Automated Defect Detection From Ultrasonic Images Using Deep Learning. Medak, D., Posilovic, L., Subasic, M., Budimir, M., & Loncaric, S. IEEE TRANSACTIONS ON ULTRASONICS FERROELECTRICS AND FREQUENCY CONTROL, 68(10):3126–3134, IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC, 445 HOES LANE, PISCATAWAY, NJ 08855-4141 USA, October, 2021. doi abstract bibtex Nondestructive evaluation (NDE) is a set of techniques used for material inspection and defect detection without causing damage to the inspected component. One of the commonly used nondestructive techniques is called ultrasonic inspection. The acquisition of ultrasonic data was mostly automated in recent years, but the analysis of the collected data is still performed manually. This process is thus very expensive, inconsistent, and prone to human errors. An automated system would significantly increase the efficiency of analysis, but the methods presented so far fail to generalize well on new cases and are not used in real-life inspection. Many of the similar data analysis problems were recently tackled by deep learning methods. This approach outperforms classical methods but requires lots of training data, which is difficult to obtain in the NDE domain. In this work, we train a deep learning architecture EfficientDet to automatically detect defects from ultrasonic images. We showed how some of the hyperparameters can be tweaked in order to improve the detection of defects with extreme aspect ratios that are common in ultrasonic images. The proposed object detector was trained on the largest dataset of ultrasonic images that was so far seen in the literature. In order to collect the dataset, six steel blocks containing 68 defects were scanned with a phased-array probe. More than 4000 VC-B-scans were acquired and used for training and evaluation of EfficientDet. The proposed model achieved 89.6% of mean average precision (mAP) during fivefold cross validation, which is a significant improvement compared to some similar methods that were previously used for this task. A detailed performance overview for each of the folds revealed that EfficientDet-D0 successfully detects all of the defects present in the inspected material.
@article{WOS:000701250000010,
abstract = {Nondestructive evaluation (NDE) is a set of techniques used for material
inspection and defect detection without causing damage to the inspected
component. One of the commonly used nondestructive techniques is called
ultrasonic inspection. The acquisition of ultrasonic data was mostly
automated in recent years, but the analysis of the collected data is
still performed manually. This process is thus very expensive,
inconsistent, and prone to human errors. An automated system would
significantly increase the efficiency of analysis, but the methods
presented so far fail to generalize well on new cases and are not used
in real-life inspection. Many of the similar data analysis problems were
recently tackled by deep learning methods. This approach outperforms
classical methods but requires lots of training data, which is difficult
to obtain in the NDE domain. In this work, we train a deep learning
architecture EfficientDet to automatically detect defects from
ultrasonic images. We showed how some of the hyperparameters can be
tweaked in order to improve the detection of defects with extreme aspect
ratios that are common in ultrasonic images. The proposed object
detector was trained on the largest dataset of ultrasonic images that
was so far seen in the literature. In order to collect the dataset, six
steel blocks containing 68 defects were scanned with a phased-array
probe. More than 4000 VC-B-scans were acquired and used for training and
evaluation of EfficientDet. The proposed model achieved 89.6\% of mean
average precision (mAP) during fivefold cross validation, which is a
significant improvement compared to some similar methods that were
previously used for this task. A detailed performance overview for each
of the folds revealed that EfficientDet-D0 successfully detects all of
the defects present in the inspected material.},
address = {445 HOES LANE, PISCATAWAY, NJ 08855-4141 USA},
author = {Medak, Duje and Posilovic, Luka and Subasic, Marko and Budimir, Marko and Loncaric, Sven},
doi = {10.1109/TUFFC.2021.3081750},
issn = {0885-3010},
journal = {IEEE TRANSACTIONS ON ULTRASONICS FERROELECTRICS AND FREQUENCY CONTROL},
keywords = {Acoustics; Deep learning; Detectors; Probes; Train},
month = oct,
number = {10},
pages = {3126--3134},
publisher = {IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC},
title = {{Automated Defect Detection From Ultrasonic Images Using Deep Learning}},
type = {Article},
volume = {68},
year = {2021}
}
Downloads: 0
{"_id":"cEyJGNneGxyGL82hE","bibbaseid":"medak-posilovic-subasic-budimir-loncaric-automateddefectdetectionfromultrasonicimagesusingdeeplearning-2021","author_short":["Medak, D.","Posilovic, L.","Subasic, M.","Budimir, M.","Loncaric, S."],"bibdata":{"bibtype":"article","type":"Article","abstract":"Nondestructive evaluation (NDE) is a set of techniques used for material inspection and defect detection without causing damage to the inspected component. One of the commonly used nondestructive techniques is called ultrasonic inspection. The acquisition of ultrasonic data was mostly automated in recent years, but the analysis of the collected data is still performed manually. This process is thus very expensive, inconsistent, and prone to human errors. An automated system would significantly increase the efficiency of analysis, but the