep model, and application. We also present detailed insights on existing work and propose grand challenges for future research.},
	language = {en},
	urldate = {2022-03-17},
	journal = {Pattern Recognition Letters},
	author = {Wang, Jindong and Chen, Yiqiang and Hao, Shuji and Peng, Xiaohui and Hu, Lisha},
	month = mar,
	year = {2019},
	keywords = {Activity recognition, Deep learning, Pattern recognition, Pervasive computing},
	pages = {3--11},
}

Sensor-based activity recognition seeks the profound high-level knowledge about human activities from multitudes of low-level sensor readings. Conventional pattern recognition approaches have made tremendous progress in the past years. However, those methods often heavily rely on heuristic hand-crafted feature extraction, which could hinder their generalization performance. Additionally, existing methods are undermined for unsupervised and incremental learning tasks. Recently, the recent advancement of deep learning makes it possible to perform automatic high-level feature extraction thus achieves promising performance in many areas. Since then, deep learning based methods have been widely adopted for the sensor-based activity recognition tasks. This paper surveys the recent advance of deep learning based sensor-based activity recognition. We summarize existing literature from three aspects: sensor modality, deep model, and application. We also present detailed insights on existing work and propose grand challenges for future research.

@inproceedings{costa_junior_novelty_2019,
	title = {Novelty {Detection} for {Multi}-{Label} {Stream} {Classification}},
	doi = {10.1109/BRACIS.2019.00034},
	abstract = {In Multi-Label Stream Classification (MLSC) examples arriving in a stream can be simultaneously classified into multiple classes. This is a very challenging task, especially considering that new classes can emerge during the stream (Concept Evolution), and known classes can change over time (Concept Drift). In real situations, these characteristics come together with a scenario with Infinitely Delayed Labels, where we can never access the true class labels of the examples to update classifiers. In order to overcome these issues, this paper proposes a new method called MultI-label learNing Algorithm for Data Streams with Binary Relevance transformation (MINAS-BR). Our proposal uses a new Novelty Detection (ND) procedure to detect concept evolution and concept drift, being updated in an unsupervised fashion. We also propose a new methodology to evaluate MLSC methods in scenarios with Infinitely Delayed Labels. Experiments over synthetic data sets attested the potential of MINAS-BR, which was able to adapt to different concept drift and concept evolution scenarios, obtaining superior or competitive performances in comparison to literature baselines.},
	booktitle = {2019 8th {Brazilian} {Conference} on {Intelligent} {Systems} ({BRACIS})},
	author = {Costa Júnior, Joel D. and Faria, Elaine R. and Silva, Jonathan A. and Gama, João and Cerri, Ricardo},
	month = oct,
	year = {2019},
	note = {ISSN: 2643-6264},
	keywords = {Adaptation models, Computational complexity, Computational modeling, Concept Evolution, Data models, Infinitely Delayed Labels, Multi-label Stream Classification, Novelty Detection, Task analysis, Training, Training data},
	pages = {144--149},
}

In Multi-Label Stream Classification (MLSC) examples arriving in a stream can be simultaneously classified into multiple classes. This is a very challenging task, especially considering that new classes can emerge during the stream (Concept Evolution), and known classes can change over time (Concept Drift). In real situations, these characteristics come together with a scenario with Infinitely Delayed Labels, where we can never access the true class labels of the examples to update classifiers. In order to overcome these issues, this paper proposes a new method called MultI-label learNing Algorithm for Data Streams with Binary Relevance transformation (MINAS-BR). Our proposal uses a new Novelty Detection (ND) procedure to detect concept evolution and concept drift, being updated in an unsupervised fashion. We also propose a new methodology to evaluate MLSC methods in scenarios with Infinitely Delayed Labels. Experiments over synthetic data sets attested the potential of MINAS-BR, which was able to adapt to different concept drift and concept evolution scenarios, obtaining superior or competitive performances in comparison to literature baselines.

@article{carnein_optimizing_2019,
	title = {Optimizing {Data} {Stream} {Representation}: {An} {Extensive} {Survey} on {Stream} {Clustering} {Algorithms}},
	volume = {61},
	issn = {1867-0202},
	shorttitle = {Optimizing {Data} {Stream} {Representation}},
	url = {https://doi.org/10.1007/s12599-019-00576-5},
	doi = {10.1007/s12599-019-00576-5},
	abstract = {Analyzing data streams has received considerable attention over the past decades due to the widespread usage of sensors, social media and other streaming data sources. A core research area in this field is stream clustering which aims to recognize patterns in an unordered, infinite and evolving stream of observations. Clustering can be a crucial support in decision making, since it aims for an optimized aggregated representation of a continuous data stream over time and allows to identify patterns in large and high-dimensional data. A multitude of algorithms and approaches has been developed that are able to find and maintain clusters over time in the challenging streaming scenario. This survey explores, summarizes and categorizes a total of 51 stream clustering algorithms and identifies core research threads over the past decades. In particular, it identifies categories of algorithms based on distance thresholds, density grids and statistical models as well as algorithms for high dimensional data. Furthermore, it discusses applications scenarios, available software and how to configure stream clustering algorithms. This survey is considerably more extensive than comparable studies, more up-to-date and highlights how concepts are interrelated and have been developed over time.},
	language = {en},
	number = {3},
	urldate = {2022-03-15},
	journal = {Business \& Information Systems Engineering},
	author = {Carnein, Matthias and Trautmann, Heike},
	month = jun,
	year = {2019},
	pages = {277--297},
}

Analyzing data streams has received considerable attention over the past decades due to the widespread usage of sensors, social media and other streaming data sources. A core research area in this field is stream clustering which aims to recognize patterns in an unordered, infinite and evolving stream of observations. Clustering can be a crucial support in decision making, since it aims for an optimized aggregated representation of a continuous data stream over time and allows to identify patterns in large and high-dimensional data. A multitude of algorithms and approaches has been developed that are able to find and maintain clusters over time in the challenging streaming scenario. This survey explores, summarizes and categorizes a total of 51 stream clustering algorithms and identifies core research threads over the past decades. In particular, it identifies categories of algorithms based on distance thresholds, density grids and statistical models as well as algorithms for high dimensional data. Furthermore, it discusses applications scenarios, available software and how to configure stream clustering algorithms. This survey is considerably more extensive than comparable studies, more up-to-date and highlights how concepts are interrelated and have been developed over time.

@inproceedings{krishnamurthy_application_2019,
	title = {Application of {Machine} {Learning} and {Spatial} {Bootstrapping} to {Image} {Processing} for {Predictive} {Maintenance}},
	doi = {10.1109/BigData47090.2019.9006439},
	abstract = {Image processing and machine learning have become valuable tools for predictive maintenance applications for a wide variety of industrial and commercial components. We present a novel light transmission image processing methodology utilizing statistical distance algorithms (Wasserstein distance (WD), Kolmogorov-Smirnov statistic (K-S)) for physical attribute correlation combined with Bayesian linear regression to estimate wear level and lifetime prediction for air filters. Robustness of this machine learning algorithm was evaluated using spatial block bootstrapping to generate synthetic training data to estimate the 95\% prediction interval for air filter lifetime. Validation of this lifetime prediction was performed using imaging measurements on a test air filter, which showed good agreement with the machine learning model. The proposed machine learning based image analytics framework effectively enables robust predictions of component wear for predictive maintenance.},
	booktitle = {2019 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Krishnamurthy, Vikram and Nezafati, Kusha and Singh, Vikrant},
	month = dec,
	year = {2019},
	keywords = {Image Processing, Imaging, Kolmogorov-Smirnov statistic, Machine Learning, Machine learning, Measurement, Predictive Maintenance, Predictive maintenance, Sensors, Spatial Bootstrapping, Training, Training data, Wasserstein Distance},
	pages = {4395--4401},
}

Image processing and machine learning have become valuable tools for predictive maintenance applications for a wide variety of industrial and commercial components. We present a novel light transmission image processing methodology utilizing statistical distance algorithms (Wasserstein distance (WD), Kolmogorov-Smirnov statistic (K-S)) for physical attribute correlation combined with Bayesian linear regression to estimate wear level and lifetime prediction for air filters. Robustness of this machine learning algorithm was evaluated using spatial block bootstrapping to generate synthetic training data to estimate the 95% prediction interval for air filter lifetime. Validation of this lifetime prediction was performed using imaging measurements on a test air filter, which showed good agreement with the machine learning model. The proposed machine learning based image analytics framework effectively enables robust predictions of component wear for predictive maintenance.

@article{zang_methods_2019,
	title = {Methods for fault diagnosis of high-speed railways: {A} review},
	volume = {233},
	issn = {1748-006X},
	shorttitle = {Methods for fault diagnosis of high-speed railways},
	url = {https://doi.org/10.1177/1748006X18823932},
	doi = {10.1177/1748006X18823932},
	abstract = {High-speed railways have a high demand for safety, but they are complex systems when it comes to fault diagnosis. The failure propagation path is difficult to trace which makes it hard to detect and identify a fault in the traditional way like signal-based methods. In recent years, artificial intelligence methods have been successfully applied in system health diagnosis and prognosis. Fault diagnosis methods based on artificial intelligence methods provide a new inspiration for fault diagnosis in the high-speed railway systems. In this article, the current research status of fault diagnosis was introduced, and the practical application of fault diagnosis methods in high-speed railways was summarized. Then taking the train control system as an example, fault diagnosis based on the artificial intelligence methods was discussed using several case studies; the results proved that the fusion of different methods has the potential to improve the diagnostic accuracy. Finally, the future research direction of fault diagnosis for high-speed railways was proposed.},
	language = {en},
	number = {5},
	urldate = {2022-03-05},
	journal = {Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability},
	author = {Zang, Yu and Shangguan, Wei and Cai, Baigen and Wang, Huashen and Pecht, Michael G},
	month = oct,
	year = {2019},
	note = {Publisher: SAGE Publications},
	keywords = {High-speed railways, artificial intelligence, case study, development trend, fault diagnosis},
	pages = {908--922},
}

High-speed railways have a high demand for safety, but they are complex systems when it comes to fault diagnosis. The failure propagation path is difficult to trace which makes it hard to detect and identify a fault in the traditional way like signal-based methods. In recent years, artificial intelligence methods have been successfully applied in system health diagnosis and prognosis. Fault diagnosis methods based on artificial intelligence methods provide a new inspiration for fault diagnosis in the high-speed railway systems. In this article, the current research status of fault diagnosis was introduced, and the practical application of fault diagnosis methods in high-speed railways was summarized. Then taking the train control system as an example, fault diagnosis based on the artificial intelligence methods was discussed using several case studies; the results proved that the fusion of different methods has the potential to improve the diagnostic accuracy. Finally, the future research direction of fault diagnosis for high-speed railways was proposed.

@article{ding_intelligent_2019,
	title = {Intelligent fault diagnosis for rotating machinery using deep {Q}-network based health state classification: {A} deep reinforcement learning approach},
	volume = {42},
	issn = {1474-0346},
	shorttitle = {Intelligent fault diagnosis for rotating machinery using deep {Q}-network based health state classification},
	url = {https://www.sciencedirect.com/science/article/pii/S1474034619305506},
	doi = {10.1016/j.aei.2019.100977},
	abstract = {Fault diagnosis methods for rotating machinery have always been a hot research topic, and artificial intelligence-based approaches have attracted increasing attention from both researchers and engineers. Among those related studies and methods, artificial neural networks, especially deep learning-based methods, are widely used to extract fault features or classify fault features obtained by other signal processing techniques. Although such methods could solve the fault diagnosis problems of rotating machinery, there are still two deficiencies. (1) Unable to establish direct linear or non-linear mapping between raw data and the corresponding fault modes, the performance of such fault diagnosis methods highly depends on the quality of the extracted features. (2) The optimization of neural network architecture and parameters, especially for deep neural networks, requires considerable manual modification and expert experience, which limits the applicability and generalization of such methods. As a remarkable breakthrough in artificial intelligence, AlphaGo, a representative achievement of deep reinforcement learning, provides inspiration and direction for the aforementioned shortcomings. Combining the advantages of deep learning and reinforcement learning, deep reinforcement learning is able to build an end-to-end fault diagnosis architecture that can directly map raw fault data to the corresponding fault modes. Thus, based on deep reinforcement learning, a novel intelligent diagnosis method is proposed that is able to overcome the shortcomings of the aforementioned diagnosis methods. Validation tests of the proposed method are carried out using datasets of two types of rotating machinery, rolling bearings and hydraulic pumps, which contain a large number of measured raw vibration signals under different health states and working conditions. The diagnosis results show that the proposed method is able to obtain intelligent fault diagnosis agents that can mine the relationships between the raw vibration signals and fault modes autonomously and effectively. Considering that the learning process of the proposed method depends only on the replayed memories of the agent and the overall rewards, which represent much weaker feedback than that obtained by the supervised learning-based method, the proposed method is promising in establishing a general fault diagnosis architecture for rotating machinery.},
	language = {en},
	urldate = {2022-03-03},
	journal = {Advanced Engineering Informatics},
	author = {Ding, Yu and Ma, Liang and Ma, Jian and Suo, Mingliang and Tao, Laifa and Cheng, Yujie and Lu, Chen},
	month = oct,
	year = {2019},
	keywords = {Deep Q-network, Deep reinforcement learning, Fault diagnosis, Rotating machinery},
	pages = {100977},
}

Fault diagnosis methods for rotating machinery have always been a hot research topic, and artificial intelligence-based approaches have attracted increasing attention from both researchers and engineers. Among those related studies and methods, artificial neural networks, especially deep learning-based methods, are widely used to extract fault features or classify fault features obtained by other signal processing techniques. Although such methods could solve the fault diagnosis problems of rotating machinery, there are still two deficiencies. (1) Unable to establish direct linear or non-linear mapping between raw data and the corresponding fault modes, the performance of such fault diagnosis methods highly depends on the quality of the extracted features. (2) The optimization of neural network architecture and parameters, especially for deep neural networks, requires considerable manual modification and expert experience, which limits the applicability and generalization of such methods. As a remarkable breakthrough in artificial intelligence, AlphaGo, a representative achievement of deep reinforcement learning, provides inspiration and direction for the aforementioned shortcomings. Combining the advantages of deep learning and reinforcement learning, deep reinforcement learning is able to build an end-to-end fault diagnosis architecture that can directly map raw fault data to the corresponding fault modes. Thus, based on deep reinforcement learning, a novel intelligent diagnosis method is proposed that is able to overcome the shortcomings of the aforementioned diagnosis methods. Validation tests of the proposed method are carried out using datasets of two types of rotating machinery, rolling bearings and hydraulic pumps, which contain a large number of measured raw vibration signals under different health states and working conditions. The diagnosis results show that the proposed method is able to obtain intelligent fault diagnosis agents that can mine the relationships between the raw vibration signals and fault modes autonomously and effectively. Considering that the learning process of the proposed method depends only on the replayed memories of the agent and the overall rewards, which represent much weaker feedback than that obtained by the supervised learning-based method, the proposed method is promising in establishing a general fault diagnosis architecture for rotating machinery.

@inproceedings{zhang_equipment_2019,
	address = {Cham},
	series = {Lecture {Notes} in {Computer} {Science}},
	title = {Equipment {Health} {Indicator} {Learning} {Using} {Deep} {Reinforcement} {Learning}},
	isbn = {978-3-030-10997-4},
	doi = {10.1007/978-3-030-10997-4_30},
	abstract = {Predictive Maintenance (PdM) is gaining popularity in industrial operations as it leverages the power of Machine Learning and Internet of Things (IoT) to predict the future health status of equipment. Health Indicator Learning (HIL) plays an important role in PdM as it learns a health curve representing the health conditions of equipment over time, so that health degradation is visually monitored and optimal planning can be performed accordingly to minimize the equipment downtime. However, HIL is a hard problem due to the fact that there is usually no way to access the actual health of the equipment during most of its operation. Traditionally, HIL is addressed by hand-crafting domain-specific performance indicators or through physical modeling, which is expensive and inapplicable for some industries. In this paper, we propose a purely data-driven approach for solving the HIL problem based on Deep Reinforcement Learning (DRL). Our key insight is that the HIL problem can be mapped to a credit assignment problem. Then DRL learns from failures by naturally backpropagating the credit of failures into intermediate states. In particular, given the observed time series of sensor, operating and event (failure) data, we learn a sequence of health indicators that represent the underlying health conditions of physical equipment. We demonstrate that the proposed methods significantly outperform the state-of-the-art methods for HIL and provide explainable insights about the equipment health. In addition, we propose the use of the learned health indicators to predict when the equipment is going to reach its end-of-life, and demonstrate how an explainable health curve is way more useful for a decision maker than a single-number prediction by a black-box model. The proposed approach has a great potential in a broader range of systems (e.g., economical and biological) as a general framework for the automatic learning of the underlying performance of complex systems.},
	language = {en},
	booktitle = {Machine {Learning} and {Knowledge} {Discovery} in {Databases}},
	publisher = {Springer International Publishing},
	author = {Zhang, Chi and Gupta, Chetan and Farahat, Ahmed and Ristovski, Kosta and Ghosh, Dipanjan},
	editor = {Brefeld, Ulf and Curry, Edward and Daly, Elizabeth and MacNamee, Brian and Marascu, Alice and Pinelli, Fabio and Berlingerio, Michele and Hurley, Neil},
	year = {2019},
	keywords = {Deep Reinforcement Learning, Health indicator learning, Predictive Maintenance},
	pages = {488--504},
}

Predictive Maintenance (PdM) is gaining popularity in industrial operations as it leverages the power of Machine Learning and Internet of Things (IoT) to predict the future health status of equipment. Health Indicator Learning (HIL) plays an important role in PdM as it learns a health curve representing the health conditions of equipment over time, so that health degradation is visually monitored and optimal planning can be performed accordingly to minimize the equipment downtime. However, HIL is a hard problem due to the fact that there is usually no way to access the actual health of the equipment during most of its operation. Traditionally, HIL is addressed by hand-crafting domain-specific performance indicators or through physical modeling, which is expensive and inapplicable for some industries. In this paper, we propose a purely data-driven approach for solving the HIL problem based on Deep Reinforcement Learning (DRL). Our key insight is that the HIL problem can be mapped to a credit assignment problem. Then DRL learns from failures by naturally backpropagating the credit of failures into intermediate states. In particular, given the observed time series of sensor, operating and event (failure) data, we learn a sequence of health indicators that represent the underlying health conditions of physical equipment. We demonstrate that the proposed methods significantly outperform the state-of-the-art methods for HIL and provide explainable insights about the equipment health. In addition, we propose the use of the learned health indicators to predict when the equipment is going to reach its end-of-life, and demonstrate how an explainable health curve is way more useful for a decision maker than a single-number prediction by a black-box model. The proposed approach has a great potential in a broader range of systems (e.g., economical and biological) as a general framework for the automatic learning of the underlying performance of complex systems.

@inproceedings{tian_concept_2019,
	address = {New York, NY, USA},
	series = {{CIKM} '19},
	title = {Concept {Drift} {Adaption} for {Online} {Anomaly} {Detection} in {Structural} {Health} {Monitoring}},
	isbn = {978-1-4503-6976-3},
	url = {https://doi.org/10.1145/3357384.3357816},
	doi = {10.1145/3357384.3357816},
	abstract = {Despite its success for anomaly detection in the scenario where only data representing normal behavior are available, one-class support vector machine (OCSVM) still has challenge in dealing with non-stationary data stream, where the underlying distributions of data are time-varying. Existing OCSVM-based online learning methods incrementally update the model to address the challenge, however, they solely rely on the location relationship between a test sample and error support vectors. To better accommodate normal behavior evolution, online anomaly detection in non-stationary data stream is formulated as a concept drift adaptation problem in this paper. It is proposed that OCSVM-based incremental learning is only performed in the case of a normal drift. For an incoming sample, its relative relationship with three sets of vectors in OCSVM, namely margin support vectors, error support vectors, and reserve vectors is fully utilized to estimate whether a normal drift is emerging. Extensive experiments in the field of structural health monitoring have been conducted and the results have shown that the proposed simple approach outperforms the existing OCSVM-based online learning algorithms for anomaly detection.},
	urldate = {2021-03-26},
	booktitle = {Proceedings of the 28th {ACM} {International} {Conference} on {Information} and {Knowledge} {Management}},
	publisher = {Association for Computing Machinery},
	author = {Tian, Hongda and Khoa, Nguyen Lu Dang and Anaissi, Ali and Wang, Yang and Chen, Fang},
	month = nov,
	year = {2019},
	keywords = {anomaly detection, concept drift, data stream, ecml, incremental/online learning, one-class support vector machine},
	pages = {2813--2821},
}

Despite its success for anomaly detection in the scenario where only data representing normal behavior are available, one-class support vector machine (OCSVM) still has challenge in dealing with non-stationary data stream, where the underlying distributions of data are time-varying. Existing OCSVM-based online learning methods incrementally update the model to address the challenge, however, they solely rely on the location relationship between a test sample and error support vectors. To better accommodate normal behavior evolution, online anomaly detection in non-stationary data stream is formulated as a concept drift adaptation problem in this paper. It is proposed that OCSVM-based incremental learning is only performed in the case of a normal drift. For an incoming sample, its relative relationship with three sets of vectors in OCSVM, namely margin support vectors, error support vectors, and reserve vectors is fully utilized to estimate whether a normal drift is emerging. Extensive experiments in the field of structural health monitoring have been conducted and the results have shown that the proposed simple approach outperforms the existing OCSVM-based online learning algorithms for anomaly detection.

@article{cheng_machine_2019,
	title = {Machine {Health} {Monitoring} {Using} {Adaptive} {Kernel} {Spectral} {Clustering} and {Deep} {Long} {Short}-{Term} {Memory} {Recurrent} {Neural} {Networks}},
	volume = {15},
	issn = {1941-0050},
	doi = {10.1109/TII.2018.2866549},
	abstract = {Machine health monitoring is of great importance in industrial informatics field. Recently, deep learning methods applied to machine health monitoring have been proven effective. However, the existing methods face enormous difficulties in extracting heterogeneous features indicating the variation until failure and revealing the inherent high-dimensional features of massive signals, which affect the accuracy and efficiency of machine health monitoring. In this paper, a novel data-driven machine health monitoring method is proposed using adaptive kernel spectral clustering (AKSC) and deep long short-term memory recurrent neural networks (LSTM-RNN). This method include three steps: First, features in the time domain, frequency domain, and time-frequency domain are, respectively, extracted from massive measured signals. And, an Euclidean distance based algorithm is designed to select degradation features. Second, the AKSC algorithm is introduced to adaptively identify machine anomaly behaviors from multiple degradation features. Third, a new deep learning model (LSTM-RNN) is constructed to update and predict the failure time of the machine. The effectiveness of the proposed method is validated using a set of test-to-failure experimental data. The results show that the performance of the proposed method is competitive with other existing methods.},
	number = {2},
	journal = {IEEE Transactions on Industrial Informatics},
	author = {Cheng, Yiwei and Zhu, Haiping and Wu, Jun and Shao, Xinyu},
	month = feb,
	year = {2019},
	note = {Conference Name: IEEE Transactions on Industrial Informatics},
	keywords = {Adaptive kernel spectral clustering (AKSC), Degradation, Feature extraction, Frequency-domain analysis, Machine learning, Monitoring, Time-domain analysis, anomaly detection, deep long short-term memory recurrent neural networks (LSTM-RNN), ecml, failure prognostics, machine health monitoring},
	pages = {987--997},
}

Machine health monitoring is of great importance in industrial informatics field. Recently, deep learning methods applied to machine health monitoring have been proven effective. However, the existing methods face enormous difficulties in extracting heterogeneous features indicating the variation until failure and revealing the inherent high-dimensional features of massive signals, which affect the accuracy and efficiency of machine health monitoring. In this paper, a novel data-driven machine health monitoring method is proposed using adaptive kernel spectral clustering (AKSC) and deep long short-term memory recurrent neural networks (LSTM-RNN). This method include three steps: First, features in the time domain, frequency domain, and time-frequency domain are, respectively, extracted from massive measured signals. And, an Euclidean distance based algorithm is designed to select degradation features. Second, the AKSC algorithm is introduced to adaptively identify machine anomaly behaviors from multiple degradation features. Third, a new deep learning model (LSTM-RNN) is constructed to update and predict the failure time of the machine. The effectiveness of the proposed method is validated using a set of test-to-failure experimental data. The results show that the performance of the proposed method is competitive with other existing methods.

@article{tsui_big_2019,
	title = {Big {Data} {Opportunities}: {System} {Health} {Monitoring} and {Management}},
	volume = {7},
	issn = {2169-3536},
	shorttitle = {Big {Data} {Opportunities}},
	doi = {10.1109/ACCESS.2019.2917891},
	abstract = {The concept of a system, generally defined as an organized set of detailed methods, procedures, and routines that are created to carry out a specific activity or solve a specific problem, has been successfully applied to many domains, ranging from mechanical systems to public health. System health monitoring and management (SHMM) refers to the framework of continuous surveillance, analysis, and interpretation of relevant data for system maintenance, management, and strategic planning. This framework is essential to ensure that an entire system is stable and under control. A fundamental problem in SHMM is the optimal use of correlated active and passive data in tasks including prediction and forecasting, monitoring and surveillance, fault detection and diagnostics, engineering management, and supply chain management. In this paper, we provide a new perspective on SHMM in a big data environment, discuss its relationship with other disciplines, and present several of its applications to complex systems.},
	journal = {IEEE Access},
	author = {Tsui, Kwok Leung and Zhao, Yang and Wang, Dong},
	year = {2019},
	note = {Conference Name: IEEE Access},
	keywords = {Active and passive data, Big Data, Complex systems, Data analysis, Data models, Informatics, Surveillance, big data, complex systems, ecml, system health monitoring and management},
	pages = {68853--68867},
}

The concept of a system, generally defined as an organized set of detailed methods, procedures, and routines that are created to carry out a specific activity or solve a specific problem, has been successfully applied to many domains, ranging from mechanical systems to public health. System health monitoring and management (SHMM) refers to the framework of continuous surveillance, analysis, and interpretation of relevant data for system maintenance, management, and strategic planning. This framework is essential to ensure that an entire system is stable and under control. A fundamental problem in SHMM is the optimal use of correlated active and passive data in tasks including prediction and forecasting, monitoring and surveillance, fault detection and diagnostics, engineering management, and supply chain management. In this paper, we provide a new perspective on SHMM in a big data environment, discuss its relationship with other disciplines, and present several of its applications to complex systems.

@article{feng_online_2019,
	title = {Online {State}-of-{Health} {Estimation} for {Li}-{Ion} {Battery} {Using} {Partial} {Charging} {Segment} {Based} on {Support} {Vector} {Machine}},
	volume = {68},
	issn = {1939-9359},
	doi = {10.1109/TVT.2019.2927120},
	abstract = {The online estimation of battery state-of-health (SOH) is an ever significant issue for the intelligent energy management of the autonomous electric vehicles. Machine-learning based approaches are promising for the online SOH estimation. This paper proposes a machine-learning based algorithm for the online SOH estimation of Li-ion battery. A predictive diagnosis model used in the algorithm is established based on support vector machine (SVM). The support vectors, which reflects the intrinsic characteristics of the Li-ion battery, are determined from the charging data of fresh cells. Furthermore, the coefficients of the SVMs for cells at different SOH are identified once the support vectors are determined. The algorithm functions by comparing partial charging curves with the stored SVMs. Similarity factor is defined after comparison to quantify the SOH of the data under evaluation. The operation of the algorithm only requires partial charging curves, e.g., 15 min charging curves, making fast on-board diagnosis of battery SOH into reality. The partial charging curves can be intercepted from a wide range of voltage section, thereby relieving the pain that there is little chance that the driver charges the battery pack from a predefined state-of-charge. Train, validation, and test are conducted for two commercial Li-ion batteries with Li(NiCoMn)1/3O2 cathode and graphite anode, indicating that the algorithm can estimate the battery SOH with less than 2\% error for 80\% of all the cases, and less than 3\% error for 95\% of all the cases.},
	number = {9},
	journal = {IEEE Transactions on Vehicular Technology},
	author = {Feng, Xuning and Weng, Caihao and He, Xiangming and Han, Xuebing and Lu, Languang and Ren, Dongsheng and Ouyang, Minggao},
	month = sep,
	year = {2019},
	note = {Conference Name: IEEE Transactions on Vehicular Technology},
	keywords = {Calibration, Electric vehicle, Estimation, Lithium-ion batteries, State of charge, Support vector machines, Voltage measurement, batteries, ecml, energy storage, state estimation, state-of-health},
	pages = {8583--8592},
}

The online estimation of battery state-of-health (SOH) is an ever significant issue for the intelligent energy management of the autonomous electric vehicles. Machine-learning based approaches are promising for the online SOH estimation. This paper proposes a machine-learning based algorithm for the online SOH estimation of Li-ion battery. A predictive diagnosis model used in the algorithm is established based on support vector machine (SVM). The support vectors, which reflects the intrinsic characteristics of the Li-ion battery, are determined from the charging data of fresh cells. Furthermore, the coefficients of the SVMs for cells at different SOH are identified once the support vectors are determined. The algorithm functions by comparing partial charging curves with the stored SVMs. Similarity factor is defined after comparison to quantify the SOH of the data under evaluation. The operation of the algorithm only requires partial charging curves, e.g., 15 min charging curves, making fast on-board diagnosis of battery SOH into reality. The partial charging curves can be intercepted from a wide range of voltage section, thereby relieving the pain that there is little chance that the driver charges the battery pack from a predefined state-of-charge. Train, validation, and test are conducted for two commercial Li-ion batteries with Li(NiCoMn)1/3O2 cathode and graphite anode, indicating that the algorithm can estimate the battery SOH with less than 2% error for 80% of all the cases, and less than 3% error for 95% of all the cases.

@article{zhao_deep_2019,
	title = {Deep learning and its applications to machine health monitoring},
	volume = {115},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327018303108},
	doi = {10.1016/j.ymssp.2018.05.050},
	abstract = {Since 2006, deep learning (DL) has become a rapidly growing research direction, redefining state-of-the-art performances in a wide range of areas such as object recognition, image segmentation, speech recognition and machine translation. In modern manufacturing systems, data-driven machine health monitoring is gaining in popularity due to the widespread deployment of low-cost sensors and their connection to the Internet. Meanwhile, deep learning provides useful tools for processing and analyzing these big machinery data. The main purpose of this paper is to review and summarize the emerging research work of deep learning on machine health monitoring. After the brief introduction of deep learning techniques, the applications of deep learning in machine health monitoring systems are reviewed mainly from the following aspects: Auto-encoder (AE) and its variants, Restricted Boltzmann Machines and its variants including Deep Belief Network (DBN) and Deep Boltzmann Machines (DBM), Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). In addition, an experimental study on the performances of these approaches has been conducted, in which the data and code have been online. Finally, some new trends of DL-based machine health monitoring methods are discussed.},
	language = {en},
	urldate = {2022-02-09},
	journal = {Mechanical Systems and Signal Processing},
	author = {Zhao, Rui and Yan, Ruqiang and Chen, Zhenghua and Mao, Kezhi and Wang, Peng and Gao, Robert X.},
	month = jan,
	year = {2019},
	keywords = {Big data, Deep learning, Machine health monitoring, ecml, health monitoring},
	pages = {213--237},
}

Since 2006, deep learning (DL) has become a rapidly growing research direction, redefining state-of-the-art performances in a wide range of areas such as object recognition, image segmentation, speech recognition and machine translation. In modern manufacturing systems, data-driven machine health monitoring is gaining in popularity due to the widespread deployment of low-cost sensors and their connection to the Internet. Meanwhile, deep learning provides useful tools for processing and analyzing these big machinery data. The main purpose of this paper is to review and summarize the emerging research work of deep learning on machine health monitoring. After the brief introduction of deep learning techniques, the applications of deep learning in machine health monitoring systems are reviewed mainly from the following aspects: Auto-encoder (AE) and its variants, Restricted Boltzmann Machines and its variants including Deep Belief Network (DBN) and Deep Boltzmann Machines (DBM), Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). In addition, an experimental study on the performances of these approaches has been conducted, in which the data and code have been online. Finally, some new trends of DL-based machine health monitoring methods are discussed.

@article{luo_development_2019,
	title = {Development of clustering-based sensor fault detection and diagnosis strategy for chilled water system},
	volume = {186},
	issn = {0378-7788},
	url = {https://www.sciencedirect.com/science/article/pii/S0378778818329207},
	doi = {10.1016/j.enbuild.2019.01.006},
	abstract = {This paper presents a new clustering-based sensor fault detection and diagnosis (SFDD) strategy for chilled water system. For data clustering, k-means algorithm was used and the optimal quantity of clusters was determined by Davis-Bouldin value. With the cluster centroid dataset, the featuring centroid score (CS) was determined for the fault-free sensor reading dataset thus the threshold for fault detection could be set. The database for sensor fault detection was then formed. By characterizing the CS patterns of different types of sensor fault, the database for sensor fault diagnosis was generated accordingly. Various sensor fault types could be handled, including bias, drift, precision degradation and complete failure. In this study, the developed SFDD strategy was applied to the sensor of primary chilled water return temperature in a water-cooled chilled water system. With the databases of sensor fault detection and diagnosis, the real-time measured sensor readings can be examined. Once sensor fault is detected, the fault type can be confirmed within a day at soonest or 2 days at most. The smallest detectable absolute bias value, absolute drifting rate and precision degradation error could be down to 0.25 °C, 0.025 °C/h and 0.1 °C respectively, demonstrating robustness of the proposed SFDD strategy.},
	language = {en},
	urldate = {2022-01-14},
	journal = {Energy and Buildings},
	author = {Luo, X. J. and Fong, K. F. and Sun, Y. J. and Leung, M. K. H.},
	month = mar,
	year = {2019},
	keywords = {-means clustering, Centroid score, Chilled water system, Data mining, Fault detection and diagnosis, Sensor},
	pages = {17--36},
}

This paper presents a new clustering-based sensor fault detection and diagnosis (SFDD) strategy for chilled water system. For data clustering, k-means algorithm was used and the optimal quantity of clusters was determined by Davis-Bouldin value. With the cluster centroid dataset, the featuring centroid score (CS) was determined for the fault-free sensor reading dataset thus the threshold for fault detection could be set. The database for sensor fault detection was then formed. By characterizing the CS patterns of different types of sensor fault, the database for sensor fault diagnosis was generated accordingly. Various sensor fault types could be handled, including bias, drift, precision degradation and complete failure. In this study, the developed SFDD strategy was applied to the sensor of primary chilled water return temperature in a water-cooled chilled water system. With the databases of sensor fault detection and diagnosis, the real-time measured sensor readings can be examined. Once sensor fault is detected, the fault type can be confirmed within a day at soonest or 2 days at most. The smallest detectable absolute bias value, absolute drifting rate and precision degradation error could be down to 0.25 °C, 0.025 °C/h and 0.1 °C respectively, demonstrating robustness of the proposed SFDD strategy.

@article{zhao_pyod_2019,
	title = {{PyOD}: {A} {Python} {Toolbox} for {Scalable} {Outlier} {Detection}},
	shorttitle = {{PyOD}},
	url = {http://arxiv.org/abs/1901.01588},
	abstract = {PyOD is an open-source Python toolbox for performing scalable outlier detection on multivariate data. Uniquely, it provides access to a wide range of outlier detection algorithms, including established outlier ensembles and more recent neural network-based approaches, under a single, well-documented API designed for use by both practitioners and researchers. With robustness and scalability in mind, best practices such as unit testing, continuous integration, code coverage, maintainability checks, interactive examples and parallelization are emphasized as core components in the toolbox's development. PyOD is compatible with both Python 2 and 3 and can be installed through Python Package Index (PyPI) or https://github.com/yzhao062/pyod.},
	urldate = {2022-01-10},
	journal = {arXiv:1901.01588 [cs, stat]},
	author = {Zhao, Yue and Nasrullah, Zain and Li, Zheng},
	month = jun,
	year = {2019},
	note = {arXiv: 1901.01588},
	keywords = {Computer Science - Information Retrieval, Computer Science - Machine Learning, Statistics - Machine Learning},
}

PyOD is an open-source Python toolbox for performing scalable outlier detection on multivariate data. Uniquely, it provides access to a wide range of outlier detection algorithms, including established outlier ensembles and more recent neural network-based approaches, under a single, well-documented API designed for use by both practitioners and researchers. With robustness and scalability in mind, best practices such as unit testing, continuous integration, code coverage, maintainability checks, interactive examples and parallelization are emphasized as core components in the toolbox's development. PyOD is compatible with both Python 2 and 3 and can be installed through Python Package Index (PyPI) or https://github.com/yzhao062/pyod.

@misc{chen_tennessee_2019,
	title = {Tennessee {Eastman} simulation dataset},
	url = {https://dx.doi.org/10.21227/4519-z502},
	doi = {10.21227/4519-z502},
	abstract = {Tenessee Eastman (TE) process simulates actual chemical processes and is widely used as a benchmark in test fault diagnosis and process control. The overall process consists of five operating units: reactor, condenser, vapor-liquid separator, recycle compressor and product stripper. It has standard training and test data sets for fault detection and diagnosis, classification, etc. Each data set is under different operating conditions。},
	language = {en},
	urldate = {2021-12-08},
	author = {Chen, Xiaolu},
	year = {2019},
	note = {Publication Title: IEEE Dataport
Publisher: IEEE},
}

Tenessee Eastman (TE) process simulates actual chemical processes and is widely used as a benchmark in test fault diagnosis and process control. The overall process consists of five operating units: reactor, condenser, vapor-liquid separator, recycle compressor and product stripper. It has standard training and test data sets for fault detection and diagnosis, classification, etc. Each data set is under different operating conditions。

@article{guan_particle_2019,
	title = {Particle swarm {Optimized} {Density}-based {Clustering} and {Classification}: {Supervised} and unsupervised learning approaches},
	volume = {44},
	issn = {2210-6502},
	shorttitle = {Particle swarm {Optimized} {Density}-based {Clustering} and {Classification}},
	url = {https://www.sciencedirect.com/science/article/pii/S2210650217302638},
	doi = {10.1016/j.swevo.2018.09.008},
	abstract = {Two pattern recognition technologies in the field of machine learning, clustering and classification, have been applied in many domains. Density-based clustering is an essential clustering algorithm. The best known density-based clustering method is Density-Based Spatial Clustering of Applications with Noise (DBSCAN), which can find arbitrary shaped clusters in datasets. DBSCAN has three drawbacks: firstly, the parameters for DBSCAN are hard to set; secondly, the number of clusters cannot be controlled by the users; and thirdly, DBSCAN cannot directly be used as a classifier. In this paper a novel Particle swarm Optimized Density-based Clustering and Classification (PODCC) is proposed, designed to offset the drawbacks of DBSCAN. Particle Swarm Optimization (PSO), a widely used Evolutionary and Swarm Algorithm (ESA), has been applied in optimization problems in different research domains including data analytics. In PODCC, a variant of PSO, SPSO-2011, is used to search the parameter space so as to identify the best parameters for density-based clustering and classification. PODCC can function in terms of both Supervised and Unsupervised Learnings by applying the appropriate fitness functions proposed in this paper. With the proposed fitness function, users can set the number of clusters as input for PODCC. The proposed method was evaluated by testing ten synthetic datasets and ten benchmarking datasets selected from various open sources. The experimental results indicate that the proposed PODCC can perform better than some established methods, especially with respect to imbalanced datasets.},
	language = {en},
	urldate = {2021-11-29},
	journal = {Swarm and Evolutionary Computation},
	author = {Guan, Chun and Yuen, Kevin Kam Fung and Coenen, Frans},
	month = feb,
	year = {2019},
	keywords = {Classification, Density-based clustering, Imbalanced dataset, Parameter tuning, Particle Swarm Optimization},
	pages = {876--896},
}

Two pattern recognition technologies in the field of machine learning, clustering and classification, have been applied in many domains. Density-based clustering is an essential clustering algorithm. The best known density-based clustering method is Density-Based Spatial Clustering of Applications with Noise (DBSCAN), which can find arbitrary shaped clusters in datasets. DBSCAN has three drawbacks: firstly, the parameters for DBSCAN are hard to set; secondly, the number of clusters cannot be controlled by the users; and thirdly, DBSCAN cannot directly be used as a classifier. In this paper a novel Particle swarm Optimized Density-based Clustering and Classification (PODCC) is proposed, designed to offset the drawbacks of DBSCAN. Particle Swarm Optimization (PSO), a widely used Evolutionary and Swarm Algorithm (ESA), has been applied in optimization problems in different research domains including data analytics. In PODCC, a variant of PSO, SPSO-2011, is used to search the parameter space so as to identify the best parameters for density-based clustering and classification. PODCC can function in terms of both Supervised and Unsupervised Learnings by applying the appropriate fitness functions proposed in this paper. With the proposed fitness function, users can set the number of clusters as input for PODCC. The proposed method was evaluated by testing ten synthetic datasets and ten benchmarking datasets selected from various open sources. The experimental results indicate that the proposed PODCC can perform better than some established methods, especially with respect to imbalanced datasets.

@article{helske_mixture_2019,
	title = {Mixture {Hidden} {Markov} {Models} for {Sequence} {Data}: {The} {seqHMM} {Package} in {R}},
	volume = {88},
	copyright = {Copyright (c) 2019 Satu Helske, Jouni Helske},
	issn = {1548-7660},
	shorttitle = {Mixture {Hidden} {Markov} {Models} for {Sequence} {Data}},
	url = {https://doi.org/10.18637/jss.v088.i03},
	doi = {10.18637/jss.v088.i03},
	abstract = {Sequence analysis is being more and more widely used for the analysis of social sequences and other multivariate categorical time series data. However, it is often complex to describe, visualize, and compare large sequence data, especially when there are multiple parallel sequences per subject. Hidden (latent) Markov models (HMMs) are able to detect underlying latent structures and they can be used in various longitudinal settings: to account for measurement error, to detect unobservable states, or to compress information across several types of observations. Extending to mixture hidden Markov models (MHMMs) allows clustering data into homogeneous subsets, with or without external covariates. The seqHMM package in R is designed for the efficient modeling of sequences and other categorical time series data containing one or multiple subjects with one or multiple interdependent sequences using HMMs and MHMMs. Also other restricted variants of the MHMM can be fitted, e.g., latent class models, Markov models, mixture Markov models, or even ordinary multinomial regression models with suitable parameterization of the HMM. Good graphical presentations of data and models are useful during the whole analysis process from the first glimpse at the data to model fitting and presentation of results. The package provides easy options for plotting parallel sequence data, and proposes visualizing HMMs as directed graphs.},
	language = {en},
	urldate = {2021-11-17},
	journal = {Journal of Statistical Software},
	author = {Helske, Satu and Helske, Jouni},
	month = jan,
	year = {2019},
	keywords = {R},
	pages = {1--32},
}

Sequence analysis is being more and more widely used for the analysis of social sequences and other multivariate categorical time series data. However, it is often complex to describe, visualize, and compare large sequence data, especially when there are multiple parallel sequences per subject. Hidden (latent) Markov models (HMMs) are able to detect underlying latent structures and they can be used in various longitudinal settings: to account for measurement error, to detect unobservable states, or to compress information across several types of observations. Extending to mixture hidden Markov models (MHMMs) allows clustering data into homogeneous subsets, with or without external covariates. The seqHMM package in R is designed for the efficient modeling of sequences and other categorical time series data containing one or multiple subjects with one or multiple interdependent sequences using HMMs and MHMMs. Also other restricted variants of the MHMM can be fitted, e.g., latent class models, Markov models, mixture Markov models, or even ordinary multinomial regression models with suitable parameterization of the HMM. Good graphical presentations of data and models are useful during the whole analysis process from the first glimpse at the data to model fitting and presentation of results. The package provides easy options for plotting parallel sequence data, and proposes visualizing HMMs as directed graphs.

@article{parpinelli_review_2019,
	title = {A review of techniques for online control of parameters in swarm intelligence and evolutionary computation algorithms},
	copyright = {Copyright © 2018 Inderscience Enterprises Ltd.},
	url = {https://www.inderscienceonline.com/doi/abs/10.1504/IJBIC.2019.097731},
	abstract = {The two major groups representing biologically inspired algorithms are swarm intelligence (SI) and evolutionary computation (EC). Both SI and EC share common features such as the use of stochastic components during the optimisation process and various parameters for configuration. The setup of parameters in swarm and in evolutionary algorithms has an important role in defining their behaviour, guiding the search and biasing the quality of final solutions. In addition, an appropriate setting for the parameters may change during the optimisation process making this task even harder. The present work brings an up-to-date discussion focusing on online parameter control strategies applied in SI and EC. Also, this review analyses and points out the key techniques and algorithms used and suggests some directions for future research.},
	language = {en},
	urldate = {2021-11-11},
	journal = {International Journal of Bio-Inspired Computation},
	author = {Parpinelli, Rafael Stubs and Plichoski, Guilherme Felippe and Silva, Renan Samuel Da and Narloch, Pedro Henrique},
	month = feb,
	year = {2019},
	note = {Publisher: Inderscience Publishers (IEL)},
	keywords = {online, swarm intelligence},
}

The two major groups representing biologically inspired algorithms are swarm intelligence (SI) and evolutionary computation (EC). Both SI and EC share common features such as the use of stochastic components during the optimisation process and various parameters for configuration. The setup of parameters in swarm and in evolutionary algorithms has an important role in defining their behaviour, guiding the search and biasing the quality of final solutions. In addition, an appropriate setting for the parameters may change during the optimisation process making this task even harder. The present work brings an up-to-date discussion focusing on online parameter control strategies applied in SI and EC. Also, this review analyses and points out the key techniques and algorithms used and suggests some directions for future research.

@inproceedings{shahin_input-output_2019,
	title = {Input-{Output} {Hidden} {Markov} {Model} for {Diagnosis} of {Complex} {System}},
	abstract = {Prognosis system state of degradation and estimating its remaining useful life requires the system health assessment. For a correct prognostic, a good diagnostic as health assessment is required. Complex systems are difficult to manage for modeling reasons considering complexity, environmental and operational conditions. This paper deals with a stochastic model for generic modeling purposes and considers operating conditions in order to determine the system health. The proposed model is an Input-Output Hidden Markov Model that is able to model a degradation process of complex systems given operational conditions and allows assessing the system health. Well-known algorithms dedicated to HMM are adapted to IOHMM for multiple observation sequences and inputs.},
	author = {Shahin, Kamrul Islam and Simon, Christophe and Weber, Philippe},
	month = jun,
	year = {2019},
}

Prognosis system state of degradation and estimating its remaining useful life requires the system health assessment. For a correct prognostic, a good diagnostic as health assessment is required. Complex systems are difficult to manage for modeling reasons considering complexity, environmental and operational conditions. This paper deals with a stochastic model for generic modeling purposes and considers operating conditions in order to determine the system health. The proposed model is an Input-Output Hidden Markov Model that is able to model a degradation process of complex systems given operational conditions and allows assessing the system health. Well-known algorithms dedicated to HMM are adapted to IOHMM for multiple observation sequences and inputs.