methods presented so far fail to generalize well on new cases and are not used in real-life inspection. Many of the similar data analysis problems were recently tackled by deep learning methods. This approach outperforms classical methods but requires lots of training data, which is difficult to obtain in the NDE domain. In this work, we train a deep learning architecture EfficientDet to automatically detect defects from ultrasonic images. We showed how some of the hyperparameters can be tweaked in order to improve the detection of defects with extreme aspect ratios that are common in ultrasonic images. The proposed object detector was trained on the largest dataset of ultrasonic images that was so far seen in the literature. In order to collect the dataset, six steel blocks containing 68 defects were scanned with a phased-array probe. More than 4000 VC-B-scans were acquired and used for training and evaluation of EfficientDet. The proposed model achieved 89.6% of mean average precision (mAP) during fivefold cross validation, which is a significant improvement compared to some similar methods that were previously used for this task. A detailed performance overview for each of the folds revealed that EfficientDet-D0 successfully detects all of the defects present in the inspected material.","address":"445 HOES LANE, PISCATAWAY, NJ 08855-4141 USA","author":[{"propositions":[],"lastnames":["Medak"],"firstnames":["Duje"],"suffixes":[]},{"propositions":[],"lastnames":["Posilovic"],"firstnames":["Luka"],"suffixes":[]},{"propositions":[],"lastnames":["Subasic"],"firstnames":["Marko"],"suffixes":[]},{"propositions":[],"lastnames":["Budimir"],"firstnames":["Marko"],"suffixes":[]},{"propositions":[],"lastnames":["Loncaric"],"firstnames":["Sven"],"suffixes":[]}],"doi":"10.1109/TUFFC.2021.3081750","issn":"0885-3010","journal":"IEEE TRANSACTIONS ON ULTRASONICS FERROELECTRICS AND FREQUENCY CONTROL","keywords":"Acoustics; Deep learning; Detectors; Probes; Train","month":"October","number":"10","pages":"3126–3134","publisher":"IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC","title":"Automated Defect Detection From Ultrasonic Images Using Deep Learning","volume":"68","year":"2021","bibtex":"@article{WOS:000701250000010,\nabstract = {Nondestructive evaluation (NDE) is a set of techniques used for material\ninspection and defect detection without causing damage to the inspected\ncomponent. One of the commonly used nondestructive techniques is called\nultrasonic inspection. The acquisition of ultrasonic data was mostly\nautomated in recent years, but the analysis of the collected data is\nstill performed manually. This process is thus very expensive,\ninconsistent, and prone to human errors. An automated system would\nsignificantly increase the efficiency of analysis, but the methods\npresented so far fail to generalize well on new cases and are not used\nin real-life inspection. Many of the similar data analysis problems were\nrecently tackled by deep learning methods. This approach outperforms\nclassical methods but requires lots of training data, which is difficult\nto obtain in the NDE domain. In this work, we train a deep learning\narchitecture EfficientDet to automatically detect defects from\nultrasonic images. We showed how some of the hyperparameters can be\ntweaked in order to improve the detection of defects with extreme aspect\nratios that are common in ultrasonic images. The proposed object\ndetector was trained on the largest dataset of ultrasonic images that\nwas so far seen in the literature. In order to collect the dataset, six\nsteel blocks containing 68 defects were scanned with a phased-array\nprobe. More than 4000 VC-B-scans were acquired and used for training and\nevaluation of EfficientDet. The proposed model achieved 89.6\\% of mean\naverage precision (mAP) during fivefold cross validation, which is a\nsignificant improvement compared to some similar methods that were\npreviously used for this task. A detailed performance overview for each\nof the folds revealed that EfficientDet-D0 successfully detects all of\nthe defects present in the inspected material.},\naddress = {445 HOES LANE, PISCATAWAY, NJ 08855-4141 USA},\nauthor = {Medak, Duje and Posilovic, Luka and Subasic, Marko and Budimir, Marko and Loncaric, Sven},\ndoi = {10.1109/TUFFC.2021.3081750},\nissn = {0885-3010},\njournal = {IEEE TRANSACTIONS ON ULTRASONICS FERROELECTRICS AND FREQUENCY CONTROL},\nkeywords = {Acoustics; Deep learning; Detectors; Probes; Train},\nmonth = oct,\nnumber = {10},\npages = {3126--3134},\npublisher = {IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC},\ntitle = {{Automated Defect Detection From Ultrasonic Images Using Deep Learning}},\ntype = {Article},\nvolume = {68},\nyear = {2021}\n}\n","author_short":["Medak, D.","Posilovic, L.","Subasic, M.","Budimir, M.","Loncaric, S."],"key":"WOS:000701250000010","id":"WOS:000701250000010","bibbaseid":"medak-posilovic-subasic-budimir-loncaric-automateddefectdetectionfromultrasonicimagesusingdeeplearning-2021","role":"author","urls":{},"keyword":["Acoustics; Deep learning; Detectors; Probes; Train"],"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/f/rTKwnBhyGKSSptYtt/My Collection.bib","dataSources":["DY3AeP9t2QujfB78L","MpTxM7aR49zFotABd"],"keywords":["acoustics; deep learning; detectors; probes; train"],"search_terms":["automated","defect","detection","ultrasonic","images","using","deep","learning","medak","posilovic","subasic","budimir","loncaric"],"title":"Automated Defect Detection From Ultrasonic Images Using Deep Learning","year":2021}