@article{zhao_condition-based_2019,
	title = {A condition-based opportunistic maintenance strategy for multi-component system},
	volume = {18},
	issn = {1475-9217},
	url = {https://doi.org/10.1177/1475921717751871},
	doi = {10.1177/1475921717751871},
	abstract = {As a new dynamic maintenance strategy, the condition-based opportunistic maintenance strategy for multi-component system is presented in this work. In the strategy, the degeneration of each component is described by Weibull proportional hazards model or Weibull proportional intensity model, and the condition indicator is defined to characterize the operating state of each component. Then, when and how to maintain a component can be confirmed by comparing the value of the condition indicator with that of the maintenance threshold function. Condition-based maintenance will be implemented on a component if the value of its condition indicator exceeds that of its condition-based maintenance threshold function. Meanwhile, opportunistic maintenance will also be implemented on a component if the value of its condition indicator exceeds that of its opportunistic maintenance threshold function. The two maintenance threshold functions can be determined by minimizing maintenance cost. Finally, taking the wind turbine as an example of a multi-component system, simulation analyses are described to validate the feasibility and effectiveness of the condition-based opportunistic maintenance strategy.},
	language = {en},
	number = {1},
	urldate = {2021-10-28},
	journal = {Structural Health Monitoring},
	author = {Zhao, Hongshan and Xu, Fanhao and Liang, Botong and Zhang, Jianping and Song, Peng},
	month = jan,
	year = {2019},
	note = {Publisher: SAGE Publications},
	keywords = {Condition indicator, Multi-component, condition-based maintenance, condition-based opportunistic maintenance, maintenance threshold function, multicomponent, opportunistic maintenance, weibull},
	pages = {270--283},
}

As a new dynamic maintenance strategy, the condition-based opportunistic maintenance strategy for multi-component system is presented in this work. In the strategy, the degeneration of each component is described by Weibull proportional hazards model or Weibull proportional intensity model, and the condition indicator is defined to characterize the operating state of each component. Then, when and how to maintain a component can be confirmed by comparing the value of the condition indicator with that of the maintenance threshold function. Condition-based maintenance will be implemented on a component if the value of its condition indicator exceeds that of its condition-based maintenance threshold function. Meanwhile, opportunistic maintenance will also be implemented on a component if the value of its condition indicator exceeds that of its opportunistic maintenance threshold function. The two maintenance threshold functions can be determined by minimizing maintenance cost. Finally, taking the wind turbine as an example of a multi-component system, simulation analyses are described to validate the feasibility and effectiveness of the condition-based opportunistic maintenance strategy.

@article{nguyen_multi-label_2019,
	title = {Multi-label classification via incremental clustering on an evolving data stream},
	volume = {95},
	issn = {0031-3203},
	url = {https://www.sciencedirect.com/science/article/pii/S0031320319302328},
	doi = {10.1016/j.patcog.2019.06.001},
	abstract = {With the advancement of storage and processing technology, an enormous amount of data is collected on a daily basis in many applications. Nowadays, advanced data analytics have been used to mine the collected data for useful information and make predictions, contributing to the competitive advantages of companies. The increasing data volume, however, has posed many problems to classical batch learning systems, such as the need to retrain the model completely with the newly arrived samples or the impracticality of storing and accessing a large volume of data. This has prompted interest on incremental learning that operates on data streams. In this study, we develop an incremental online multi-label classification (OMLC) method based on a weighted clustering model. The model is made to adapt to the change of data via the decay mechanism in which each sample's weight dwindles away over time. The clustering model therefore always focuses more on newly arrived samples. In the classification process, only clusters whose weights are greater than a threshold (called mature clusters) are employed to assign labels for the samples. In our method, not only is the clustering model incrementally maintained with the revealed ground truth labels of the arrived samples, the number of predicted labels in a sample are also adjusted based on the Hoeffding inequality and the label cardinality. The experimental results show that our method is competitive compared to several well-known benchmark algorithms on six performance measures in both the stationary and the concept drift settings.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Pattern Recognition},
	author = {Nguyen, Tien Thanh and Dang, Manh Truong and Luong, Anh Vu and Liew, Alan Wee-Chung and Liang, Tiancai and McCall, John},
	month = nov,
	year = {2019},
	keywords = {Clustering, Concept drift, Data stream, Incremental learning, Multi-label classification, Online learning, cluster convergence, hoeffding, mature cluster},
	pages = {96--113},
}

With the advancement of storage and processing technology, an enormous amount of data is collected on a daily basis in many applications. Nowadays, advanced data analytics have been used to mine the collected data for useful information and make predictions, contributing to the competitive advantages of companies. The increasing data volume, however, has posed many problems to classical batch learning systems, such as the need to retrain the model completely with the newly arrived samples or the impracticality of storing and accessing a large volume of data. This has prompted interest on incremental learning that operates on data streams. In this study, we develop an incremental online multi-label classification (OMLC) method based on a weighted clustering model. The model is made to adapt to the change of data via the decay mechanism in which each sample's weight dwindles away over time. The clustering model therefore always focuses more on newly arrived samples. In the classification process, only clusters whose weights are greater than a threshold (called mature clusters) are employed to assign labels for the samples. In our method, not only is the clustering model incrementally maintained with the revealed ground truth labels of the arrived samples, the number of predicted labels in a sample are also adjusted based on the Hoeffding inequality and the label cardinality. The experimental results show that our method is competitive compared to several well-known benchmark algorithms on six performance measures in both the stationary and the concept drift settings.

@article{zeni_fixing_2019,
	title = {Fixing {Mislabeling} by {Human} {Annotators} {Leveraging} {Conflict} {Resolution} and {Prior} {Knowledge}},
	volume = {3},
	url = {https://doi.org/10.1145/3314419},
	doi = {10.1145/3314419},
	abstract = {According to the "human in the loop" paradigm, machine learning algorithms can improve when leveraging on human intelligence, usually in the form of labels or annotation from domain experts. However, in the case of research areas such as ubiquitous computing or lifelong learning, where the annotator is not an expert and is continuously asked for feedback, humans can provide significant fractions of incorrect labels. We propose to address this issue in a series of experiments where students are asked to provide information about their behavior via a dedicated mobile application. Their trustworthiness is tested by employing an architecture where the machine uses all its available knowledge to check the correctness of its own and the user labeling to build a uniform confidence measure for both of them to be used when a contradiction arises. The overarching system runs through a series of modes with progressively higher confidence and features a conflict resolution component to settle the inconsistencies. The results are very promising and show the pervasiveness of annotation mistakes, the extreme diversity of the users' behaviors which provides evidence of the impracticality of a uniform fits-it-all solution, and the substantially improved performance of a skeptical supervised learning strategy.},
	number = {1},
	urldate = {2021-10-18},
	journal = {Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies},
	author = {Zeni, Mattia and Zhang, Wanyi and Bignotti, Enrico and Passerini, Andrea and Giunchiglia, Fausto},
	month = mar,
	year = {2019},
	keywords = {Annotation Errors, Collaborative and Social Computing, Ubiquitous and Mobile Devices},
	pages = {32:1--32:23},
}

According to the "human in the loop" paradigm, machine learning algorithms can improve when leveraging on human intelligence, usually in the form of labels or annotation from domain experts. However, in the case of research areas such as ubiquitous computing or lifelong learning, where the annotator is not an expert and is continuously asked for feedback, humans can provide significant fractions of incorrect labels. We propose to address this issue in a series of experiments where students are asked to provide information about their behavior via a dedicated mobile application. Their trustworthiness is tested by employing an architecture where the machine uses all its available knowledge to check the correctness of its own and the user labeling to build a uniform confidence measure for both of them to be used when a contradiction arises. The overarching system runs through a series of modes with progressively higher confidence and features a conflict resolution component to settle the inconsistencies. The results are very promising and show the pervasiveness of annotation mistakes, the extreme diversity of the users' behaviors which provides evidence of the impracticality of a uniform fits-it-all solution, and the substantially improved performance of a skeptical supervised learning strategy.

@article{yang_learning_2019,
	title = {Learning structured and non-redundant representations with deep neural networks},
	volume = {86},
	issn = {0031-3203},
	url = {https://www.sciencedirect.com/science/article/pii/S0031320318303169},
	doi = {10.1016/j.patcog.2018.08.017},
	abstract = {This paper proposes a novel regularizer named Structured Decorrelation Constraint, to address both the generalization and optimization of deep neural networks, including multiple-layer perceptrons and convolutional neural networks. Our proposed regularizer reduces overfitting by breaking the co-adaptions between the neurons with an explicit penalty. As a result, the network is capable of learning non-redundant representations. Meanwhile, the proposed regularizer encourages the networks to learn structured high-level features to aid the networks’ optimization during training. To this end, neurons are constrained to behave obeying a group prior. Our regularizer applies to various types of layers, including fully connected layers, convolutional layers and normalization layers. The loss of our regularizer can be directly minimized along with the network’s classification loss by stochastic gradient descent. Experiments show that the proposed regularizer obviously relieves the overfitting problem of the existing deep networks. It yields much better performance on extensive datasets than the conventional regularizers like Dropout.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Pattern Recognition},
	author = {Yang, Jihai and Xiong, Wei and Li, Shijun and Xu, Chang},
	month = feb,
	year = {2019},
	keywords = {Decorrelation, Deep networks, Overfitting},
	pages = {224--235},
}

This paper proposes a novel regularizer named Structured Decorrelation Constraint, to address both the generalization and optimization of deep neural networks, including multiple-layer perceptrons and convolutional neural networks. Our proposed regularizer reduces overfitting by breaking the co-adaptions between the neurons with an explicit penalty. As a result, the network is capable of learning non-redundant representations. Meanwhile, the proposed regularizer encourages the networks to learn structured high-level features to aid the networks’ optimization during training. To this end, neurons are constrained to behave obeying a group prior. Our regularizer applies to various types of layers, including fully connected layers, convolutional layers and normalization layers. The loss of our regularizer can be directly minimized along with the network’s classification loss by stochastic gradient descent. Experiments show that the proposed regularizer obviously relieves the overfitting problem of the existing deep networks. It yields much better performance on extensive datasets than the conventional regularizers like Dropout.

@article{li_incremental_2019,
	title = {Incremental semi-supervised learning on streaming data},
	volume = {88},
	issn = {0031-3203},
	url = {https://www.sciencedirect.com/science/article/pii/S0031320318303923},
	doi = {10.1016/j.patcog.2018.11.006},
	abstract = {In streaming data classification, most of the existing methods assume that all arrived evolving data are completely labeled. One challenge is that some applications where only small amount of labeled examples are available for training. Incremental semi-supervised learning algorithms have been proposed for regularizing neural networks by incorporating various side information, such as pairwise constraints or user-provided labels. However, it is hard to put them into practice, especially for non-stationary environments due to the effectiveness and parameter sensitivity of such algorithms. In this paper, we propose a novel incremental semi-supervised learning framework on streaming data. Each layer of model is comprised of a generative network, a discriminant structure and the bridge. The generative network uses dynamic feature learning based on autoencoders to learn generative features from streaming data which has been demonstrated its potential in learning latent feature representations. In addition, the discriminant structure regularizes the network construction via building pairwise similarity and dissimilarity constraints. It is also used for facilitating the parameter learning of the generative network. The network and structure are integrated into a joint learning framework and bridged by enforcing the correlation of their parameters, which balances the flexible incorporation of supervision information and numerical tractability for non-stationary environments as well as explores the intrinsic data structure. Moreover, an efficient algorithm is designed to solve the proposed optimization problem and we also give an ensemble method. Particularly, when multiple layers of model are stacked, the performance is significantly boosted. Finally, to validate the effectiveness of the proposed method, extensive experiments are conducted on synthetic and real-life datasets. The experimental results demonstrate that the performance of the proposed algorithms is superior to some state-of-the-art approaches.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Pattern Recognition},
	author = {Li, Yanchao and Wang, Yongli and Liu, Qi and Bi, Cheng and Jiang, Xiaohui and Sun, Shurong},
	month = apr,
	year = {2019},
	keywords = {Classification, Dynamic feature learning, Semi-supervised learning, Streaming data},
	pages = {383--396},
}

In streaming data classification, most of the existing methods assume that all arrived evolving data are completely labeled. One challenge is that some applications where only small amount of labeled examples are available for training. Incremental semi-supervised learning algorithms have been proposed for regularizing neural networks by incorporating various side information, such as pairwise constraints or user-provided labels. However, it is hard to put them into practice, especially for non-stationary environments due to the effectiveness and parameter sensitivity of such algorithms. In this paper, we propose a novel incremental semi-supervised learning framework on streaming data. Each layer of model is comprised of a generative network, a discriminant structure and the bridge. The generative network uses dynamic feature learning based on autoencoders to learn generative features from streaming data which has been demonstrated its potential in learning latent feature representations. In addition, the discriminant structure regularizes the network construction via building pairwise similarity and dissimilarity constraints. It is also used for facilitating the parameter learning of the generative network. The network and structure are integrated into a joint learning framework and bridged by enforcing the correlation of their parameters, which balances the flexible incorporation of supervision information and numerical tractability for non-stationary environments as well as explores the intrinsic data structure. Moreover, an efficient algorithm is designed to solve the proposed optimization problem and we also give an ensemble method. Particularly, when multiple layers of model are stacked, the performance is significantly boosted. Finally, to validate the effectiveness of the proposed method, extensive experiments are conducted on synthetic and real-life datasets. The experimental results demonstrate that the performance of the proposed algorithms is superior to some state-of-the-art approaches.

@article{shao_synchronization-based_2019,
	title = {Synchronization-based clustering on evolving data stream},
	volume = {501},
	issn = {0020-0255},
	url = {https://www.sciencedirect.com/science/article/pii/S0020025518307400},
	doi = {10.1016/j.ins.2018.09.035},
	abstract = {Clustering streams of data is of increasing importance in many applications. In this paper, we propose a new synchronization-based clustering approach for evolving data streams, called SyncTree, which maintains all micro-clusters at different levels of granularity depending upon the data recency. Instead of using a sliding window or decay function to focus on recent data, SyncTree summarizes all continuously-arriving objects as synchronized micro-clusters sequentially in a batch fashion. Owing to the powerful concept of synchronization, the derived micro-clusters truly reflect the intrinsic cluster structure rather than summarize statistics of data, and old micro-clusters can be intuitively summarized at a higher level by iterative clustering to fit memory constraints. Building upon the hierarchical micro-clusters, SyncTree allows investigating the cluster structure of the data stream between any two time stamps in the past, and also provides a principled way to analyze the cluster evolution. Empirical results demonstrate that our method has good performance compared to state-of-the-art algorithms.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Information Sciences},
	author = {Shao, Junming and Tan, Yue and Gao, Lianli and Yang, Qinli and Plant, Claudia and Assent, Ira},
	month = oct,
	year = {2019},
	keywords = {Clustering, Data stream, Evolving analysis, Synchronization},
	pages = {573--587},
}

Clustering streams of data is of increasing importance in many applications. In this paper, we propose a new synchronization-based clustering approach for evolving data streams, called SyncTree, which maintains all micro-clusters at different levels of granularity depending upon the data recency. Instead of using a sliding window or decay function to focus on recent data, SyncTree summarizes all continuously-arriving objects as synchronized micro-clusters sequentially in a batch fashion. Owing to the powerful concept of synchronization, the derived micro-clusters truly reflect the intrinsic cluster structure rather than summarize statistics of data, and old micro-clusters can be intuitively summarized at a higher level by iterative clustering to fit memory constraints. Building upon the hierarchical micro-clusters, SyncTree allows investigating the cluster structure of the data stream between any two time stamps in the past, and also provides a principled way to analyze the cluster evolution. Empirical results demonstrate that our method has good performance compared to state-of-the-art algorithms.

@article{xu_density-based_2019,
	title = {A density-based competitive data stream clustering network with self-adaptive distance metric},
	volume = {110},
	issn = {0893-6080},
	url = {https://www.sciencedirect.com/science/article/pii/S0893608018303307},
	doi = {10.1016/j.neunet.2018.11.008},
	abstract = {Data stream clustering is a branch of clustering where patterns are processed as an ordered sequence. In this paper, we propose an unsupervised learning neural network named Density Based Self Organizing Incremental Neural Network(DenSOINN) for data stream clustering tasks. DenSOINN is a self organizing competitive network that grows incrementally to learn suitable nodes to fit the distribution of learning data, combining online unsupervised learning and topology learning by means of competitive Hebbian learning rule. By adopting a density-based clustering mechanism, DenSOINN discovers arbitrarily shaped clusters and diminishes the negative effect of noise. In addition, we adopt a self-adaptive distance framework to obtain good performance for learning unnormalized input data. Experiments show that the DenSOINN can achieve high standard performance comparing to state-of-the-art methods.},
	language = {en},
	urldate = {2021-10-18},
	journal = {Neural Networks},
	author = {Xu, Baile and Shen, Furao and Zhao, Jinxi},
	month = feb,
	year = {2019},
	keywords = {Clustering methods, Competitive neural networks, Stream learning, Unsupervised learning},
	pages = {141--158},
}

Data stream clustering is a branch of clustering where patterns are processed as an ordered sequence. In this paper, we propose an unsupervised learning neural network named Density Based Self Organizing Incremental Neural Network(DenSOINN) for data stream clustering tasks. DenSOINN is a self organizing competitive network that grows incrementally to learn suitable nodes to fit the distribution of learning data, combining online unsupervised learning and topology learning by means of competitive Hebbian learning rule. By adopting a density-based clustering mechanism, DenSOINN discovers arbitrarily shaped clusters and diminishes the negative effect of noise. In addition, we adopt a self-adaptive distance framework to obtain good performance for learning unnormalized input data. Experiments show that the DenSOINN can achieve high standard performance comparing to state-of-the-art methods.

@article{breitenstein_approaching_2019,
	title = {Approaching neural net feature interpretation using stacked autoencoders: gene expression profiling of systemic lupus erythematosus patients},
	volume = {2019},
	issn = {2153-4063},
	shorttitle = {Approaching neural net feature interpretation using stacked autoencoders},
	url = {https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6568105/},
	abstract = {Systemic lupus erythematosus (SLE) is a rare, autoimmune disorder known to affect most organ sites. Complicating clinical management is a poorly differentiated, heterogenous SLE disease state. While some small molecule drugs and biologics are available for treatment, additional therapeutic options are needed. Parsing complex biological signatures using powerful, yet human interpretable approaches is critical to advancing our understanding of SLE etiology and identifying therapeutic repositioning opportunities. To approach this goal, we developed a semi-supervised deep neural network pipeline for gene expression profiling of SLE patients and subsequent characterization of individual gene features. Our pipeline performed exemplar multinomial classification of SLE patients in independent balanced validation (F1=0.956) and unbalanced, under-powered testing (F1=0.944) cohorts. A stacked autoencoder disambiguated individual feature representativeness by regenerating an input-like(A ‘) feature matrix. A to A’ comparisons suggest the top associated features to be key features in gene expression profiling using neural nets.},
	urldate = {2021-10-15},
	journal = {AMIA Summits on Translational Science Proceedings},
	author = {Breitenstein, Matthew K. and Hu, Vincent JY and Bhatnagar, Roopal and Ratnagiri, Madhavi},
	month = may,
	year = {2019},
	pmid = {31258997},
	pmcid = {PMC6568105},
	pages = {435--442},
}

Systemic lupus erythematosus (SLE) is a rare, autoimmune disorder known to affect most organ sites. Complicating clinical management is a poorly differentiated, heterogenous SLE disease state. While some small molecule drugs and biologics are available for treatment, additional therapeutic options are needed. Parsing complex biological signatures using powerful, yet human interpretable approaches is critical to advancing our understanding of SLE etiology and identifying therapeutic repositioning opportunities. To approach this goal, we developed a semi-supervised deep neural network pipeline for gene expression profiling of SLE patients and subsequent characterization of individual gene features. Our pipeline performed exemplar multinomial classification of SLE patients in independent balanced validation (F1=0.956) and unbalanced, under-powered testing (F1=0.944) cohorts. A stacked autoencoder disambiguated individual feature representativeness by regenerating an input-like(A ‘) feature matrix. A to A’ comparisons suggest the top associated features to be key features in gene expression profiling using neural nets.

@article{yu_online_2019,
	title = {Online {Fault} {Diagnosis} for {Industrial} {Processes} {With} {Bayesian} {Network}-{Based} {Probabilistic} {Ensemble} {Learning} {Strategy}},
	volume = {16},
	issn = {1558-3783},
	doi = {10.1109/TASE.2019.2915286},
	abstract = {The efficient mitigation of the detrimental effects of a fault in complex systems requires online fault diagnosis techniques that are able to identify the cause of an observable anomaly. However, an individual diagnosis model can only acquire a limited diagnostic effect and may be insufficient for a particular application. In this paper, a Bayesian network-based probabilistic ensemble learning (PEL-BN) strategy is proposed to address the aforementioned issue. First, an ensemble index is proposed to evaluate the candidate diagnosis models in a probabilistic manner so that the diagnosis models with better diagnosis performance can be selected. Then, based on the selected classifiers, the architecture of the Bayesian network can be constructed using the proposed three types of basic topologies. Finally, the advantages of different diagnosis models are integrated using the developed Bayesian network, and thus, the fault causes of the observable anomaly can be accurately inferred. In addition, the proposed method can effectively capture the mixed fault characteristics of multifaults (MFs) by integrating decisions derived from different diagnosis models. Hence, this method can also provide a feasible solution for diagnosing MFs in real industrial processes. A simulation process and a real industrial process are adopted to verify the performance of the proposed method, and the experimental results illustrate that the proposed PEL-BN strategy improves the diagnosis performance of single faults and is a feasible solution for MF diagnosis. Note to Practitioners-The focus of this paper is to develop a probabilistic ensemble learning strategy based on the Bayesian network (PEL-BN) to diagnose different kinds of faults in industrial processes. The PEL-BN strategy can automatically select the base classifiers to establish the architecture of the Bayesian network. In this way, the conclusions of these base classifiers can be effectively integrated to provide better diagnosis performance. In addition, the proposed method is also a feasible technique for diagnosing MFs resulted from the joint effects of multiple faults.},
	number = {4},
	journal = {IEEE Transactions on Automation Science and Engineering},
	author = {Yu, Wanke and Zhao, Chunhui},
	month = oct,
	year = {2019},
	note = {Conference Name: IEEE Transactions on Automation Science and Engineering},
	keywords = {Base classifiers selection, Bayes methods, Bayesian network, Fault diagnosis, Network topology, Probabilistic logic, diagnosis, diagnostics, ensemble, fault diagnosis probabilistic ensemble learning},
	pages = {1922--1932},
}

The efficient mitigation of the detrimental effects of a fault in complex systems requires online fault diagnosis techniques that are able to identify the cause of an observable anomaly. However, an individual diagnosis model can only acquire a limited diagnostic effect and may be insufficient for a particular application. In this paper, a Bayesian network-based probabilistic ensemble learning (PEL-BN) strategy is proposed to address the aforementioned issue. First, an ensemble index is proposed to evaluate the candidate diagnosis models in a probabilistic manner so that the diagnosis models with better diagnosis performance can be selected. Then, based on the selected classifiers, the architecture of the Bayesian network can be constructed using the proposed three types of basic topologies. Finally, the advantages of different diagnosis models are integrated using the developed Bayesian network, and thus, the fault causes of the observable anomaly can be accurately inferred. In addition, the proposed method can effectively capture the mixed fault characteristics of multifaults (MFs) by integrating decisions derived from different diagnosis models. Hence, this method can also provide a feasible solution for diagnosing MFs in real industrial processes. A simulation process and a real industrial process are adopted to verify the performance of the proposed method, and the experimental results illustrate that the proposed PEL-BN strategy improves the diagnosis performance of single faults and is a feasible solution for MF diagnosis. Note to Practitioners-The focus of this paper is to develop a probabilistic ensemble learning strategy based on the Bayesian network (PEL-BN) to diagnose different kinds of faults in industrial processes. The PEL-BN strategy can automatically select the base classifiers to establish the architecture of the Bayesian network. In this way, the conclusions of these base classifiers can be effectively integrated to provide better diagnosis performance. In addition, the proposed method is also a feasible technique for diagnosing MFs resulted from the joint effects of multiple faults.

@inproceedings{chen_railway_2019,
	title = {Railway {Vehicle} {Door} {Fault} {Diagnosis} {Method} with {Bayesian} {Network}},
	doi = {10.1109/ICCRE.2019.8724211},
	abstract = {Recent years, more attention has been given to fault diagnosis of railway vehicle door system. In order to handle the uncertainties in fault diagnosis of the door system, a fault diagnosis method based on Bayesian Network was proposed. Fault data provided by a subway company was counted. A model based on Bayesian Network of railway vehicle door was built up and the prior probability of failure was calculated. Inputting fault evidence in the Bayesian model, the posterior probability of each fault would be obtained. Simulation experiment and engineering application show that Bayesian Network can reason through the fault of door system correctly and the result can provide reference and advice for fault diagnosis and maintenance of railway vehicle door.},
	booktitle = {2019 4th {International} {Conference} on {Control} and {Robotics} {Engineering} ({ICCRE})},
	author = {Chen, Ruwen and Zhu, Songqing and Hao, Fei and Zhu, Bin and Zhao, Zhendong and Xu, Youxiong},
	month = apr,
	year = {2019},
	keywords = {Bayes methods, Bayesian Network, Fault diagnosis, Fault trees, Logic gates, Maintenance engineering, Rails, Switches, door, fault diagnosis, posterior probability, prior probability, railway, railway vehicle door},
	pages = {70--74},
}

Recent years, more attention has been given to fault diagnosis of railway vehicle door system. In order to handle the uncertainties in fault diagnosis of the door system, a fault diagnosis method based on Bayesian Network was proposed. Fault data provided by a subway company was counted. A model based on Bayesian Network of railway vehicle door was built up and the prior probability of failure was calculated. Inputting fault evidence in the Bayesian model, the posterior probability of each fault would be obtained. Simulation experiment and engineering application show that Bayesian Network can reason through the fault of door system correctly and the result can provide reference and advice for fault diagnosis and maintenance of railway vehicle door.

@article{atoui_single_2019,
	title = {A single {Bayesian} network classifier for monitoring with unknown classes},
	volume = {85},
	issn = {0952-1976},
	url = {https://www.sciencedirect.com/science/article/pii/S0952197619301800},
	doi = {10.1016/j.engappai.2019.07.016},
	abstract = {In this paper, the Conditional Gaussian Networks (CGNs), a form of Bayesian Networks (BN), are used as a statistical process monitoring approach to detect and diagnose faults. The proposed approach improves the structure of Bayesian networks and generalizes a few results regarding statistical tests and the use of an exclusion criterion. The proposed framework is evaluated using data from the benchmark Tennessee Eastman Process (TEP) with various scenarios.},
	language = {en},
	urldate = {2021-10-14},
	journal = {Engineering Applications of Artificial Intelligence},
	author = {Atoui, M. Amine and Cohen, Achraf and Verron, Sylvain and Kobi, Abdessamad},
	month = oct,
	year = {2019},
	keywords = {Bayesian networks, Classification, Exclusion criteria, Fault detection and diagnosis},
	pages = {681--690},
}

In this paper, the Conditional Gaussian Networks (CGNs), a form of Bayesian Networks (BN), are used as a statistical process monitoring approach to detect and diagnose faults. The proposed approach improves the structure of Bayesian networks and generalizes a few results regarding statistical tests and the use of an exclusion criterion. The proposed framework is evaluated using data from the benchmark Tennessee Eastman Process (TEP) with various scenarios.

@article{dineva_fault_2019,
	title = {Fault {Diagnosis} of {Rotating} {Electrical} {Machines} {Using} {Multi}-{Label} {Classification}},
	volume = {9},
	copyright = {http://creativecommons.org/licenses/by/3.0/},
	url = {https://www.mdpi.com/2076-3417/9/23/5086},
	doi = {10.3390/app9235086},
	abstract = {Fault Detection and Diagnosis of electrical machine and drive systems are of utmost importance in modern industrial automation. The widespread use of Machine Learning techniques has made it possible to replace traditional motor fault detection techniques with more efficient solutions that are capable of early fault recognition by using large amounts of sensory data. However, the detection of concurrent failures is still a challenge in the presence of disturbing noises or when the multiple faults cause overlapping features. Multi-label classification has recently gained popularity in various application domains as an efficient method for fault detection and monitoring of systems with promising results. The contribution of this work is to propose a novel methodology for multi-label classification for simultaneously diagnosing multiple faults and evaluating the fault severity under noisy conditions. In this research, the Electrical Signature Analysis as well as traditional vibration data have been considered for modeling. Furthermore, the performance of various multi-label classification models is compared. Current and vibration signals are acquired under normal and fault conditions. The applicability of the proposed method is experimentally validated under diverse fault conditions such as unbalance and misalignment.},
	language = {en},
	number = {23},
	urldate = {2021-10-11},
	journal = {Applied Sciences},
	author = {Dineva, Adrienn and Mosavi, Amir and Gyimesi, Mate and Vajda, Istvan and Nabipour, Narjes and Rabczuk, Timon},
	month = jan,
	year = {2019},
	note = {Number: 23
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {big data, data science, drive systems and power electronics, electric machine, energy conversion, fault classifiers, fault severity, machine learning, multi-label classification, multiple fault detection, rotating electrical machines, soft computing},
	pages = {5086},
}

Fault Detection and Diagnosis of electrical machine and drive systems are of utmost importance in modern industrial automation. The widespread use of Machine Learning techniques has made it possible to replace traditional motor fault detection techniques with more efficient solutions that are capable of early fault recognition by using large amounts of sensory data. However, the detection of concurrent failures is still a challenge in the presence of disturbing noises or when the multiple faults cause overlapping features. Multi-label classification has recently gained popularity in various application domains as an efficient method for fault detection and monitoring of systems with promising results. The contribution of this work is to propose a novel methodology for multi-label classification for simultaneously diagnosing multiple faults and evaluating the fault severity under noisy conditions. In this research, the Electrical Signature Analysis as well as traditional vibration data have been considered for modeling. Furthermore, the performance of various multi-label classification models is compared. Current and vibration signals are acquired under normal and fault conditions. The applicability of the proposed method is experimentally validated under diverse fault conditions such as unbalance and misalignment.

@inproceedings{sarma_dbscan_2019,
	title = {μ{DBSCAN}: {An} {Exact} {Scalable} {DBSCAN} {Algorithm} for {Big} {Data} {Exploiting} {Spatial} {Locality}},
	shorttitle = {μ{DBSCAN}},
	doi = {10.1109/CLUSTER.2019.8891020},
	abstract = {DBSCAN is one of the most popular and effective clustering algorithms that is capable of identifying arbitrary-shaped clusters and noise efficiently. However, its super-linear complexity makes it infeasible for applications involving clustering of Big Data. A major portion of the computation time of DBSCAN is taken up by the neighborhood queries, which becomes a bottleneck to its performance. We address this issue in our proposed micro-cluster based DBSCAN algorithm, μDBSCAN, which identifies core-points even without performing neighbourhood queries and becomes instrumental in reducing the run-time of the algorithm. It also significantly reduces the computation time per neighbourhood query while producing exact DBSCAN clusters. Moreover, the micro-cluster based solution makes it scalable for high dimensional data. We also propose a highly scalable distributed implementation of μDBSCAN, μDBSCAN-D, to exploit a commodity cluster infrastructure. Experimental results demonstrate tremendous improvements in performance of our proposed algorithms as compared to their respective state-of-the-art solutions for various standard datasets. μDBSCAN-D is an exact parallel solution for DBSCAN which is capable of processing massive amounts of data efficiently (1 billion data points in 41 minutes on a 32 node cluster), while producing a clustering that is same as that of traditional DBSCAN.},
	booktitle = {2019 {IEEE} {International} {Conference} on {Cluster} {Computing} ({CLUSTER})},
	author = {Sarma, Aditya and Goyal, Poonam and Kumari, Sonal and Wani, Anand and Challa, Jagat Sesh and Islam, Saiyedul and Goyal, Navneet},
	month = sep,
	year = {2019},
	note = {ISSN: 2168-9253},
	keywords = {Approximation algorithms, Big Data, Clustering algorithms, Density-based Clustering, Distributed Computing, Distributed databases, Exact Clustering Algorithm, Optimized Neighborhood Query, Parallel algorithms, Partitioning algorithms, Spatial Locality, Time complexity},
	pages = {1--11},
}

DBSCAN is one of the most popular and effective clustering algorithms that is capable of identifying arbitrary-shaped clusters and noise efficiently. However, its super-linear complexity makes it infeasible for applications involving clustering of Big Data. A major portion of the computation time of DBSCAN is taken up by the neighborhood queries, which becomes a bottleneck to its performance. We address this issue in our proposed micro-cluster based DBSCAN algorithm, μDBSCAN, which identifies core-points even without performing neighbourhood queries and becomes instrumental in reducing the run-time of the algorithm. It also significantly reduces the computation time per neighbourhood query while producing exact DBSCAN clusters. Moreover, the micro-cluster based solution makes it scalable for high dimensional data. We also propose a highly scalable distributed implementation of μDBSCAN, μDBSCAN-D, to exploit a commodity cluster infrastructure. Experimental results demonstrate tremendous improvements in performance of our proposed algorithms as compared to their respective state-of-the-art solutions for various standard datasets. μDBSCAN-D is an exact parallel solution for DBSCAN which is capable of processing massive amounts of data efficiently (1 billion data points in 41 minutes on a 32 node cluster), while producing a clustering that is same as that of traditional DBSCAN.

@inproceedings{aliperti_fuzzy_2019,
	title = {A {Fuzzy} {Density}-based {Clustering} {Algorithm} for {Streaming} {Data}},
	doi = {10.1109/FUZZ-IEEE.2019.8858909},
	abstract = {The exploitation of data streams, nowadays provided nonstop by a myriad of diverse applications, asks for specific analysis methods. In this paper, we propose SF-DBSCAN, a fuzzy version of the DBSCAN algorithm, aimed to perform unsupervised analysis of streaming data. Fuzziness is introduced by fuzzy borders of density-based clusters. We describe and discuss the proposed algorithm, which evolves the clusters at each occurrence of a new object. Three synthetic datasets are used to show the ability of SF-DBSCAN to successfully track changes of data distribution, thus properly addressing concept drift. SF-DBSCAN is compared with a basic, crisp streaming version of DBSCAN with regard to modelling effectiveness.},
	booktitle = {2019 {IEEE} {International} {Conference} on {Fuzzy} {Systems} ({FUZZ}-{IEEE})},
	author = {Aliperti, Andrea and Bechini, Alessio and Marcelloni, Francesco and Renda, Alessandro},
	month = jun,
	year = {2019},
	note = {ISSN: 1558-4739},
	keywords = {Clustering algorithms, Data structures, Memory management, Partitioning algorithms, Proposals, Sensitivity, Shape},
	pages = {1--6},
}

The exploitation of data streams, nowadays provided nonstop by a myriad of diverse applications, asks for specific analysis methods. In this paper, we propose SF-DBSCAN, a fuzzy version of the DBSCAN algorithm, aimed to perform unsupervised analysis of streaming data. Fuzziness is introduced by fuzzy borders of density-based clusters. We describe and discuss the proposed algorithm, which evolves the clusters at each occurrence of a new object. Three synthetic datasets are used to show the ability of SF-DBSCAN to successfully track changes of data distribution, thus properly addressing concept drift. SF-DBSCAN is compared with a basic, crisp streaming version of DBSCAN with regard to modelling effectiveness.

@article{zaremoodi_concept-evolution_2019,
	title = {Concept-evolution detection in non-stationary data streams: a fuzzy clustering approach},
	volume = {60},
	issn = {0219-3116},
	shorttitle = {Concept-evolution detection in non-stationary data streams},
	url = {https://doi.org/10.1007/s10115-018-1266-y},
	doi = {10.1007/s10115-018-1266-y},
	abstract = {We have entered the era of networked communications where concepts such as big data and social networks are emerging. The explosion and profusion of available data in a broad range of application domains cause data streams to become an inevitable part of the most real-world applications. In the classification of data streams, there are four major challenges: infinite length, concept drift, recurring and evolving concepts. This paper proposes a novel method to address the mentioned challenges with a focus on the last one. Unlike the existing methods for detection of evolving concepts, we cast joint classification and detection of evolving concepts into optimizing an objective function by extending a fuzzy agglomerative clustering method. Moreover, rather than keeping instances or hyper-sphere summaries of previously seen classes, we just maintain boundaries in the kernel space and generate instances of each class on demand. This approach enhances the accuracy and reduces the memory usage of the proposed method. We empirically evaluated and showed the effectiveness of the proposed approach on several synthetic and real datasets. Experimental results on synthetic and real datasets show the superiority of the proposed method over the related state-of-the-art methods in this area.},
	language = {en},
	number = {3},
	urldate = {2021-10-01},
	journal = {Knowledge and Information Systems},
	author = {ZareMoodi, Poorya and Kamali Siahroudi, Sajjad and Beigy, Hamid},
	month = sep,
	year = {2019},
	pages = {1329--1352},
}

We have entered the era of networked communications where concepts such as big data and social networks are emerging. The explosion and profusion of available data in a broad range of application domains cause data streams to become an inevitable part of the most real-world applications. In the classification of data streams, there are four major challenges: infinite length, concept drift, recurring and evolving concepts. This paper proposes a novel method to address the mentioned challenges with a focus on the last one. Unlike the existing methods for detection of evolving concepts, we cast joint classification and detection of evolving concepts into optimizing an objective function by extending a fuzzy agglomerative clustering method. Moreover, rather than keeping instances or hyper-sphere summaries of previously seen classes, we just maintain boundaries in the kernel space and generate instances of each class on demand. This approach enhances the accuracy and reduces the memory usage of the proposed method. We empirically evaluated and showed the effectiveness of the proposed approach on several synthetic and real datasets. Experimental results on synthetic and real datasets show the superiority of the proposed method over the related state-of-the-art methods in this area.

@article{li_deep_2019,
	title = {Deep learning-based remaining useful life estimation of bearings using multi-scale feature extraction},
	volume = {182},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832018308299},
	doi = {10.1016/j.ress.2018.11.011},
	abstract = {Accurate evaluation of machine degradation during long-time operation is of great importance. With the rapid development of modern industries, physical model is becoming less capable of describing sophisticated systems, and data-driven approaches have been widely developed. This paper proposes a novel intelligent remaining useful life (RUL) prediction method based on deep learning. The time-frequency domain information is explored for prognostics, and multi-scale feature extraction is implemented using convolutional neural networks. Experiments on a popular rolling bearing dataset prepared from the PRONOSTIA platform are carried out to show the effectiveness of the proposed method, and its superiority is demonstrated by the comparisons with other approaches. In general, high accuracy on the RUL prediction is achieved, and the proposed method is promising for industrial applications.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Li, Xiang and Zhang, Wei and Ding, Qian},
	month = feb,
	year = {2019},
	keywords = {Deep learning, Multi-scale feature extraction, Prognostics and health management, Remaining useful life, Rolling bearing},
	pages = {208--218},
}

Accurate evaluation of machine degradation during long-time operation is of great importance. With the rapid development of modern industries, physical model is becoming less capable of describing sophisticated systems, and data-driven approaches have been widely developed. This paper proposes a novel intelligent remaining useful life (RUL) prediction method based on deep learning. The time-frequency domain information is explored for prognostics, and multi-scale feature extraction is implemented using convolutional neural networks. Experiments on a popular rolling bearing dataset prepared from the PRONOSTIA platform are carried out to show the effectiveness of the proposed method, and its superiority is demonstrated by the comparisons with other approaches. In general, high accuracy on the RUL prediction is achieved, and the proposed method is promising for industrial applications.

@article{al-dulaimi_multimodal_2019,
	title = {A multimodal and hybrid deep neural network model for {Remaining} {Useful} {Life} estimation},
	volume = {108},
	issn = {0166-3615},
	url = {https://www.sciencedirect.com/science/article/pii/S0166361518304925},
	doi = {10.1016/j.compind.2019.02.004},
	abstract = {Aging critical infrastructures and valuable machineries together with recent catastrophic incidents such as the collapse of Morandi bridge calls for an urgent quest to design advanced and innovative data-driven solutions and efficiently incorporate multi-sensor streaming data sources for condition-based maintenance. Remaining Useful Life (RUL) is a crucial measure used in this regard within manufacturing and industrial systems, and its accurate estimation enables improved decision-making for operations and maintenance. Capitalizing on the recent success of multiple-model (also referred to as hybrid or mixture of experts) deep learning techniques, the paper proposes a hybrid deep neural network framework for RUL estimation, referred to as the Hybrid Deep Neural Network Model (HDNN). The proposed HDNN framework is the first hybrid deep neural network model designed for RUL estimation that integrates two deep learning models simultaneously and in a parallel fashion. More specifically, in contrary to the majority of existing data-driven prognostic approaches for RUL estimation, which are developed based on a single deep model and can hardly maintain good generalization performance across various prognostic scenarios, the proposed HDNN framework consists of two parallel paths (one LSTM and one CNN) followed by a fully connected multilayer fusion neural network which acts as the fusion centre combining the output of the two paths to form the target RUL. The HDNN uses the LSTM path to extract temporal features while simultaneously the CNN is utilized to extract spatial features. The proposed HDNN framework is tested on the NASA commercial modular aero-propulsion system simulation (C-MAPSS) dataset. Our comprehensive experiments and comparisons with several recently proposed RUL estimation methodologies developed based on the same data-sets show that the proposed HDNN framework significantly outperforms all its counterparts in the complicated prognostic scenarios with increased number of operating conditions and fault modes.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Computers in Industry},
	author = {Al-Dulaimi, Ali and Zabihi, Soheil and Asif, Amir and Mohammadi, Arash},
	month = jun,
	year = {2019},
	keywords = {Convolutional Neural Networks (CNN), Deep learning, Hybrid models, Long Short-Term Memory Neural Network (LSTM), Machine Health Monitoring, Prognostic Health Management, Remaining Useful Life (RUL)},
	pages = {186--196},
}

Aging critical infrastructures and valuable machineries together with recent catastrophic incidents such as the collapse of Morandi bridge calls for an urgent quest to design advanced and innovative data-driven solutions and efficiently incorporate multi-sensor streaming data sources for condition-based maintenance. Remaining Useful Life (RUL) is a crucial measure used in this regard within manufacturing and industrial systems, and its accurate estimation enables improved decision-making for operations and maintenance. Capitalizing on the recent success of multiple-model (also referred to as hybrid or mixture of experts) deep learning techniques, the paper proposes a hybrid deep neural network framework for RUL estimation, referred to as the Hybrid Deep Neural Network Model (HDNN). The proposed HDNN framework is the first hybrid deep neural network model designed for RUL estimation that integrates two deep learning models simultaneously and in a parallel fashion. More specifically, in contrary to the majority of existing data-driven prognostic approaches for RUL estimation, which are developed based on a single deep model and can hardly maintain good generalization performance across various prognostic scenarios, the proposed HDNN framework consists of two parallel paths (one LSTM and one CNN) followed by a fully connected multilayer fusion neural network which acts as the fusion centre combining the output of the two paths to form the target RUL. The HDNN uses the LSTM path to extract temporal features while simultaneously the CNN is utilized to extract spatial features. The proposed HDNN framework is tested on the NASA commercial modular aero-propulsion system simulation (C-MAPSS) dataset. Our comprehensive experiments and comparisons with several recently proposed RUL estimation methodologies developed based on the same data-sets show that the proposed HDNN framework significantly outperforms all its counterparts in the complicated prognostic scenarios with increased number of operating conditions and fault modes.

@article{yu_remaining_2019,
	title = {Remaining useful life estimation using a bidirectional recurrent neural network based autoencoder scheme},
	volume = {129},
	issn = {0888-3270},
	url = {https://www.sciencedirect.com/science/article/pii/S0888327019303061},
	doi = {10.1016/j.ymssp.2019.05.005},
	abstract = {System remaining useful life (RUL) estimation is one of the major prognostic activities in industrial applications. In this paper, we propose a sensor-based data-driven scheme using a deep learning tool and the similarity-based curve matching technique to estimate the RUL of a system. The whole procedure consists of two steps: in the first step, a bidirectional recurrent neural network based autoencoder is trained in an unsupervised way to convert the multi-sensor (high-dimensional) readings collected from historical run-to-failure instances (i.e. multiple units of the same system) to low-dimensional embeddings, which are used to construct the one-dimensional health index (HI) values to reflect various health degradation patterns of the instances. In the second step, the test HI curve obtained from sensor readings collected from an on-line instance is compared with the degradation patterns built in the offline phase using the similarity-based curve matching technique, from which the RUL of the test unit can be estimated at an early stage. The proposed scheme was tested on two publicly available run-to-failure datasets: the turbofan engine datasets (simulation datasets) and the milling datasets (experimental datasets). The prognostic performance of the proposed procedure was directly compared with the existing state-of-art prognostic models in terms of various prognostic metrics on the two datasets respectively. The comparison results demonstrate the competitiveness of the proposed method used for RUL estimation of systems.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Mechanical Systems and Signal Processing},
	author = {Yu, Wennian and Kim, II Yong and Mechefske, Chris},
	month = aug,
	year = {2019},
	keywords = {Autoencoder, Bidirectional recurrent neural network, Health index, Remaining useful life},
	pages = {764--780},
}

System remaining useful life (RUL) estimation is one of the major prognostic activities in industrial applications. In this paper, we propose a sensor-based data-driven scheme using a deep learning tool and the similarity-based curve matching technique to estimate the RUL of a system. The whole procedure consists of two steps: in the first step, a bidirectional recurrent neural network based autoencoder is trained in an unsupervised way to convert the multi-sensor (high-dimensional) readings collected from historical run-to-failure instances (i.e. multiple units of the same system) to low-dimensional embeddings, which are used to construct the one-dimensional health index (HI) values to reflect various health degradation patterns of the instances. In the second step, the test HI curve obtained from sensor readings collected from an on-line instance is compared with the degradation patterns built in the offline phase using the similarity-based curve matching technique, from which the RUL of the test unit can be estimated at an early stage. The proposed scheme was tested on two publicly available run-to-failure datasets: the turbofan engine datasets (simulation datasets) and the milling datasets (experimental datasets). The prognostic performance of the proposed procedure was directly compared with the existing state-of-art prognostic models in terms of various prognostic metrics on the two datasets respectively. The comparison results demonstrate the competitiveness of the proposed method used for RUL estimation of systems.

@article{nguyen_new_2019,
	title = {A new dynamic predictive maintenance framework using deep learning for failure prognostics},
	volume = {188},
	issn = {0951-8320},
	url = {https://www.sciencedirect.com/science/article/pii/S0951832018311050},
	doi = {10.1016/j.ress.2019.03.018},
	abstract = {In Prognostic Health and Management (PHM) literature, the predictive maintenance studies can be classified into two groups. The first group focuses on the prognostics step but does not consider the maintenance decisions. The second group addresses the maintenance optimization question based on the assumptions that the prognostics information or the degradation models of the system are already known. However, none of the two groups provides a complete framework (from data-driven prognostics to maintenance decisions) investigating the impact of the imperfect prognostics on maintenance decision. Therefore, this paper aims to fill this gap of literature. It presents a novel dynamic predicive maintenance framework based on sensor measurements. In this framework, the prognostics step, based on the Long Short-Term Memory network, is oriented towards the requirements of operation planners. It provides the probabilities that the system can fail in different time horizons to decide the moment for preparing and performing maintenance activities. The proposed framework is validated on a real application case study. Its performance is highlighted when compared with two benchmark maintenance policies: classical periodic and ideal predicted maintenance. In addition, the impact of the imperfect prognostics information on maintenance decisions is discussed in this paper.},
	language = {en},
	urldate = {2021-09-28},
	journal = {Reliability Engineering \& System Safety},
	author = {Nguyen, Khanh T. P. and Medjaher, Kamal},
	month = aug,
	year = {2019},
	keywords = {Deep learning, Inventory management, PHM, Predictive maintenance, Prognostics information, Residual life prediction},
	pages = {251--262},
}

In Prognostic Health and Management (PHM) literature, the predictive maintenance studies can be classified into two groups. The first group focuses on the prognostics step but does not consider the maintenance decisions. The second group addresses the maintenance optimization question based on the assumptions that the prognostics information or the degradation models of the system are already known. However, none of the two groups provides a complete framework (from data-driven prognostics to maintenance decisions) investigating the impact of the imperfect prognostics on maintenance decision. Therefore, this paper aims to fill this gap of literature. It presents a novel dynamic predicive maintenance framework based on sensor measurements. In this framework, the prognostics step, based on the Long Short-Term Memory network, is oriented towards the requirements of operation planners. It provides the probabilities that the system can fail in different time horizons to decide the moment for preparing and performing maintenance activities. The proposed framework is validated on a real application case study. Its performance is highlighted when compared with two benchmark maintenance policies: classical periodic and ideal predicted maintenance. In addition, the impact of the imperfect prognostics information on maintenance decisions is discussed in this paper.

@article{cao_smart_2019,
	title = {Smart {Condition} {Monitoring} for {Industry} 4.0 {Manufacturing} {Processes}: {An} {Ontology}-{Based} {Approach}},
	volume = {50},
	shorttitle = {Smart {Condition} {Monitoring} for {Industry} 4.0 {Manufacturing} {Processes}},
	doi = {10.1080/01969722.2019.1565118},
	abstract = {Following the trend of Industry 4.0, automation in different manufacturing processes has triggered the use of intelligent condition monitoring systems, which are crucial for improving productivity and availability of production systems. To develop such an intelligent system, semantic technologies are of paramount importance. This paper introduces an ontology that will be used to develop an intelligent condition monitoring system. The proposed ontology formalizes domain knowledge related to condition monitoring tasks of manufacturing processes. After introducing the ontology in detail, we evaluate the proposed ontology by instantiating it with a case study: a conditional maintenance task of bearings in rotating machinery.},
	journal = {Cybernetics and Systems},
	author = {Cao, Qiushi and Giustozzi, Franco and Zanni-Merk, Cecilia and De Bertrand de Beuvron, François and Reich, Christoph},
	month = feb,
	year = {2019},
	pages = {1--15},
}

Following the trend of Industry 4.0, automation in different manufacturing processes has triggered the use of intelligent condition monitoring systems, which are crucial for improving productivity and availability of production systems. To develop such an intelligent system, semantic technologies are of paramount importance. This paper introduces an ontology that will be used to develop an intelligent condition monitoring system. The proposed ontology formalizes domain knowledge related to condition monitoring tasks of manufacturing processes. After introducing the ontology in detail, we evaluate the proposed ontology by instantiating it with a case study: a conditional maintenance task of bearings in rotating machinery.

@incollection{nalmpantis_signal2vec_2019,
	title = {{Signal2Vec}: {Time} {Series} {Embedding} {Representation}},
	isbn = {978-3-642-54671-6},
	shorttitle = {{Signal2Vec}},
	abstract = {The rise of Internet-of-Things (IoT) and the exponential increase of devices using sensors, has lead to an increasing interest in data mining of time series. In this context, several representation methods have been proposed. Signal2vec is a novel framework, which can represent any time-series in a vector space. It is unsupervised, computationally efficient, scalable and generic. The framework is evaluated via a theoretical analysis and real world applications, with a focus on energy data. The experimental results are compared against a baseline using raw data and two other popular representations, SAX and PAA. Signal2vec is superior not only in terms of performance, but also in efficiency, due to dimensionality reduction.},
	author = {Nalmpantis, Christoforos and Vrakas, Dimitris},
	month = may,
	year = {2019},
	doi = {10.1007/978-3-030-20257-6_7},
	pages = {80--90},
}

The rise of Internet-of-Things (IoT) and the exponential increase of devices using sensors, has lead to an increasing interest in data mining of time series. In this context, several representation methods have been proposed. Signal2vec is a novel framework, which can represent any time-series in a vector space. It is unsupervised, computationally efficient, scalable and generic. The framework is evaluated via a theoretical analysis and real world applications, with a focus on energy data. The experimental results are compared against a baseline using raw data and two other popular representations, SAX and PAA. Signal2vec is superior not only in terms of performance, but also in efficiency, due to dimensionality reduction.

@inproceedings{turgis_industrialization_2019,
	title = {Industrialization of {Condition} {Based} {Maintenance} for {Complex} {Systems} in a {Complex} {Maintenance} {Environment}, {Example} of {NAT}},
	abstract = {NAT train is composed by several complex systems like passenger access systems and batteries. In mass transit rolling stock, they are considered as critical systems as they are directly related to reliability and belongs to top contributors on train maintenance costs. These systems are complex by definition, as they are composed of various sub systems, themselves composed of various technologies binding to numerous functionalities and stressful environment. To realize maintenance of such systems into a large rolling fleet (more than 250 trains, 4000 passenger access systems to cover) with operational constraints due to mass transit, the maintenance plan cannot be fully covered by conditioned based maintenance. In order to reduce downtime, lower maintenance costs and to increase reliability, a mixed maintenance solution was proposed to optimize dependencies between systematic, corrective and condition-based maintenance.},
	author = {Turgis, Fabien and Audier, Pierre and Coutadeur, Quentin and Verdun, Cyril},
	month = dec,
	year = {2019},
}

NAT train is composed by several complex systems like passenger access systems and batteries. In mass transit rolling stock, they are considered as critical systems as they are directly related to reliability and belongs to top contributors on train maintenance costs. These systems are complex by definition, as they are composed of various sub systems, themselves composed of various technologies binding to numerous functionalities and stressful environment. To realize maintenance of such systems into a large rolling fleet (more than 250 trains, 4000 passenger access systems to cover) with operational constraints due to mass transit, the maintenance plan cannot be fully covered by conditioned based maintenance. In order to reduce downtime, lower maintenance costs and to increase reliability, a mixed maintenance solution was proposed to optimize dependencies between systematic, corrective and condition-based maintenance.

@article{ismail_fawaz_deep_2019,
	title = {Deep learning for time series classification: a review},
	volume = {33},
	issn = {1573-756X},
	shorttitle = {Deep learning for time series classification},
	url = {https://doi.org/10.1007/s10618-019-00619-1},
	doi = {10.1007/s10618-019-00619-1},
	abstract = {Time Series Classification (TSC) is an important and challenging problem in data mining. With the increase of time series data availability, hundreds of TSC algorithms have been proposed. Among these methods, only a few have considered Deep Neural Networks (DNNs) to perform this task. This is surprising as deep learning has seen very successful applications in the last years. DNNs have indeed revolutionized the field of computer vision especially with the advent of novel deeper architectures such as Residual and Convolutional Neural Networks. Apart from images, sequential data such as text and audio can also be processed with DNNs to reach state-of-the-art performance for document classification and speech recognition. In this article, we study the current state-of-the-art performance of deep learning algorithms for TSC by presenting an empirical study of the most recent DNN architectures for TSC. We give an overview of the most successful deep learning applications in various time series domains under a unified taxonomy of DNNs for TSC. We also provide an open source deep learning framework to the TSC community where we implemented each of the compared approaches and evaluated them on a univariate TSC benchmark (the UCR/UEA archive) and 12 multivariate time series datasets. By training 8730 deep learning models on 97 time series datasets, we propose the most exhaustive study of DNNs for TSC to date.},
	language = {en},
	number = {4},
	urldate = {2020-12-12},
	journal = {Data Mining and Knowledge Discovery},
	author = {Ismail Fawaz, Hassan and Forestier, Germain and Weber, Jonathan and Idoumghar, Lhassane and Muller, Pierre-Alain},
	month = jul,
	year = {2019},
	pages = {917--963},
}

Time Series Classification (TSC) is an important and challenging problem in data mining. With the increase of time series data availability, hundreds of TSC algorithms have been proposed. Among these methods, only a few have considered Deep Neural Networks (DNNs) to perform this task. This is surprising as deep learning has seen very successful applications in the last years. DNNs have indeed revolutionized the field of computer vision especially with the advent of novel deeper architectures such as Residual and Convolutional Neural Networks. Apart from images, sequential data such as text and audio can also be processed with DNNs to reach state-of-the-art performance for document classification and speech recognition. In this article, we study the current state-of-the-art performance of deep learning algorithms for TSC by presenting an empirical study of the most recent DNN architectures for TSC. We give an overview of the most successful deep learning applications in various time series domains under a unified taxonomy of DNNs for TSC. We also provide an open source deep learning framework to the TSC community where we implemented each of the compared approaches and evaluated them on a univariate TSC benchmark (the UCR/UEA archive) and 12 multivariate time series datasets. By training 8730 deep learning models on 97 time series datasets, we propose the most exhaustive study of DNNs for TSC to date.

@inproceedings{korycki_active_2019,
	title = {Active {Learning} with {Abstaining} {Classifiers} for {Imbalanced} {Drifting} {Data} {Streams}},
	doi = {10.1109/BigData47090.2019.9006453},
	abstract = {Learning from data streams is one of the most promising and challenging domains in modern machine learning. Proliferating online data sources provide us access to real-time knowledge we have never had before. At the same time, new obstacles emerge and we have to overcome them in order to fully and effectively utilize the potential of the data. Prohibitive time and memory constraints or non-stationary distributions are only some of the problems. When dealing with classification tasks, one has to remember that effective adaptation has to be achieved on weak foundations of partially labeled and often imbalanced data. In our work, we propose an online framework for binary classification, that aims to handle the complex problem of working with dynamic, sparsely labeled and imbalanced streams. The main part of it is a novel active learning strategy (MD-OAL) that is able to prioritize labeling of minority instances and, as a result, improve the balance of the learning process. We combine the strategy with a dynamic ensemble of base learners that can abstain from making decisions, if they are very uncertain. We adjust the abstaining mechanism in favor of minority instances, providing an effective method for handling remaining imbalance and a concept drift simultaneously. The conducted evaluation shows that in the challenging and realistic scenarios our framework outperforms state-of-the-art algorithms, providing higher resilience to the combined effect of limited labeling and imbalance.},
	booktitle = {2019 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Korycki, Ł and Cano, A. and Krawczyk, B.},
	month = dec,
	year = {2019},
	keywords = {Computer science, Data mining, Data models, Heuristic algorithms, Labeling, Machine learning, Uncertainty, abstaining classifiers, active learning, active learning strategy, binary classification, classification tasks, data stream mining, ensemble learning., imbalanced data, imbalanced drifting data streams, learning (artificial intelligence), machine learning, online data sources, pattern classification, real-time knowledge},
	pages = {2334--2343},
}

Learning from data streams is one of the most promising and challenging domains in modern machine learning. Proliferating online data sources provide us access to real-time knowledge we have never had before. At the same time, new obstacles emerge and we have to overcome them in order to fully and effectively utilize the potential of the data. Prohibitive time and memory constraints or non-stationary distributions are only some of the problems. When dealing with classification tasks, one has to remember that effective adaptation has to be achieved on weak foundations of partially labeled and often imbalanced data. In our work, we propose an online framework for binary classification, that aims to handle the complex problem of working with dynamic, sparsely labeled and imbalanced streams. The main part of it is a novel active learning strategy (MD-OAL) that is able to prioritize labeling of minority instances and, as a result, improve the balance of the learning process. We combine the strategy with a dynamic ensemble of base learners that can abstain from making decisions, if they are very uncertain. We adjust the abstaining mechanism in favor of minority instances, providing an effective method for handling remaining imbalance and a concept drift simultaneously. The conducted evaluation shows that in the challenging and realistic scenarios our framework outperforms state-of-the-art algorithms, providing higher resilience to the combined effect of limited labeling and imbalance.

@article{hu_novel_2019,
	title = {A {Novel} {Segmentation} and {Representation} {Approach} for {Streaming} {Time} {Series}},
	volume = {7},
	issn = {2169-3536},
	doi = {10.1109/ACCESS.2018.2828320},
	abstract = {Along with the coming of Internet of Everything era, massive numbers of pervasive connected devices in various fields are continuously producing oceans of time series stream data. In order to carry out different kinds of data mining tasks (similarity search, classification, clustering, and prediction) based on streaming time series efficiently and effectively, segmentation and representation which segment a streaming time series into several subsequences and provide approximative representation for the raw data, should be done as the first step. With the virtue of solid theoretical foundations, piecewise linear representation (PLR) has been gained success in yielding more compact representation and fewer segments. However, the current state of art PLR methods have their own flaws: For one thing, most of current PLR methods focus on the guaranteed error bound instead of the holistic approximation error, which may lead to excessive fitting errors of segments and loss of factual research significance. For another, most of current PLR methods process streaming time series with some fixed criteria, which cannot provide a more flexible way to represent streaming time series. Motivated by the above analysis, we propose a novel continuous segmentation and multi-resolution representation approach based on turning points, which subdivides the streaming time series by a set of temporal feature points and represents the time series flexibly. Our method can not only generate more accurate approximation than the state-of-the-art of PLR algorithm, but also represent the streaming time series in a more flexible way to meet different needs of users. Extensive experiments on different kinds of typical time series datasets have been conducted to demonstrate the superiorities of our method.},
	journal = {IEEE Access},
	author = {Hu, Yupeng and Guan, Peiyuan and Zhan, Peng and Ding, Yiming and Li, Xueqing},
	year = {2019},
	note = {Conference Name: IEEE Access},
	keywords = {Approximation algorithms, Data mining, Indexes, Internet of Everything era, Internet of Things, Microsoft Windows, Task analysis, Time series analysis, Turning, continuous segmentation approach, data mining, data structures, multi-resolution representation, multiresolution representation approach, online segmentation, piecewise linear representation, piecewise linear techniques, streaming time series, time series, time series stream data, typical time series datasets},
	pages = {184423--184437},
}

Along with the coming of Internet of Everything era, massive numbers of pervasive connected devices in various fields are continuously producing oceans of time series stream data. In order to carry out different kinds of data mining tasks (similarity search, classification, clustering, and prediction) based on streaming time series efficiently and effectively, segmentation and representation which segment a streaming time series into several subsequences and provide approximative representation for the raw data, should be done as the first step. With the virtue of solid theoretical foundations, piecewise linear representation (PLR) has been gained success in yielding more compact representation and fewer segments. However, the current state of art PLR methods have their own flaws: For one thing, most of current PLR methods focus on the guaranteed error bound instead of the holistic approximation error, which may lead to excessive fitting errors of segments and loss of factual research significance. For another, most of current PLR methods process streaming time series with some fixed criteria, which cannot provide a more flexible way to represent streaming time series. Motivated by the above analysis, we propose a novel continuous segmentation and multi-resolution representation approach based on turning points, which subdivides the streaming time series by a set of temporal feature points and represents the time series flexibly. Our method can not only generate more accurate approximation than the state-of-the-art of PLR algorithm, but also represent the streaming time series in a more flexible way to meet different needs of users. Extensive experiments on different kinds of typical time series datasets have been conducted to demonstrate the superiorities of our method.

@article{mabrok_pattern_2019,
	title = {Pattern detection for time series trajectories in human in the loop applications},
	volume = {37},
	issn = {1064-1246},
	url = {https://content.iospress.com/articles/journal-of-intelligent-and-fuzzy-systems/ifs179070},
	doi = {10.3233/JIFS-179070},
	abstract = {Extracting repeated unknown maneuvers or patterns preformed by a human operator in a cyber-physical systems can lead to better understanding of the behavior of the human operator who is controlling or sharing tasks with dynamical systems. These repea},
	language = {en},
	number = {1},
	urldate = {2020-10-01},
	journal = {Journal of Intelligent \& Fuzzy Systems},
	author = {Mabrok, Mohamed A. and Abdel-Aty, Abdel-Haleem},
	month = jan,
	year = {2019},
	note = {Publisher: IOS Press},
	pages = {115--123},
}

Extracting repeated unknown maneuvers or patterns preformed by a human operator in a cyber-physical systems can lead to better understanding of the behavior of the human operator who is controlling or sharing tasks with dynamical systems. These repea

@article{gan_survey_2019,
	title = {A {Survey} of {Parallel} {Sequential} {Pattern} {Mining}},
	url = {http://arxiv.org/abs/1805.10515},
	doi = {10.1145/3314107},
	abstract = {With the growing popularity of shared resources, large volumes of complex data of different types are collected automatically. Traditional data mining algorithms generally have problems and challenges including huge memory cost, low processing speed, and inadequate hard disk space. As a fundamental task of data mining, sequential pattern mining (SPM) is used in a wide variety of real-life applications. However, it is more complex and challenging than other pattern mining tasks, i.e., frequent itemset mining and association rule mining, and also suffers from the above challenges when handling the large-scale data. To solve these problems, mining sequential patterns in a parallel or distributed computing environment has emerged as an important issue with many applications. In this paper, an in-depth survey of the current status of parallel sequential pattern mining (PSPM) is investigated and provided, including detailed categorization of traditional serial SPM approaches, and state of the art parallel SPM. We review the related work of parallel sequential pattern mining in detail, including partition-based algorithms for PSPM, Apriori-based PSPM, pattern growth based PSPM, and hybrid algorithms for PSPM, and provide deep description (i.e., characteristics, advantages, disadvantages and summarization) of these parallel approaches of PSPM. Some advanced topics for PSPM, including parallel quantitative / weighted / utility sequential pattern mining, PSPM from uncertain data and stream data, hardware acceleration for PSPM, are further reviewed in details. Besides, we review and provide some well-known open-source software of PSPM. Finally, we summarize some challenges and opportunities of PSPM in the big data era.},
	urldate = {2020-10-01},
	journal = {arXiv:1805.10515 [cs]},
	author = {Gan, Wensheng and Lin, Jerry Chun-Wei and Fournier-Viger, Philippe and Chao, Han-Chieh and Yu, Philip S.},
	month = apr,
	year = {2019},
	note = {arXiv: 1805.10515},
	keywords = {Computer Science - Databases},
}

With the growing popularity of shared resources, large volumes of complex data of different types are collected automatically. Traditional data mining algorithms generally have problems and challenges including huge memory cost, low processing speed, and inadequate hard disk space. As a fundamental task of data mining, sequential pattern mining (SPM) is used in a wide variety of real-life applications. However, it is more complex and challenging than other pattern mining tasks, i.e., frequent itemset mining and association rule mining, and also suffers from the above challenges when handling the large-scale data. To solve these problems, mining sequential patterns in a parallel or distributed computing environment has emerged as an important issue with many applications. In this paper, an in-depth survey of the current status of parallel sequential pattern mining (PSPM) is investigated and provided, including detailed categorization of traditional serial SPM approaches, and state of the art parallel SPM. We review the related work of parallel sequential pattern mining in detail, including partition-based algorithms for PSPM, Apriori-based PSPM, pattern growth based PSPM, and hybrid algorithms for PSPM, and provide deep description (i.e., characteristics, advantages, disadvantages and summarization) of these parallel approaches of PSPM. Some advanced topics for PSPM, including parallel quantitative / weighted / utility sequential pattern mining, PSPM from uncertain data and stream data, hardware acceleration for PSPM, are further reviewed in details. Besides, we review and provide some well-known open-source software of PSPM. Finally, we summarize some challenges and opportunities of PSPM in the big data era.

@book{li_sample_2019,
	title = {Sample {Adaptive} {Multiple} {Kernel} {Learning} for {Failure} {Prediction} of {Railway} {Points}},
	author = {Li, Zhibin and Zhang, Jian and Wu, Qiang and Gong, Yongshun and Yi, Jinfeng and Kirsch, Christina},
	year = {2019},
	note = {\_eprint: 1907.01162},
}

@article{gomes_machine_2019,
	title = {Machine learning for streaming data: state of the art, challenges, and opportunities},
	volume = {21},
	issn = {1931-0145},
	shorttitle = {Machine learning for streaming data},
	url = {https://doi.org/10.1145/3373464.3373470},
	doi = {10.1145/3373464.3373470},
	abstract = {Incremental learning, online learning, and data stream learning are terms commonly associated with learning algorithms that update their models given a continuous influx of data without performing multiple passes over data. Several works have been devoted to this area, either directly or indirectly as characteristics of big data processing, i.e., Velocity and Volume. Given the current industry needs, there are many challenges to be addressed before existing methods can be efficiently applied to real-world problems. In this work, we focus on elucidating the connections among the current stateof- the-art on related fields; and clarifying open challenges in both academia and industry. We treat with special care topics that were not thoroughly investigated in past position and survey papers. This work aims to evoke discussion and elucidate the current research opportunities, highlighting the relationship of different subareas and suggesting courses of action when possible.},
	number = {2},
	urldate = {2020-03-17},
	journal = {ACM SIGKDD Explorations Newsletter},
	author = {Gomes, Heitor Murilo and Read, Jesse and Bifet, Albert and Barddal, Jean Paul and Gama, João},
	month = nov,
	year = {2019},
	pages = {6--22},
}

Incremental learning, online learning, and data stream learning are terms commonly associated with learning algorithms that update their models given a continuous influx of data without performing multiple passes over data. Several works have been devoted to this area, either directly or indirectly as characteristics of big data processing, i.e., Velocity and Volume. Given the current industry needs, there are many challenges to be addressed before existing methods can be efficiently applied to real-world problems. In this work, we focus on elucidating the connections among the current stateof- the-art on related fields; and clarifying open challenges in both academia and industry. We treat with special care topics that were not thoroughly investigated in past position and survey papers. This work aims to evoke discussion and elucidate the current research opportunities, highlighting the relationship of different subareas and suggesting courses of action when possible.

@article{carvalho_systematic_2019,
	title = {A systematic literature review of machine learning methods applied to predictive maintenance},
	volume = {137},
	issn = {0360-8352},
	url = {http://www.sciencedirect.com/science/article/pii/S0360835219304838},
	doi = {https://doi.org/10.1016/j.cie.2019.106024},
	journal = {Computers \& Industrial Engineering},
	author = {Carvalho, Thyago P. and Soares, Fabrízzio A. A. M. N. and Vita, Roberto and Francisco, Roberto da P. and Basto, João P. and Alcalá, Symone G. S.},
	year = {2019},
	keywords = {Artificial intelligence, Machine learning, PdM, Predictive maintenance, Systematic literature review},
	pages = {106024},
}

@article{noman_overview_2019,
	title = {Overview of predictive condition based maintenance research using bibliometric indicators},
	volume = {31},
	issn = {1018-3639},
	url = {http://www.sciencedirect.com/science/article/pii/S1018363917303720},
	doi = {https://doi.org/10.1016/j.jksues.2018.02.003},
	number = {4},
	journal = {Journal of King Saud University - Engineering Sciences},
	author = {Noman, Mohammed A. and Nasr, Emad S. Abouel and Al-Shayea, Adel and Kaid, Husam},
	year = {2019},
	pages = {355 -- 367},
}

@article{fernandes_data_2019,
	title = {Data analysis and feature selection for predictive maintenance: {A} case-study in the metallurgic industry},
	volume = {46},
	issn = {0268-4012},
	url = {http://www.sciencedirect.com/science/article/pii/S0268401218304699},
	doi = {https://doi.org/10.1016/j.ijinfomgt.2018.10.006},
	abstract = {Proactive Maintenance practices are becoming more standard in industrial environments, with a direct and profound impact on the competitivity within the sector. These practices demand the continuous monitorization of industrial equipment, which generates extensive amounts of data. This information can be processed into useful knowledge with the use of machine learning algorithms. However, before the algorithms can effectively be applied, the data must go through an exploratory phase: assessing the meaning of the features and to which degree they are redundant. In this paper, we present the findings of the analysis conducted on a real-world dataset from a metallurgic company. A number of data analysis and feature selection methods are employed, uncovering several relationships, which are systematized in a rule-based model, and reducing the feature space from an initial 47-feature dataset to a 32-feature dataset.},
	journal = {International Journal of Information Management},
	author = {Fernandes, Marta and Canito, Alda and Bolón-Canedo, Verónica and Conceição, Luís and Praça, Isabel and Marreiros, Goreti},
	year = {2019},
	keywords = {Data analysis, Feature selection, Predictive maintenance, Rule-based model},
	pages = {252 -- 262},
}

Proactive Maintenance practices are becoming more standard in industrial environments, with a direct and profound impact on the competitivity within the sector. These practices demand the continuous monitorization of industrial equipment, which generates extensive amounts of data. This information can be processed into useful knowledge with the use of machine learning algorithms. However, before the algorithms can effectively be applied, the data must go through an exploratory phase: assessing the meaning of the features and to which degree they are redundant. In this paper, we present the findings of the analysis conducted on a real-world dataset from a metallurgic company. A number of data analysis and feature selection methods are employed, uncovering several relationships, which are systematized in a rule-based model, and reducing the feature space from an initial 47-feature dataset to a 32-feature dataset.

@article{allah_bukhsh_predictive_2019,
	title = {Predictive maintenance using tree-based classification techniques: {A} case of railway switches},
	volume = {101},
	issn = {0968-090X},
	shorttitle = {Predictive maintenance using tree-based classification techniques},
	url = {http://www.sciencedirect.com/science/article/pii/S0968090X18309057},
	doi = {10.1016/j.trc.2019.02.001},
	abstract = {With growing service demands, rapid deterioration due to extensive usage, and limited maintenance due to budget cuts, the railway infrastructure is in a critical state and require continuous maintenance. The infrastructure managers have to come up with smart maintenance decisions in order to improve the assets’ condition, spend optimal cost and keep the network available. Currently, the infrastructure managers lack the tools and decision support models that could assist them in taking (un) planned maintenance decisions effectively and efficiently. Recently, many literature studies have proposed to employ the machine learning techniques to estimate the performance state of an asset, predict the maintenance need, possible failure modes, and such similar aspects in advance. Most of these studies have utilised additional data collection measures to record the assets’ behaviour. Though useful for experimentation, it is expensive and impractical to mount monitoring devices on multiple assets across the network. Therefore, the objective of this study is to develop predictive models that utilise existing data from a railway agency and yield interpretable results. We propose to leverage the tree-based classification techniques of machine learning in order to predict maintenance need, activity type and trigger’s status of railway switches. Using the data from an in-use business process, predictive models based on the decision tree, random forest, and gradient boosted trees are developed. Moreover, to facilitate in models interpretability, we provided a detail explanation of models’ predictions by features importance analysis and instance level details. Our solution approach of predictive models development and their results explanation have wider applicability and can be used for other asset types and different (maintenance) planning scenarios.},
	language = {en},
	urldate = {2020-09-19},
	journal = {Transportation Research Part C: Emerging Technologies},
	author = {Allah Bukhsh, Zaharah and Saeed, Aaqib and Stipanovic, Irina and Doree, Andre G.},
	month = apr,
	year = {2019},
	keywords = {Classification, Data-driven, Decision support, LIME, Machine learning, Predictive maintenance, Railway infrastructure, Switches and crossings},
	pages = {35--54},
}

With growing service demands, rapid deterioration due to extensive usage, and limited maintenance due to budget cuts, the railway infrastructure is in a critical state and require continuous maintenance. The infrastructure managers have to come up with smart maintenance decisions in order to improve the assets’ condition, spend optimal cost and keep the network available. Currently, the infrastructure managers lack the tools and decision support models that could assist them in taking (un) planned maintenance decisions effectively and efficiently. Recently, many literature studies have proposed to employ the machine learning techniques to estimate the performance state of an asset, predict the maintenance need, possible failure modes, and such similar aspects in advance. Most of these studies have utilised additional data collection measures to record the assets’ behaviour. Though useful for experimentation, it is expensive and impractical to mount monitoring devices on multiple assets across the network. Therefore, the objective of this study is to develop predictive models that utilise existing data from a railway agency and yield interpretable results. We propose to leverage the tree-based classification techniques of machine learning in order to predict maintenance need, activity type and trigger’s status of railway switches. Using the data from an in-use business process, predictive models based on the decision tree, random forest, and gradient boosted trees are developed. Moreover, to facilitate in models interpretability, we provided a detail explanation of models’ predictions by features importance analysis and instance level details. Our solution approach of predictive models development and their results explanation have wider applicability and can be used for other asset types and different (maintenance) planning scenarios.

@inproceedings{ismail_fawaz_transfer_2018,
	title = {Transfer learning for time series classification},
	url = {https://ieeexplore.ieee.org/document/8621990},
	doi = {10.1109/BigData.2018.8621990},
	abstract = {Transfer learning for deep neural networks is the process of first training a base network on a source dataset, and then transferring the learned features (the network’s weights) to a second network to be trained on a target dataset. This idea has been shown to improve deep neural network’s generalization capabilities in many computer vision tasks such as image recognition and object localization. Apart from these applications, deep Convolutional Neural Networks (CNNs) have also recently gained popularity in the Time Series Classification (TSC) community. However, unlike for image recognition problems, transfer learning techniques have not yet been investigated thoroughly for the TSC task. This is surprising as the accuracy of deep learning models for TSC could potentially be improved if the model is fine-tuned from a pre-trained neural network instead of training it from scratch. In this paper, we fill this gap by investigating how to transfer deep CNNs for the TSC task. To evaluate the potential of transfer learning, we performed extensive experiments using the UCR archive which is the largest publicly available TSC benchmark containing 85 datasets. For each dataset in the archive, we pre-trained a model and then fine-tuned it on the other datasets resulting in 7140 different deep neural networks. These experiments revealed that transfer learning can improve or degrade the models predictions depending on the dataset used for transfer. Therefore, in an effort to predict the best source dataset for a given target dataset, we propose a new method relying on Dynamic Time Warping to measure inter-datasets similarities. We describe how our method can guide the transfer to choose the best source dataset leading to an improvement in accuracy on 71 out of 85 datasets.},
	urldate = {2023-10-17},
	booktitle = {2018 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Ismail Fawaz, Hassan and Forestier, Germain and Weber, Jonathan and Idoumghar, Lhassane and Muller, Pierre-Alain},
	month = dec,
	year = {2018},
	pages = {1367--1376},
}

Transfer learning for deep neural networks is the process of first training a base network on a source dataset, and then transferring the learned features (the network’s weights) to a second network to be trained on a target dataset. This idea has been shown to improve deep neural network’s generalization capabilities in many computer vision tasks such as image recognition and object localization. Apart from these applications, deep Convolutional Neural Networks (CNNs) have also recently gained popularity in the Time Series Classification (TSC) community. However, unlike for image recognition problems, transfer learning techniques have not yet been investigated thoroughly for the TSC task. This is surprising as the accuracy of deep learning models for TSC could potentially be improved if the model is fine-tuned from a pre-trained neural network instead of training it from scratch. In this paper, we fill this gap by investigating how to transfer deep CNNs for the TSC task. To evaluate the potential of transfer learning, we performed extensive experiments using the UCR archive which is the largest publicly available TSC benchmark containing 85 datasets. For each dataset in the archive, we pre-trained a model and then fine-tuned it on the other datasets resulting in 7140 different deep neural networks. These experiments revealed that transfer learning can improve or degrade the models predictions depending on the dataset used for transfer. Therefore, in an effort to predict the best source dataset for a given target dataset, we propose a new method relying on Dynamic Time Warping to measure inter-datasets similarities. We describe how our method can guide the transfer to choose the best source dataset leading to an improvement in accuracy on 71 out of 85 datasets.

@misc{nichol_first-order_2018,
	title = {On {First}-{Order} {Meta}-{Learning} {Algorithms}},
	url = {http://arxiv.org/abs/1803.02999},
	doi = {10.48550/arXiv.1803.02999},
	abstract = {This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously unseen task sampled from this distribution. We analyze a family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates. This family includes and generalizes first-order MAML, an approximation to MAML obtained by ignoring second-order derivatives. It also includes Reptile, a new algorithm that we introduce here, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task. We expand on the results from Finn et al. showing that first-order meta-learning algorithms perform well on some well-established benchmarks for few-shot classification, and we provide theoretical analysis aimed at understanding why these algorithms work.},
	urldate = {2023-10-03},
	publisher = {arXiv},
	author = {Nichol, Alex and Achiam, Joshua and Schulman, John},
	month = oct,
	year = {2018},
	note = {arXiv:1803.02999 [cs]},
	keywords = {Computer Science - Machine Learning},
}

This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously unseen task sampled from this distribution. We analyze a family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates. This family includes and generalizes first-order MAML, an approximation to MAML obtained by ignoring second-order derivatives. It also includes Reptile, a new algorithm that we introduce here, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task. We expand on the results from Finn et al. showing that first-order meta-learning algorithms perform well on some well-established benchmarks for few-shot classification, and we provide theoretical analysis aimed at understanding why these algorithms work.

@article{losing_incremental_2018,
	title = {Incremental on-line learning: {A} review and comparison of state of the art algorithms},
	volume = {275},
	issn = {0925-2312},
	shorttitle = {Incremental on-line learning},
	url = {http://www.sciencedirect.com/science/article/pii/S0925231217315928},
	doi = {10.1016/j.neucom.2017.06.084},
	abstract = {Recently, incremental and on-line learning gained more attention especially in the context of big data and learning from data streams, conflicting with the traditional assumption of complete data availability. Even though a variety of different methods are available, it often remains unclear which of them is suitable for a specific task and how they perform in comparison to each other. We analyze the key properties of eight popular incremental methods representing different algorithm classes. Thereby, we evaluate them with regards to their on-line classification error as well as to their behavior in the limit. Further, we discuss the often neglected issue of hyperparameter optimization specifically for each method and test how robustly it can be done based on a small set of examples. Our extensive evaluation on data sets with different characteristics gives an overview of the performance with respect to accuracy, convergence speed as well as model complexity, facilitating the choice of the best method for a given application.},
	language = {en},
	urldate = {2020-03-17},
	journal = {Neurocomputing},
	author = {Losing, Viktor and Hammer, Barbara and Wersing, Heiko},
	month = jan,
	year = {2018},
	keywords = {Data streams, Hyperparameter optimization, Incremental learning, Model selection, On-line learning},
	pages = {1261--1274},
}

Recently, incremental and on-line learning gained more attention especially in the context of big data and learning from data streams, conflicting with the traditional assumption of complete data availability. Even though a variety of different methods are available, it often remains unclear which of them is suitable for a specific task and how they perform in comparison to each other. We analyze the key properties of eight popular incremental methods representing different algorithm classes. Thereby, we evaluate them with regards to their on-line classification error as well as to their behavior in the limit. Further, we discuss the often neglected issue of hyperparameter optimization specifically for each method and test how robustly it can be done based on a small set of examples. Our extensive evaluation on data sets with different characteristics gives an overview of the performance with respect to accuracy, convergence speed as well as model complexity, facilitating the choice of the best method for a given application.

@inproceedings{dosilovic_explainable_2018,
	title = {Explainable artificial intelligence: {A} survey},
	shorttitle = {Explainable artificial intelligence},
	doi = {10.23919/MIPRO.2018.8400040},
	abstract = {In the last decade, with availability of large datasets and more computing power, machine learning systems have achieved (super)human performance in a wide variety of tasks. Examples of this rapid development can be seen in image recognition, speech analysis, strategic game planning and many more. The problem with many state-of-the-art models is a lack of transparency and interpretability. The lack of thereof is a major drawback in many applications, e.g. healthcare and finance, where rationale for model's decision is a requirement for trust. In the light of these issues, explainable artificial intelligence (XAI) has become an area of interest in research community. This paper summarizes recent developments in XAI in supervised learning, starts a discussion on its connection with artificial general intelligence, and gives proposals for further research directions.},
	booktitle = {2018 41st {International} {Convention} on {Information} and {Communication} {Technology}, {Electronics} and {Microelectronics} ({MIPRO})},
	author = {Došilović, Filip Karlo and Brčić, Mario and Hlupić, Nikica},
	month = may,
	year = {2018},
	keywords = {Decision trees, Machine learning, Optimization, Predictive models, Supervised learning, Support vector machines, comprehensibility, explainability, explainable artificial intelligence, interpretability},
	pages = {0210--0215},
}

In the last decade, with availability of large datasets and more computing power, machine learning systems have achieved (super)human performance in a wide variety of tasks. Examples of this rapid development can be seen in image recognition, speech analysis, strategic game planning and many more. The problem with many state-of-the-art models is a lack of transparency and interpretability. The lack of thereof is a major drawback in many applications, e.g. healthcare and finance, where rationale for model's decision is a requirement for trust. In the light of these issues, explainable artificial intelligence (XAI) has become an area of interest in research community. This paper summarizes recent developments in XAI in supervised learning, starts a discussion on its connection with artificial general intelligence, and gives proposals for further research directions.

@misc{noauthor_bs_2018,
	title = {{BS} {EN} 13306 ({Maintenance} - {Maintenance} terminology)},
	publisher = {BSI Standards Limited},
	year = {2018},
}

@inproceedings{montiel_learning_2018,
	title = {Learning {Fast} and {Slow}: {A} {Unified} {Batch}/{Stream} {Framework}},
	shorttitle = {Learning {Fast} and {Slow}},
	doi = {10.1109/BigData.2018.8622222},
	abstract = {Data ubiquity highlights the need of efficient and adaptable data-driven solutions. In this paper, we present FAST AND SLOW LEARNING (FSL), a novel unified framework that sheds light on the symbiosis between batch and stream learning. FSL works by employing Fast (stream) and Slow (batch) Learners, emulating the mechanisms used by humans to make decisions. We showcase the applicability of FSL on the task of classification by introducing the FAST AND SLOW CLASSIFIER (FSC). A Fast Learner provides predictions on the spot, continuously updating its model and adapting to changes in the data. On the other hand, the Slow Learner provides predictions considering a wider spectrum of seen data, requiring more time and data to create complex models. Once that enough data has been collected, FSC trains the Slow Learner and starts tracking the performance of both learners. A drift detection mechanism triggers the creation of new Slow models when the current Slow model becomes obsolete. FSC selects between Fast and Slow Learners according to their performance on new incoming data. Test results on real and synthetic data show that FSC effectively drives the positive interaction of stream and batch models for learning from evolving data streams.},
	booktitle = {2018 {IEEE} {International} {Conference} on {Big} {Data} ({Big} {Data})},
	author = {Montiel, Jacob and Bifet, Albert and Losing, Viktor and Read, Jesse and Abdessalem, Talel},
	month = dec,
	year = {2018},
	keywords = {Adaptation models, Batch Learning, Classification, Concept Drift, Data models, Machine Learning, Machine learning, Power capacitors, Predictive models, Stream Learning, Task analysis, Training},
	pages = {1065--1072},
}

Data ubiquity highlights the need of efficient and adaptable data-driven solutions. In this paper, we present FAST AND SLOW LEARNING (FSL), a novel unified framework that sheds light on the symbiosis between batch and stream learning. FSL works by employing Fast (stream) and Slow (batch) Learners, emulating the mechanisms used by humans to make decisions. We showcase the applicability of FSL on the task of classification by introducing the FAST AND SLOW CLASSIFIER (FSC). A Fast Learner provides predictions on the spot, continuously updating its model and adapting to changes in the data. On the other hand, the Slow Learner provides predictions considering a wider spectrum of seen data, requiring more time and data to create complex models. Once that enough data has been collected, FSC trains the Slow Learner and starts tracking the performance of both learners. A drift detection mechanism triggers the creation of new Slow models when the current Slow model becomes obsolete. FSC selects between Fast and Slow Learners according to their performance on new incoming data. Test results on real and synthetic data show that FSC effectively drives the positive interaction of stream and batch models for learning from evolving data streams.

@article{chanial_connectionlens_2018,
	title = {Connectionlens: finding connections across heterogeneous data sources},
	volume = {11},
	issn = {2150-8097},
	shorttitle = {Connectionlens},
	url = {https://doi.org/10.14778/3229863.3236252},
	doi = {10.14778/3229863.3236252},
	abstract = {Nowadays, journalism is facilitated by the existence of large amounts of publicly available digital data sources. In particular, journalists can do investigative work, which typically consists on keyword-based searches over many heterogeneous, independently produced and dynamic data sources, to obtain useful, interconnecting and traceable information. We propose to demonstrate ConnectionLens, a system based on a novel algorithm for keyword search across heterogeneous data sources. Our demonstration scenarios are based on use cases suggested by journalists from the french journal Le Monde, with whom we collaborate.},
	number = {12},
	urldate = {2023-02-03},
	journal = {Proceedings of the VLDB Endowment},
	author = {Chanial, Camille and Dziri, Rédouane and Galhardas, Helena and Leblay, Julien and Nguyen, Minh-Huong Le and Manolescu, Ioana},
	month = aug,
	year = {2018},
	pages = {2030--2033},
}

Nowadays, journalism is facilitated by the existence of large amounts of publicly available digital data sources. In particular, journalists can do investigative work, which typically consists on keyword-based searches over many heterogeneous, independently produced and dynamic data sources, to obtain useful, interconnecting and traceable information. We propose to demonstrate ConnectionLens, a system based on a novel algorithm for keyword search across heterogeneous data sources. Our demonstration scenarios are based on use cases suggested by journalists from the french journal Le Monde, with whom we collaborate.

@article{hu_survey_2018,
	title = {A survey on online feature selection with streaming features},
	volume = {12},
	issn = {2095-2236},
	url = {https://doi.org/10.1007/s11704-016-5489-3},
	doi = {10.1007/s11704-016-5489-3},
	abstract = {In the era of big data, the dimensionality of data is increasing dramatically in many domains. To deal with high dimensionality, online feature selection becomes critical in big data mining. Recently, online selection of dynamic features has received much attention. In situations where features arrive sequentially over time, we need to perform online feature selection upon feature arrivals. Meanwhile, considering grouped features, it is necessary to deal with features arriving by groups. To handle these challenges, some state-of-the-art methods for online feature selection have been proposed. In this paper, we first give a brief review of traditional feature selection approaches. Then we discuss specific problems of online feature selection with feature streams in detail. A comprehensive review of existing online feature selection methods is presented by comparing with each other. Finally, we discuss several open issues in online feature selection.},
	language = {en},
	number = {3},
	urldate = {2022-07-29},
	journal = {Frontiers of Computer Science},
	author = {Hu, Xuegang and Zhou, Peng and Li, Peipei and Wang, Jing and Wu, Xindong},
	month = jun,
	year = {2018},
	keywords = {big data, feature selection, feature stream, online feature selection},
	pages = {479--493},
}

In the era of big data, the dimensionality of data is increasing dramatically in many domains. To deal with high dimensionality, online feature selection becomes critical in big data mining. Recently, online selection of dynamic features has received much attention. In situations where features arrive sequentially over time, we need to perform online feature selection upon feature arrivals. Meanwhile, considering grouped features, it is necessary to deal with features arriving by groups. To handle these challenges, some state-of-the-art methods for online feature selection have been proposed. In this paper, we first give a brief review of traditional feature selection approaches. Then we discuss specific problems of online feature selection with feature streams in detail. A comprehensive review of existing online feature selection methods is presented by comparing with each other. Finally, we discuss several open issues in online feature selection.

@article{abdullatif_clustering_2018,
	title = {Clustering of nonstationary data streams: {A} survey of fuzzy partitional methods},
	volume = {8},
	issn = {1942-4795},
	shorttitle = {Clustering of nonstationary data streams},
	url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/widm.1258},
	doi = {10.1002/widm.1258},
	abstract = {Data streams have arisen as a relevant research topic during the past decade. They are real-time, incremental in nature, temporally ordered, massive, contain outliers, and the objects in a data stream may evolve over time (concept drift). Clustering is often one of the earliest and most important steps in the streaming data analysis workflow. A comprehensive literature is available about stream data clustering; however, less attention is devoted to the fuzzy clustering approach, even though the nonstationary nature of many data streams makes it especially appealing. This survey discusses relevant data stream clustering algorithms focusing mainly on fuzzy methods, including their treatment of outliers and concept drift and shift. This article is categorized under Technologies {\textgreater} Machine Learning Technologies {\textgreater} Computational Intelligence Fundamental Concepts of Data and Knowledge {\textgreater} Data Concepts},
	language = {en},
	number = {4},
	urldate = {2022-07-29},
	journal = {WIREs Data Mining and Knowledge Discovery},
	author = {Abdullatif, Amr and Masulli, Francesco and Rovetta, Stefano},
	year = {2018},
	note = {\_eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/widm.1258},
	keywords = {data streams, fuzzy clustering, nonstationary data, survey},
	pages = {e1258},
}

Data streams have arisen as a relevant research topic during the past decade. They are real-time, incremental in nature, temporally ordered, massive, contain outliers, and the objects in a data stream may evolve over time (concept drift). Clustering is often one of the earliest and most important steps in the streaming data analysis workflow. A comprehensive literature is available about stream data clustering; however, less attention is devoted to the fuzzy clustering approach, even though the nonstationary nature of many data streams makes it especially appealing. This survey discusses relevant data stream clustering algorithms focusing mainly on fuzzy methods, including their treatment of outliers and concept drift and shift. This article is categorized under Technologies \textgreater Machine Learning Technologies \textgreater Computational Intelligence Fundamental Concepts of Data and Knowledge \textgreater Data Concepts

@article{von_birgelen_self-organizing_2018,
	series = {51st {CIRP} {Conference} on {Manufacturing} {Systems}},
	title = {Self-{Organizing} {Maps} for {Anomaly} {Localization} and {Predictive} {Maintenance} in {Cyber}-{Physical} {Production} {Systems}},
	volume = {72},
	issn = {2212-8271},
	url = {https://www.sciencedirect.com/science/article/pii/S221282711830307X},
	doi = {10.1016/j.procir.2018.03.150},
	abstract = {Modern Cyber-Physical Production Systems provide large amounts of data such as sensor and control signals or configuration parameters. The available data enables unsupervised, data-driven solutions for model-based anomaly detection, anomaly localization and predictive maintenance: models which represent the normal behaviour of the system are learned from data. Then, live data from the system can be compared to the predictions of the model to detect faults, perform fault diagnosis and derive the overall condition of a system or its components. In this paper we use self-organizing maps for the aforementioned tasks and evaluate the presented methods on several real-world systems.},
	language = {en},
	urldate = {2022-05-16},
	journal = {Procedia CIRP},
	author = {von Birgelen, Alexander and Buratti, Davide and Mager, Jens and Niggemann, Oliver},
	month = jan,
	year = {2018},
	keywords = {CPPS, SOM, anomaly detection, anomaly localization, cyber-physical production system, data-driven, diagnosis, predictive maintenance, self-organizing map},
	pages = {480--485},
}

Modern Cyber-Physical Production Systems provide large amounts of data such as sensor and control signals or configuration parameters. The available data enables unsupervised, data-driven solutions for model-based anomaly detection, anomaly localization and predictive maintenance: models which represent the normal behaviour of the system are learned from data. Then, live data from the system can be compared to the predictions of the model to detect faults, perform fault diagnosis and derive the overall condition of a system or its components. In this paper we use self-organizing maps for the aforementioned tasks and evaluate the presented methods on several real-world systems.

@article{tidriri_new_2018,
	series = {10th {IFAC} {Symposium} on {Fault} {Detection}, {Supervision} and {Safety} for {Technical} {Processes} {SAFEPROCESS} 2018},
	title = {A {New} {Multi}-{Objective} {Decision}-{Making} {Approach} {Applied} to the {Tennessee} {Eastman} {Process}},
	volume = {51},
	issn = {2405-8963},
	url = {https://www.sciencedirect.com/science/article/pii/S2405896318324169},
	doi = {10.1016/j.ifacol.2018.09.697},
	abstract = {In this paper, a generic framework and a new methodology aiming to decisions fusion of various Fault Detection and Diagnosis (FDD) methods are proposed. The framework consists of a discrete Bayesian Network (BN) and can handle all FDD methods, regardless of their a prior knowledge or requirements. The methodology expresses the FDD objectives to achieve the desired performance and results in a theoretical learning of the BN parameters. The development leads to a multi-objective problem under constraints, resolved with a lexicographic method. The effectiveness of the proposed Multi-Objective Decision-Making (MODM) approach is validated through the Tennessee Eastman Process (TEP), as a challenging industrial benchmark problem. The application shows the significant improvement in FDD performances that can be ensured by the proposed methodology, in terms of high fault detection rate and small false alarm rate.},
	language = {en},
	number = {24},
	urldate = {2022-05-02},
	journal = {IFAC-PapersOnLine},
	author = {Tidriri, Khaoula and Tiplica, Teodor and Chatti, Nizar and Verron, Sylvain},
	month = jan,
	year = {2018},
	keywords = {Complex systems, Decision fusion, Fault detection, Fault isolation, Generic framework},
	pages = {1212--1219},
}

In this paper, a generic framework and a new methodology aiming to decisions fusion of various Fault Detection and Diagnosis (FDD) methods are proposed. The framework consists of a discrete Bayesian Network (BN) and can handle all FDD methods, regardless of their a prior knowledge or requirements. The methodology expresses the FDD objectives to achieve the desired performance and results in a theoretical learning of the BN parameters. The development leads to a multi-objective problem under constraints, resolved with a lexicographic method. The effectiveness of the proposed Multi-Objective Decision-Making (MODM) approach is validated through the Tennessee Eastman Process (TEP), as a challenging industrial benchmark problem. The application shows the significant improvement in FDD performances that can be ensured by the proposed methodology, in terms of high fault detection rate and small false alarm rate.

@inproceedings{ren_intelligent_2018,
	title = {An {Intelligent} {Fault} {Detection} {Method} {Based} on {Sparse} {Auto}-{Encoder} for {Industrial} {Process} {Systems}: {A} {Case} {Study} on {Tennessee} {Eastman} {Process} {Chemical} {System}},
	volume = {01},
	shorttitle = {An {Intelligent} {Fault} {Detection} {Method} {Based} on {Sparse} {Auto}-{Encoder} for {Industrial} {Process} {Systems}},
	doi = {10.1109/IHMSC.2018.00051},
	abstract = {This paper introduced a deep learning approach to achieve fault detection with signal analysis and processing, which is based on an sparse auto-encoder and can be employed to achieve unsupervised learning to automatically extract features of complex data-sets to detect fault. This sparse auto-encoder can be employed to extract features from the unrecognized signals to achieve intelligent identification. The hidden layer of auto-encoder can be considered as an over-complete dictionary, which can be employed to reconstruct the input signals to extract data-sets features unsupervised. Furthermore, the sparse auto-encoder can be considered as the method to build up a specific architecture to describe the process industrial system, not only to avoid the requirement of large amount of data onto the training step, but also to handle the problem with small sample training data. Finally, the application of this method of Tennessee Eastman Process Chemical system can be employed to demonstrate and illustrate the effectiveness and the reliability of this proposed method, and the results have shown its excellent performance on fault detection for process industrial systems.},
	booktitle = {2018 10th {International} {Conference} on {Intelligent} {Human}-{Machine} {Systems} and {Cybernetics} ({IHMSC})},
	author = {Ren, Hao and Chai, Yi and Qu, Jianfeng and Zhang, Ke and Tang, Qiu},
	month = aug,
	year = {2018},
	keywords = {Chemicals, Dictionaries, Fault Detection, Fault detection, Feature extraction, Process control, Soft-max Classifier, Sparse Auto-Encoder, Tennessee Eastman Process System, Training},
	pages = {190--193},
}

This paper introduced a deep learning approach to achieve fault detection with signal analysis and processing, which is based on an sparse auto-encoder and can be employed to achieve unsupervised learning to automatically extract features of complex data-sets to detect fault. This sparse auto-encoder can be employed to extract features from the unrecognized signals to achieve intelligent identification. The hidden layer of auto-encoder can be considered as an over-complete dictionary, which can be employed to reconstruct the input signals to extract data-sets features unsupervised. Furthermore, the sparse auto-encoder can be considered as the method to build up a specific architecture to describe the process industrial system, not only to avoid the requirement of large amount of data onto the training step, but also to handle the problem with small sample training data. Finally, the application of this method of Tennessee Eastman Process Chemical system can be employed to demonstrate and illustrate the effectiveness and the reliability of this proposed method, and the results have shown its excellent performance on fault detection for process industrial systems.

@article{fontes_evaluation_2018,
	title = {Evaluation of a {Hybrid} {Clustering} {Approach} for a {Benchmark} {Industrial} {System}},
	volume = {57},
	issn = {0888-5885},
	url = {https://doi.org/10.1021/acs.iecr.8b00429},
	doi = {10.1021/acs.iecr.8b00429},
	abstract = {The paper discusses a novel algorithm for classifying data represented through multivariate time series based on similarity metrics. To improve over the performance of existent classification methods based on single similarity, the method used in this study is based on a combination between the principal component analysis similarity factor and the average-based Euclidian distance within a fuzzy clustering approach. Additionally, an approach is proposed to cope with the changes of these metrics over the time window, improving the similarity analysis between the objects. The method is applied to the Tennessee Eastman process, a well-known benchmark industrial system used to compare various fault detection and diagnosis approaches. The results were compared with standards multivariate techniques showing the efficiency and flexibility of the proposed method in fault detection and classification problems, when considering different types of failures, process variables, and changes in operating conditions.},
	number = {32},
	urldate = {2022-05-02},
	journal = {Industrial \& Engineering Chemistry Research},
	author = {Fontes, Cristiano Hora and Budman, Hector},
	month = aug,
	year = {2018},
	note = {Publisher: American Chemical Society},
	pages = {11039--11049},
}

The paper discusses a novel algorithm for classifying data represented through multivariate time series based on similarity metrics. To improve over the performance of existent classification methods based on single similarity, the method used in this study is based on a combination between the principal component analysis similarity factor and the average-based Euclidian distance within a fuzzy clustering approach. Additionally, an approach is proposed to cope with the changes of these metrics over the time window, improving the similarity analysis between the objects. The method is applied to the Tennessee Eastman process, a well-known benchmark industrial system used to compare various fault detection and diagnosis approaches. The results were compared with standards multivariate techniques showing the efficiency and flexibility of the proposed method in fault detection and classification problems, when considering different types of failures, process variables, and changes in operating conditions.

@article{thomas_data_2018,
	series = {Big {Data}: {Data} {Science} for {Process} {Control} and {Operations}},
	title = {Data mining and clustering in chemical process databases for monitoring and knowledge discovery},
	volume = {67},
	issn = {0959-1524},
	url = {https://www.sciencedirect.com/science/article/pii/S095915241730032X},
	doi = {10.1016/j.jprocont.2017.02.006},
	abstract = {Modern chemical plants maintain large historical databases recording past sensor measurements which advanced process monitoring techniques analyze to help plant operators and engineers interpret the meaning of live trends in databases. However, many of the best process monitoring methods require data organized into groups before training is possible. In practice, such organization rarely exists and the time required to create classified training data is an obstacle to the use of advanced process monitoring strategies. Data mining and knowledge discovery techniques drawn from computer science literature can help engineers find fault states in historical databases and group them together with little detailed knowledge of the process. This study evaluates how several data clustering and feature extraction techniques work together to reveal useful trends in industrial chemical process data. Two studies on an industrial scale separation tower and the Tennessee Eastman process simulation demonstrate data clustering and feature extraction effectively revealing significant process trends from high dimensional, multivariate data. Process knowledge and supervised clustering metrics compare the cluster results against true labels in the data to compare performance of different combinations of dimensionality reduction and data clustering approaches.},
	language = {en},
	urldate = {2022-05-02},
	journal = {Journal of Process Control},
	author = {Thomas, Michael C. and Zhu, Wenbo and Romagnoli, Jose A.},
	month = jul,
	year = {2018},
	keywords = {Data clustering, Data mining, Dimensionality reduction, Knowledge discovery},
	pages = {160--175},
}

Modern chemical plants maintain large historical databases recording past sensor measurements which advanced process monitoring techniques analyze to help plant operators and engineers interpret the meaning of live trends in databases. However, many of the best process monitoring methods require data organized into groups before training is possible. In practice, such organization rarely exists and the time required to create classified training data is an obstacle to the use of advanced process monitoring strategies. Data mining and knowledge discovery techniques drawn from computer science literature can help engineers find fault states in historical databases and group them together with little detailed knowledge of the process. This study evaluates how several data clustering and feature extraction techniques work together to reveal useful trends in industrial chemical process data. Two studies on an industrial scale separation tower and the Tennessee Eastman process simulation demonstrate data clustering and feature extraction effectively revealing significant process trends from high dimensional, multivariate data. Process knowledge and supervised clustering metrics compare the cluster results against true labels in the data to compare performance of different combinations of dimensionality reduction and data clustering approaches.

@article{tidriri_model-based_2018,
	title = {Model-based fault detection and diagnosis of complex chemical processes: {A} case study of the {Tennessee} {Eastman} process},
	volume = {232},
	issn = {0959-6518},
	shorttitle = {Model-based fault detection and diagnosis of complex chemical processes},
	url = {https://doi.org/10.1177/0959651818764510},
	doi = {10.1177/0959651818764510},
	abstract = {Fault detection and diagnosis for industrial systems has been an important field of research during the past years. Among these systems, the Tennessee Eastman process is extensively used as a realistic benchmark to test and compare different fault detection and diagnosis strategies. In this context, data-driven approach has been widely applied for fault detection and diagnosis of the Tennessee Eastman process, by exploiting the massive amount of available measurement data. However, only few published works had attempted to deal with the dynamic behavior of the whole system including the mixing zone, circulating pumps, the reactor, the separator, the stripper, and so on, because of the difficulty of modeling physical phenomena that may occur in such complex system. In this article, an accurate model of the Tennessee Eastman process, properly tailored for fault detection and diagnosis purposes, is provided. This model shows better fault detection and diagnosis performances than all the others proposed in the literature and gives better or comparable results with the data-driven approaches. This work uses the bond graph methodology to systematically develop computational and graphical model. This methodology provides a physical understanding of the system and a description of its dynamic behavior. The bond graph model is then used for monitoring purposes by generating formal fault indicators, called residuals, and algorithms for fault detection and diagnosis. Hence, abnormal situations are detected by supervising the residuals’ evolution and faults are isolated using the nature of the violated residuals. Therefore, the dynamic model of the Tennessee Eastman process can now be used as a basis to achieve accurately different analysis through the causal and structural features of the bond graph tool.},
	language = {en},
	number = {6},
	urldate = {2022-05-02},
	journal = {Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering},
	author = {Tidriri, Khaoula and Chatti, Nizar and Verron, Sylvain and Tiplica, Teodor},
	month = jul,
	year = {2018},
	note = {Publisher: IMECHE},
	keywords = {Fault detection, Tennessee Eastman process, bond graph, fault diagnosis, model-based methods},
	pages = {742--760},
}

Fault detection and diagnosis for industrial systems has been an important field of research during the past years. Among these systems, the Tennessee Eastman process is extensively used as a realistic benchmark to test and compare different fault detection and diagnosis strategies. In this context, data-driven approach has been widely applied for fault detection and diagnosis of the Tennessee Eastman process, by exploiting the massive amount of available measurement data. However, only few published works had attempted to deal with the dynamic behavior of the whole system including the mixing zone, circulating pumps, the reactor, the separator, the stripper, and so on, because of the difficulty of modeling physical phenomena that may occur in such complex system. In this article, an accurate model of the Tennessee Eastman process, properly tailored for fault detection and diagnosis purposes, is provided. This model shows better fault detection and diagnosis performances than all the others proposed in the literature and gives better or comparable results with the data-driven approaches. This work uses the bond graph methodology to systematically develop computational and graphical model. This methodology provides a physical understanding of the system and a description of its dynamic behavior. The bond graph model is then used for monitoring purposes by generating formal fault indicators, called residuals, and algorithms for fault detection and diagnosis. Hence, abnormal situations are detected by supervising the residuals’ evolution and faults are isolated using the nature of the violated residuals. Therefore, the dynamic model of the Tennessee Eastman process can now be used as a basis to achieve accurately different analysis through the causal and structural features of the bond graph tool.

@article{zou_fault_2018,
	title = {Fault {Diagnosis} of {Tennessee}-{Eastman} {Process} {Using} {Orthogonal} {Incremental} {Extreme} {Learning} {Machine} {Based} on {Driving} {Amount}},
	volume = {48},
	issn = {2168-2275},
	doi = {10.1109/TCYB.2018.2830338},
	abstract = {Fault diagnosis is important to the industrial process. This paper proposes an orthogonal incremental extreme learning machine based on driving amount (DAOI-ELM) for recognizing the faults of the Tennessee-Eastman process (TEP). The basic idea of DAOI-ELM is to incorporate the Gram-Schmidt orthogonalization method and driving amount into an incremental extreme learning machine (I-ELM). The case study for the 2-D nonlinear function and regression problems from the UCI dataset results show that DAOI-ELM can obtain better generalization ability and a more compact structure of ELM than I-ELM, convex I-ELM (CI-ELM), orthogonal I-ELM (OI-ELM), and bidirectional ELM. The experimental training and testing data are derived from the simulations of TEP. The performance of DAOI-ELM is evaluated and compared with that of the back propagation neural network, support vector machine, I-ELM, CI-ELM, and OI-ELM. The simulation results show that DAOI-ELM diagnoses the TEP faults better than other methods.},
	number = {12},
	journal = {IEEE Transactions on Cybernetics},
	author = {Zou, Weidong and Xia, Yuanqing and Li, Huifang},
	month = dec,
	year = {2018},
	note = {Conference Name: IEEE Transactions on Cybernetics},
	keywords = {Artificial neural networks, Driving amount, Fault diagnosis, Gram–Schmidt orthogonalization method, Network architecture, Predictive maintenance, Prognostics and health management, Tennessee-Eastman process (TEP), fault diagnosis},
	pages = {3403--3410},
}

Fault diagnosis is important to the industrial process. This paper proposes an orthogonal incremental extreme learning machine based on driving amount (DAOI-ELM) for recognizing the faults of the Tennessee-Eastman process (TEP). The basic idea of DAOI-ELM is to incorporate the Gram-Schmidt orthogonalization method and driving amount into an incremental extreme learning machine (I-ELM). The case study for the 2-D nonlinear function and regression problems from the UCI dataset results show that DAOI-ELM can obtain better generalization ability and a more compact structure of ELM than I-ELM, convex I-ELM (CI-ELM), orthogonal I-ELM (OI-ELM), and bidirectional ELM. The experimental training and testing data are derived from the simulations of TEP. The performance of DAOI-ELM is evaluated and compared with that of the back propagation neural network, support vector machine, I-ELM, CI-ELM, and OI-ELM. The simulation results show that DAOI-ELM diagnoses the TEP faults better than other methods.

@inproceedings{fazai_fault_2018,
	title = {Fault {Detection} of the {Tennessee} {Eastman} {Process} using {Online} {Reduced} {Kernel} {PCA}},
	doi = {10.23919/ECC.2018.8550213},
	abstract = {In this paper, we propose an online reduced kernel principal component analysis (KPCA) method for process monitoring. The developed method consists in updating the KPCA model depending on the dictionary which contains linearly independent kernel functions and then using this new reduced KPCA model for process monitoring. The process monitoring performances are studied using Tennessee Eastman Process (TEP). The results demonstrate the effectiveness of the developed online KPCA technique compared to the classical online KPCA method.},
	booktitle = {2018 {European} {Control} {Conference} ({ECC})},
	author = {Fazai, Radhia and Mansouri, Majdi and Taouali, Okba and Harkat, Mohamed-Faouzi and Bouguila, Nassreddine and Nounou, Mohamed},
	month = jun,
	year = {2018},
	keywords = {Dictionaries, Erbium, Indexes, Inductors, Kernel, Monitoring, Principal component analysis},
	pages = {2697--2702},
}

In this paper, we propose an online reduced kernel principal component analysis (KPCA) method for process monitoring. The developed method consists in updating the KPCA model depending on the dictionary which contains linearly independent kernel functions and then using this new reduced KPCA model for process monitoring. The process monitoring performances are studied using Tennessee Eastman Process (TEP). The results demonstrate the effectiveness of the developed online KPCA technique compared to the classical online KPCA method.

@article{pratama_autonomous_2018,
	title = {Autonomous {Deep} {Learning}: {Incremental} {Learning} of {Denoising} {Autoencoder} for {Evolving} {Data} {Streams}},
	shorttitle = {Autonomous {Deep} {Learning}},
	url = {http://arxiv.org/abs/1809.09081},
	abstract = {The generative learning phase of Autoencoder (AE) and its successor Denosing Autoencoder (DAE) enhances the flexibility of data stream method in exploiting unlabelled samples. Nonetheless, the feasibility of DAE for data stream analytic deserves in-depth study because it characterizes a fixed network capacity which cannot adapt to rapidly changing environments. An automated construction of a denoising autoeconder, namely deep evolving denoising autoencoder (DEVDAN), is proposed in this paper. DEVDAN features an open structure both in the generative phase and in the discriminative phase where input features can be automatically added and discarded on the fly. A network significance (NS) method is formulated in this paper and is derived from the bias-variance concept. This method is capable of estimating the statistical contribution of the network structure and its hidden units which precursors an ideal state to add or prune input features. Furthermore, DEVDAN is free of the problem- specific threshold and works fully in the single-pass learning fashion. The efficacy of DEVDAN is numerically validated using nine non-stationary data stream problems simulated under the prequential test-then-train protocol where DEVDAN is capable of delivering an improvement of classification accuracy to recently published online learning works while having flexibility in the automatic extraction of robust input features and in adapting to rapidly changing environments.},
	urldate = {2022-03-19},
	journal = {arXiv:1809.09081 [cs, stat]},
	author = {Pratama, Mahardhika and Ashfahani, Andri and Ong, Yew Soon and Ramasamy, Savitha and Lughofer, Edwin},
	month = sep,
	year = {2018},
	note = {arXiv: 1809.09081},
	keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning, Statistics - Machine Learning},
}

The generative learning phase of Autoencoder (AE) and its successor Denosing Autoencoder (DAE) enhances the flexibility of data stream method in exploiting unlabelled samples. Nonetheless, the feasibility of DAE for data stream analytic deserves in-depth study because it characterizes a fixed network capacity which cannot adapt to rapidly changing environments. An automated construction of a denoising autoeconder, namely deep evolving denoising autoencoder (DEVDAN), is proposed in this paper. DEVDAN features an open structure both in the generative phase and in the discriminative phase where input features can be automatically added and discarded on the fly. A network significance (NS) method is formulated in this paper and is derived from the bias-variance concept. This method is capable of estimating the statistical contribution of the network structure and its hidden units which precursors an ideal state to add or prune input features. Furthermore, DEVDAN is free of the problem- specific threshold and works fully in the single-pass learning fashion. The efficacy of DEVDAN is numerically validated using nine non-stationary data stream problems simulated under the prequential test-then-train protocol where DEVDAN is capable of delivering an improvement of classification accuracy to recently published online learning works while having flexibility in the automatic extraction of robust input features and in adapting to rapidly changing environments.

@article{sahoo_online_2018,
	title = {Online {Deep} {Learning}: {Learning} {Deep} {Neural} {Networks} on the {Fly}},
	shorttitle = {Online {Deep} {Learning}},
	url = {https://www.ijcai.org/proceedings/2018/369},
	abstract = {Electronic proceedings of IJCAI 2018},
	urldate = {2022-03-19},
	author = {Sahoo, Doyen and Pham, Quang and Lu, Jing and Hoi, Steven C. H.},
	year = {2018},
	pages = {2660--2666},
}

Electronic proceedings of IJCAI 2018

@article{montiel_scikit-multiflow_2018,
	title = {Scikit-multiflow: a multi-output streaming framework},
	volume = {19},
	issn = {1532-4435},
	shorttitle = {Scikit-multiflow},
	abstract = {scikit-multiflow is a framework for learning from data streams and multi-output learning in Python. Conceived to serve as a platform to encourage the democratization of stream learning research, it provides multiple state-of-the-art learning methods, data generators and evaluators for different stream learning problems, including single-output, multi-output and multi-label. scikit-multiflow builds upon popular open source frameworks including scikit-learn, MOA and MEKA. Development follows the FOSS principles. Quality is enforced by complying with PEP8 guidelines, using continuous integration and functional testing. The source code is available at https://github.com/scikit-multiflow/scikit-multiflow.},
	number = {1},
	journal = {The Journal of Machine Learning Research},
	author = {Montiel, Jacob and Read, Jesse and Bifet, Albert and Abdessalem, Talel},
	month = jan,
	year = {2018},
	keywords = {drift detection, machine learning, multi-output, python, stream data},
	pages = {2915--2914},
}

scikit-multiflow is a framework for learning from data streams and multi-output learning in Python. Conceived to serve as a platform to encourage the democratization of stream learning research, it provides multiple state-of-the-art learning methods, data generators and evaluators for different stream learning problems, including single-output, multi-output and multi-label. scikit-multiflow builds upon popular open source frameworks including scikit-learn, MOA and MEKA. Development follows the FOSS principles. Quality is enforced by complying with PEP8 guidelines, using continuous integration and functional testing. The source code is available at https://github.com/scikit-multiflow/scikit-multiflow.

@article{abdallah_activity_2018,
	title = {Activity {Recognition} with {Evolving} {Data} {Streams}: {A} {Review}},
	volume = {51},
	issn = {0360-0300},
	shorttitle = {Activity {Recognition} with {Evolving} {Data} {Streams}},
	url = {https://doi.org/10.1145/3158645},
	doi = {10.1145/3158645},
	abstract = {Activity recognition aims to provide accurate and opportune information on people’s activities by leveraging sensory data available in today’s sensory rich environments. Nowadays, activity recognition has become an emerging field in the areas of pervasive and ubiquitous computing. A typical activity recognition technique processes data streams that evolve from sensing platforms such as mobile sensors, on body sensors, and/or ambient sensors. This article surveys the two overlapped areas of research of activity recognition and data stream mining. The perspective of this article is to review the adaptation capabilities of activity recognition techniques in streaming environment. Categories of techniques are identified based on different features in both data streams and activity recognition. The pros and cons of the algorithms in each category are analysed, and the possible directions of future research are indicated.},
	number = {4},
	urldate = {2022-03-15},
	journal = {ACM Computing Surveys},
	author = {Abdallah, Zahraa S. and Gaber, Mohamed Medhat and Srinivasan, Bala and Krishnaswamy, Shonali},
	month = jul,
	year = {2018},
	keywords = {Activity recognition, adaptive learning, stream mining, transfer learning},
	pages = {71:1--71:36},
}

Activity recognition aims to provide accurate and opportune information on people’s activities by leveraging sensory data available in today’s sensory rich environments. Nowadays, activity recognition has become an emerging field in the areas of pervasive and ubiquitous computing. A typical activity recognition technique processes data streams that evolve from sensing platforms such as mobile sensors, on body sensors, and/or ambient sensors. This article surveys the two overlapped areas of research of activity recognition and data stream mining. The perspective of this article is to review the adaptation capabilities of activity recognition techniques in streaming environment. Categories of techniques are identified based on different features in both data streams and activity recognition. The pros and cons of the algorithms in each category are analysed, and the possible directions of future research are indicated.

@inproceedings{wu_design_2018,
	title = {Design on {Fault} {Diagnosis} {Expert} {System} for {Railway} {Signal} {Equipment}},
	isbn = {978-94-6252-531-3},
	url = {https://www.atlantis-press.com/proceedings/icmmct-18/25898536},
	doi = {10.2991/icmmct-18.2018.7},
	abstract = {In view of the fact that the fault of railway signal equipment is difficult to diagnose, this paper designs an expert system to simulate human experts to make decisions and improve the science of fault diagnosis. The design includes four aspects: First, a system structure composed of a database, a knowledge base, an inference engine, and an interpretation...},
	language = {en},
	urldate = {2022-03-05},
	publisher = {Atlantis Press},
	author = {Wu, Guangrong},
	month = jun,
	year = {2018},
	note = {ISSN: 2352-5401},
	pages = {36--41},
}

In view of the fact that the fault of railway signal equipment is difficult to diagnose, this paper designs an expert system to simulate human experts to make decisions and improve the science of fault diagnosis. The design includes four aspects: First, a system structure composed of a database, a knowledge base, an inference engine, and an interpretation...

@book{sutton_reinforcement_2018,
	title = {Reinforcement {Learning}, second edition: {An} {Introduction}},
	isbn = {978-0-262-35270-3},
	shorttitle = {Reinforcement {Learning}, second edition},
	abstract = {The significantly expanded and updated new edition of a widely used text on reinforcement learning, one of the most active research areas in artificial intelligence.Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives while interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the field's key ideas and algorithms. This second edition has been significantly expanded and updated, presenting new topics and updating coverage of other topics.Like the first edition, this second edition focuses on core online learning algorithms, with the more mathematical material set off in shaded boxes. Part I covers as much of reinforcement learning as possible without going beyond the tabular case for which exact solutions can be found. Many algorithms presented in this part are new to the second edition, including UCB, Expected Sarsa, and Double Learning. Part II extends these ideas to function approximation, with new sections on such topics as artificial neural networks and the Fourier basis, and offers expanded treatment of off-policy learning and policy-gradient methods. Part III has new chapters on reinforcement learning's relationships to psychology and neuroscience, as well as an updated case-studies chapter including AlphaGo and AlphaGo Zero, Atari game playing, and IBM Watson's wagering strategy. The final chapter discusses the future societal impacts of reinforcement learning.},
	language = {en},
	publisher = {MIT Press},
	author = {Sutton, Richard S. and Barto, Andrew G.},
	month = nov,
	year = {2018},
	note = {Google-Books-ID: uWV0DwAAQBAJ},
	keywords = {Computers / Artificial Intelligence / General},
}

The significantly expanded and updated new edition of a widely used text on reinforcement learning, one of the most active research areas in artificial intelligence.Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives while interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the field's key ideas and algorithms. This second edition has been significantly expanded and updated, presenting new topics and updating coverage of other topics.Like the first edition, this second edition focuses on core online learning algorithms, with the more mathematical material set off in shaded boxes. Part I covers as much of reinforcement learning as possible without going beyond the tabular case for which exact solutions can be found. Many algorithms presented in this part are new to the second edition, including UCB, Expected Sarsa, and Double Learning. Part II extends these ideas to function approximation, with new sections on such topics as artificial neural networks and the Fourier basis, and offers expanded treatment of off-policy learning and policy-gradient methods. Part III has new chapters on reinforcement learning's relationships to psychology and neuroscience, as well as an updated case-studies chapter including AlphaGo and AlphaGo Zero, Atari game playing, and IBM Watson's wagering strategy. The final chapter discusses the future societal impacts of reinforcement learning.

@inproceedings{korvesis_predictive_2018,
	title = {Predictive {Maintenance} in {Aviation}: {Failure} {Prediction} from {Post}-{Flight} {Reports}},
	shorttitle = {Predictive {Maintenance} in {Aviation}},
	doi = {10.1109/ICDE.2018.00160},
	abstract = {In this paper we present an approach to tackle the problem of event prediction for the purpose of performing predictive maintenance in aviation. Given a collection of recorded events that correspond to equipment failures, our method predicts the next occurrence of one or more events of interest (target events or critical failures). Our objective is to develop an alerting system that would notify aviation engineers well in advance for upcoming aircraft failures, providing enough time to prepare the corresponding maintenance actions. We formulate a regression problem in order to approximate the risk of occurrence of a target event, given the past occurrences of other events. In order to achieve the best results we employed a multiple instance learning scheme (multiple instance regression) along with extensive data preprocessing. We applied our method on data coming from a fleet of aircraft and our predictions involve failures of components onboard, specifically components that are related to the landing gear. The event logs correspond to post flight reports retrieved from multiple aircraft during several years of operation. To the best of our knowledge, this paper is the first attempt on aircraft failure prediction using post flight report data and finally, our findings show high potential impact on the aviation industry.},
	booktitle = {2018 {IEEE} 34th {International} {Conference} on {Data} {Engineering} ({ICDE})},
	author = {Korvesis, Panagiotis and Besseau, Stephane and Vazirgiannis, Michalis},
	month = apr,
	year = {2018},
	note = {ISSN: 2375-026X},
	keywords = {Aircraft, Electronic mail, Feature extraction, Hardware, Monitoring, Predictive maintenance, aviation, failure prediction, predictive maintenance},
	pages = {1414--1422},
}

In this paper we present an approach to tackle the problem of event prediction for the purpose of performing predictive maintenance in aviation. Given a collection of recorded events that correspond to equipment failures, our method predicts the next occurrence of one or more events of interest (target events or critical failures). Our objective is to develop an alerting system that would notify aviation engineers well in advance for upcoming aircraft failures, providing enough time to prepare the corresponding maintenance actions. We formulate a regression problem in order to approximate the risk of occurrence of a target event, given the past occurrences of other events. In order to achieve the best results we employed a multiple instance learning scheme (multiple instance regression) along with extensive data preprocessing. We applied our method on data coming from a fleet of aircraft and our predictions involve failures of components onboard, specifically components that are related to the landing gear. The event logs correspond to post flight reports retrieved from multiple aircraft during several years of operation. To the best of our knowledge, this paper is the first attempt on aircraft failure prediction using post flight report data and finally, our findings show high potential impact on the aviation industry.

@article{pan_novel_2018,
	title = {Novel battery state-of-health online estimation method using multiple health indicators and an extreme learning machine},
	volume = {160},
	issn = {0360-5442},
	url = {https://www.sciencedirect.com/science/article/pii/S0360544218312854},
	doi = {10.1016/j.energy.2018.06.220},
	abstract = {Battery health monitoring and management is critically important for electric vehicle performance and economy. This paper presents a multiple health indicators-based and machine learning-enabled state-of-health estimator for prognostics and health management. The multiple online health indicators without the influence of different loading profiles are used as effective signatures of the health estimator for effective quantification of capacity degradation. An extreme learning machine is introduced to capture the underlying correlation between the extracted health indicators and capacity degradation to improve the speed and accuracy of machine learning for online estimation. The proposed estimator is also compared to the traditional BP neural network. The associated results indicate that the maximum estimation error of the proposed health management strategy is less than 2.5\%, and it has better performance and faster speed than the BP neural network.},
	language = {en},
	urldate = {2022-02-09},
	journal = {Energy},
	author = {Pan, Haihong and Lü, Zhiqiang and Wang, Huimin and Wei, Haiyan and Chen, Lin},
	month = oct,
	year = {2018},
	keywords = {Extreme learning machine, Health indicator, Li-ion battery, State-of-Health, ecml},
	pages = {466--477},
}

Battery health monitoring and management is critically important for electric vehicle performance and economy. This paper presents a multiple health indicators-based and machine learning-enabled state-of-health estimator for prognostics and health management. The multiple online health indicators without the influence of different loading profiles are used as effective signatures of the health estimator for effective quantification of capacity degradation. An extreme learning machine is introduced to capture the underlying correlation between the extracted health indicators and capacity degradation to improve the speed and accuracy of machine learning for online estimation. The proposed estimator is also compared to the traditional BP neural network. The associated results indicate that the maximum estimation error of the proposed health management strategy is less than 2.5%, and it has better performance and faster speed than the BP neural network.

@article{zhao_machine_2018,
	title = {Machine {Health} {Monitoring} {Using} {Local} {Feature}-{Based} {Gated} {Recurrent} {Unit} {Networks}},
	volume = {65},
	issn = {1557-9948},
	doi = {10.1109/TIE.2017.2733438},
	abstract = {In modern industries, machine health monitoring systems (MHMS) have been applied wildly with the goal of realizing predictive maintenance including failures tracking, downtime reduction, and assets preservation. In the era of big machinery data, data-driven MHMS have achieved remarkable results in the detection of faults after the occurrence of certain failures (diagnosis) and prediction of the future working conditions and the remaining useful life (prognosis). The numerical representation for raw sensory data is the key stone for various successful MHMS. Conventional methods are the labor-extensive as they usually depend on handcrafted features, which require expert knowledge. Inspired by the success of deep learning methods that redefine representation learning from raw data, we propose local feature-based gated recurrent unit (LFGRU) networks. It is a hybrid approach that combines handcrafted feature design with automatic feature learning for machine health monitoring. First, features from windows of input time series are extracted. Then, an enhanced bidirectional GRU network is designed and applied on the generated sequence of local features to learn the representation. A supervised learning layer is finally trained to predict machine condition. Experiments on three machine health monitoring tasks: tool wear prediction, gearbox fault diagnosis, and incipient bearing fault detection verify the effectiveness and generalization of the proposed LFGRU.},
	number = {2},
	journal = {IEEE Transactions on Industrial Electronics},
	author = {Zhao, Rui and Wang, Dongzhe and Yan, Ruqiang and Mao, Kezhi and Shen, Fei and Wang, Jinjiang},
	month = feb,
	year = {2018},
	note = {Conference Name: IEEE Transactions on Industrial Electronics},
	keywords = {Computational modeling, Data mining, Fault diagnosis, Feature extraction, Logic gates, Monitoring, Sensors, ecml, feature engineering, feature extraction, feature leanring, gated recurrent unit (GRU), machine health monitoring (MHM), tool wear prediction},
	pages = {1539--1548},
}

In modern industries, machine health monitoring systems (MHMS) have been applied wildly with the goal of realizing predictive maintenance including failures tracking, downtime reduction, and assets preservation. In the era of big machinery data, data-driven MHMS have achieved remarkable results in the detection of faults after the occurrence of certain failures (diagnosis) and prediction of the future working conditions and the remaining useful life (prognosis). The numerical representation for raw sensory data is the key stone for various successful MHMS. Conventional methods are the labor-extensive as they usually depend on handcrafted features, which require expert knowledge. Inspired by the success of deep learning methods that redefine representation learning from raw data, we propose local feature-based gated recurrent unit (LFGRU) networks. It is a hybrid approach that combines handcrafted feature design with automatic feature learning for machine health monitoring. First, features from windows of input time series are extracted. Then, an enhanced bidirectional GRU network is designed and applied on the generated sequence of local features to learn the representation. A supervised learning layer is finally trained to predict machine condition. Experiments on three machine health monitoring tasks: tool wear prediction, gearbox fault diagnosis, and incipient bearing fault detection verify the effectiveness and generalization of the proposed LFGRU.

@article{ben_ali_online_2018,
	title = {Online automatic diagnosis of wind turbine bearings progressive degradations under real experimental conditions based on unsupervised machine learning},
	volume = {132},
	issn = {0003-682X},
	url = {https://www.sciencedirect.com/science/article/pii/S0003682X1730333X},
	doi = {10.1016/j.apacoust.2017.11.021},
	abstract = {As a critical component, failures of high-speed shaft bearing in wind turbines cause the unplanned stoppage of electrical energy production. Investigations related to naturally progressed defects of high-speed shaft bearings are relatively scarce and the online assessment in damage severities is rarely available in the literature. In this sense, this paper presents a new online vibration-based diagnosis method for wind turbine high-speed bearing monitoring. The adaptive resonance theory 2 (ART2) is proposed for an unsupervised classification of the extracted features. The Randall model is adapted considering the geometry of the tested bearing to train the ART2 in the offline step. In fact, the time domain, the frequency domain, and the time-frequency domain are investigated for a better bearing fault characterization. Indeed, the use of real measured data from a wind turbine drivetrain proves that the proposed data-driven approach is suitable for wind turbine bearings online condition monitoring even under real experimental conditions. This method reveals a better generalization capability compared to previous works even with noisy measurements.},
	language = {en},
	urldate = {2022-02-09},
	journal = {Applied Acoustics},
	author = {Ben Ali, Jaouher and Saidi, Lotfi and Harrath, Salma and Bechhoefer, Eric and Benbouzid, Mohamed},
	month = mar,
	year = {2018},
	keywords = {ART2, Fault diagnosis, Feature extraction, High speed shaft bearing, Wind turbines, ecml},
	pages = {167--181},
}

As a critical component, failures of high-speed shaft bearing in wind turbines cause the unplanned stoppage of electrical energy production. Investigations related to naturally progressed defects of high-speed shaft bearings are relatively scarce and the online assessment in damage severities is rarely available in the literature. In this sense, this paper presents a new online vibration-based diagnosis method for wind turbine high-speed bearing monitoring. The adaptive resonance theory 2 (ART2) is proposed for an unsupervised classification of the extracted features. The Randall model is adapted considering the geometry of the tested bearing to train the ART2 in the offline step. In fact, the time domain, the frequency domain, and the time-frequency domain are investigated for a better bearing fault characterization. Indeed, the use of real measured data from a wind turbine drivetrain proves that the proposed data-driven approach is suitable for wind turbine bearings online condition monitoring even under real experimental conditions. This method reveals a better generalization capability compared to previous works even with noisy measurements.

@inproceedings{rieth_issues_2018,
	address = {Cham},
	series = {Advances in {Intelligent} {Systems} and {Computing}},
	title = {Issues and {Advances} in {Anomaly} {Detection} {Evaluation} for {Joint} {Human}-{Automated} {Systems}},
	isbn = {978-3-319-60384-1},
	doi = {10.1007/978-3-319-60384-1_6},
	abstract = {As human-managed systems become more complex, automated anomaly detection can provide assistance—but only if it is effective. Rigorous evaluation of automated detection is vital for determining its effectiveness before implementation into systems. We identified recurring issues in evaluation practices limiting the conclusions that can be applied from published studies to broader application. In this paper, we demonstrate the implications of these issues and illustrate solutions. We show how receiver operating characteristic curves can reveal performance tradeoffs masked by reporting of single metric results and how using multiple simulation data examples can prevent biases that result from evaluation using single training and testing examples. We also provide methods for incorporating detection latency into tradeoff analyses. Application of these methods will help to provide researchers, engineers, and decision makers with a more objective basis for anomaly detection performance evaluation, resulting in greater utility, better performance, and cost savings in systems engineering.},
	language = {en},
	booktitle = {Advances in {Human} {Factors} in {Robots} and {Unmanned} {Systems}},
	publisher = {Springer International Publishing},
	author = {Rieth, Cory A. and Amsel, Ben D. and Tran, Randy and Cook, Maia B.},
	editor = {Chen, Jessie},
	year = {2018},
	keywords = {Anomaly detection, Automation evaluation, Receiver operating characteristic, Tennessee Eastman process simulation},
	pages = {52--63},
}

As human-managed systems become more complex, automated anomaly detection can provide assistance—but only if it is effective. Rigorous evaluation of automated detection is vital for determining its effectiveness before implementation into systems. We identified recurring issues in evaluation practices limiting the conclusions that can be applied from published studies to broader application. In this paper, we demonstrate the implications of these issues and illustrate solutions. We show how receiver operating characteristic curves can reveal performance tradeoffs masked by reporting of single metric results and how using multiple simulation data examples can prevent biases that result from evaluation using single training and testing examples. We also provide methods for incorporating detection latency into tradeoff analyses. Application of these methods will help to provide researchers, engineers, and decision makers with a more objective basis for anomaly detection performance evaluation, resulting in greater utility, better performance, and cost savings in systems engineering.

@inproceedings{gulenko_detecting_2018,
	title = {Detecting {Anomalous} {Behavior} of {Black}-{Box} {Services} {Modeled} with {Distance}-{Based} {Online} {Clustering}},
	doi = {10.1109/CLOUD.2018.00134},
	abstract = {Reliable deployment of services is especially challenging in virtualized infrastructures, where the deep tech-nological stack and the multitude of components necessitate automatic anomaly detection and remediation mechanisms. Traditional monitoring solutions observe the system and generate alarms when the collected metrics exceed predefined thresholds. The fixed thresholds rely on expert knowledge and can lead to numerous false alarms, while abnormal behavior that spans over multiple metrics, components, or system layers, may not be detected. We propose to use an unsupervised online clustering algorithm to create a model of the normal behavior of each monitored component with minimal human interaction and no impact on the monitored system. When an anomaly is detected, a human administrator or automatic remediation system can subsequently revert the component into a normal state. An experimental evaluation resulted in a high accuracy of our approach, indicating that it is suitable for anomaly detection in productive systems.},
	booktitle = {2018 {IEEE} 11th {International} {Conference} on {Cloud} {Computing} ({CLOUD})},
	author = {Gulenko, Anton and Schmidt, Florian and Acker, Alexander and Wallschläger, Marcel and Kao, Odej and Liu, Feng},
	month = jul,
	year = {2018},
	note = {ISSN: 2159-6190},
	keywords = {Anomaly detection, Cloud computing, Data collection, Data models, Measurement, Monitoring, Virtual machine monitors, anomaly detection, cloud computing, machine learning, service virtualization},
	pages = {912--915},
}

Reliable deployment of services is especially challenging in virtualized infrastructures, where the deep tech-nological stack and the multitude of components necessitate automatic anomaly detection and remediation mechanisms. Traditional monitoring solutions observe the system and generate alarms when the collected metrics exceed predefined thresholds. The fixed thresholds rely on expert knowledge and can lead to numerous false alarms, while abnormal behavior that spans over multiple metrics, components, or system layers, may not be detected. We propose to use an unsupervised online clustering algorithm to create a model of the normal behavior of each monitored component with minimal human interaction and no impact on the monitored system. When an anomaly is detected, a human administrator or automatic remediation system can subsequently revert the component into a normal state. An experimental evaluation resulted in a high accuracy of our approach, indicating that it is suitable for anomaly detection in productive systems.

@article{rodriguez-ramos_approach_2018,
	title = {An approach to fault diagnosis with online detection of novel faults using fuzzy clustering tools},
	volume = {113},
	issn = {0957-4174},
	url = {https://www.sciencedirect.com/science/article/pii/S0957417418304135},
	doi = {10.1016/j.eswa.2018.06.055},
	abstract = {This paper presents an approach to fault diagnosis with online detection of novel faults and automatic learning using fuzzy clustering techniques. In the off-line learning stage, the classifier is trained to diagnose the known faults and the normal operation state using the Density Oriented Fuzzy C-Means and the Kernel Fuzzy C-Means algorithms. In this stage, the historical data previously selected by experts, are firstly pre-processed to eliminate outliers and reduce the confusion in the classification process by using the Density Oriented Fuzzy C-Means algorithm. Later on, the Kernel Fuzzy C-Means algorithm is used for achieving greater separability among the classes and reducing the classification errors. Finally, the optimization of the two parameters used by these algorithms in the training stage is developed by using a bio-inspired optimization algorithm, namely the differential evolution. After the training, the classifier is used online (online diagnosis stage) in order to classify the new observations that are collected from the process. In this stage, the detection of novel faults based on density by using the DOFCM algorithm is applied. The algorithm analyzes the observations belonging to a window of time which were not classified into the known classes and it is determined if they are a new class or outliers. If a new class is identified, a procedure is developed to incorporate it to the known classes set. The proposed approach was validated using the Development and Application of Methods for Actuator Diagnosis in Industrial Control Systems (DAMADICS) benchmark. The results obtained indicate the feasibility of the proposed method.},
	language = {en},
	urldate = {2022-01-14},
	journal = {Expert Systems with Applications},
	author = {Rodríguez-Ramos, Adrián and da Silva Neto, Antônio José and Llanes-Santiago, Orestes},
	month = dec,
	year = {2018},
	keywords = {Automatic learning, Fuzzy clustering tools, Novel faults, On-line detection, Optimal parameters, Robust fault diagnosis},
	pages = {200--212},
}

This paper presents an approach to fault diagnosis with online detection of novel faults and automatic learning using fuzzy clustering techniques. In the off-line learning stage, the classifier is trained to diagnose the known faults and the normal operation state using the Density Oriented Fuzzy C-Means and the Kernel Fuzzy C-Means algorithms. In this stage, the historical data previously selected by experts, are firstly pre-processed to eliminate outliers and reduce the confusion in the classification process by using the Density Oriented Fuzzy C-Means algorithm. Later on, the Kernel Fuzzy C-Means algorithm is used for achieving greater separability among the classes and reducing the classification errors. Finally, the optimization of the two parameters used by these algorithms in the training stage is developed by using a bio-inspired optimization algorithm, namely the differential evolution. After the training, the classifier is used online (online diagnosis stage) in order to classify the new observations that are collected from the process. In this stage, the detection of novel faults based on density by using the DOFCM algorithm is applied. The algorithm analyzes the observations belonging to a window of time which were not classified into the known classes and it is determined if they are a new class or outliers. If a new class is identified, a procedure is developed to incorporate it to the known classes set. The proposed approach was validated using the Development and Application of Methods for Actuator Diagnosis in Industrial Control Systems (DAMADICS) benchmark. The results obtained indicate the feasibility of the proposed method.

@article{wang_particle_2018,
	title = {Particle swarm optimization algorithm: an overview},
	volume = {22},
	issn = {1433-7479},
	shorttitle = {Particle swarm optimization algorithm},
	url = {https://doi.org/10.1007/s00500-016-2474-6},
	doi = {10.1007/s00500-016-2474-6},
	abstract = {Particle swarm optimization (PSO) is a population-based stochastic optimization algorithm motivated by intelligent collective behavior of some animals such as flocks of birds or schools of fish. Since presented in 1995, it has experienced a multitude of enhancements. As researchers have learned about the technique, they derived new versions aiming to different demands, developed new applications in a host of areas, published theoretical studies of the effects of the various parameters and proposed many variants of the algorithm. This paper introduces its origin and background and carries out the theory analysis of the PSO. Then, we analyze its present situation of research and application in algorithm structure, parameter selection, topology structure, discrete PSO algorithm and parallel PSO algorithm, multi-objective optimization PSO and its engineering applications. Finally, the existing problems are analyzed and future research directions are presented.},
	language = {en},
	number = {2},
	urldate = {2021-11-29},
	journal = {Soft Computing},
	author = {Wang, Dongshu and Tan, Dapei and Liu, Lei},
	month = jan,
	year = {2018},
	pages = {387--408},
}

Particle swarm optimization (PSO) is a population-based stochastic optimization algorithm motivated by intelligent collective behavior of some animals such as flocks of birds or schools of fish. Since presented in 1995, it has experienced a multitude of enhancements. As researchers have learned about the technique, they derived new versions aiming to different demands, developed new applications in a host of areas, published theoretical studies of the effects of the various parameters and proposed many variants of the algorithm. This paper introduces its origin and background and carries out the theory analysis of the PSO. Then, we analyze its present situation of research and application in algorithm structure, parameter selection, topology structure, discrete PSO algorithm and parallel PSO algorithm, multi-objective optimization PSO and its engineering applications. Finally, the existing problems are analyzed and future research directions are presented.

@inproceedings{wadhwa_modified_2018,
	title = {Modified {DBSCAN} {Using} {Particle} {Swarm} {Optimization} for {Spatial} {Hotspot} {Identification}},
	doi = {10.1109/IC3.2018.8530558},
	abstract = {Spatial hotspots of irregular shape occur naturally in fields like epidemiology and earth science. Classical techniques for identifying hotspots are either based on scan statistics or clustering algorithms. These techniques result in hotspots of fixed shapes like circle, ellipse or straight line. Density based spatial clustering of applications with noise (DBSCAN) is one of the often used algorithms for finding non-geometric shaped clusters. It is highly sensitive to the values of its input variables (MinPoints and Epsilon) which are to be provided by the users. In this paper, we propose a Particle Swarm Optimization (PSO) based approach which automatically computes the values of MinPoints and Epsilon for given input data and finds the spatial hotspots. The modified DBSCAN approach is applied to six artificial datasets and purity of the resultant clustering is calculated. Achieved values of the purity function indicate the accuracy of the proposed method. Proposed approach is also applied to find out the hotspots for earthquake zoning.},
	booktitle = {2018 {Eleventh} {International} {Conference} on {Contemporary} {Computing} ({IC3})},
	author = {Wadhwa, Ankita and Thakur, Manish K.},
	month = aug,
	year = {2018},
	note = {ISSN: 2572-6129},
	keywords = {Clustering, Clustering algorithms, Conferences, DBSCAN, Earthquake Zoning, Earthquakes, Hotspots, Microsoft Windows, Particle Swarm optimization, Particle swarm optimization, Scan statistics, Shape, Spatial databases},
	pages = {1--3},
}

Spatial hotspots of irregular shape occur naturally in fields like epidemiology and earth science. Classical techniques for identifying hotspots are either based on scan statistics or clustering algorithms. These techniques result in hotspots of fixed shapes like circle, ellipse or straight line. Density based spatial clustering of applications with noise (DBSCAN) is one of the often used algorithms for finding non-geometric shaped clusters. It is highly sensitive to the values of its input variables (MinPoints and Epsilon) which are to be provided by the users. In this paper, we propose a Particle Swarm Optimization (PSO) based approach which automatically computes the values of MinPoints and Epsilon for given input data and finds the spatial hotspots. The modified DBSCAN approach is applied to six artificial datasets and purity of the resultant clustering is calculated. Achieved values of the purity function indicate the accuracy of the proposed method. Proposed approach is also applied to find out the hotspots for earthquake zoning.

@inproceedings{zheng_dags_2018,
	title = {{DAGs} with {NO} {TEARS}: {Continuous} {Optimization} for {Structure} {Learning}},
	volume = {31},
	url = {https://proceedings.neurips.cc/paper/2018/file/e347c51419ffb23ca3fd5050202f9c3d-Paper.pdf},
	booktitle = {Advances in {Neural} {Information} {Processing} {Systems}},
	publisher = {Curran Associates, Inc.},
	author = {Zheng, Xun and Aragam, Bryon and Ravikumar, Pradeep K and Xing, Eric P},
	editor = {Bengio, S. and Wallach, H. and Larochelle, H. and Grauman, K. and Cesa-Bianchi, N. and Garnett, R.},
	year = {2018},
}