A multi-sensor approach for early detection and notification of approaching trains. Shrestha, P., Hempel, M., Santos, J., & Sharif, H. In 2014.
doi  abstract   bibtex   
Personnel safety is a cornerstone of railroad operations. However, various activities such as track maintenance and repair require workers to be physically present on or near railroad tracks that are potentially active. This creates a significant risk to the workers, especially at very high train speeds and in places where the geographical topography reduces direct line of sight. In fact, fatalities and injuries continue to be reported. A common approach to lessen this problem is to employ workers specifically to act as lookouts - to observe approaching trains and report them back to the work site. However, maintenance operations are often lengthy and it is common for the lookouts to lose concentration and thus the possibility arises that they fail to report incoming trains. Another possibility is the failure of the communication link between the observer and the workers. In an attempt to remove the human element and the associated risks, commercial vendors have developed automated systems to detect approaching trains and trigger alarms. However, the current commercial solutions available have several drawbacks. Firstly, deploying the available solutions requires procedures that are invasive and possibly destructive for the railroad tracks. They often involve digging underneath the tracks to install devices. This also requires significant installation effort and makes the solution often nonportable. This has serious operating concerns for the railroads since maintenance operations are temporary and installing permanent devices becomes inefficient. Also, since the solutions are not portable, deploying them on a large scale is not cost effective. Secondly, the solutions have limited sensing capabilities, employing only a single detection approach such as mechanical treadles, and do not provide comprehensive information about the trains being detected. For example, if the maintenance is occurring on parallel tracks, the sensors may fail to detect which of the tracks the train is currently operating on. Also, they cannot accurately capture all the characteristics of the approaching train, such as length, velocity, identifying signature, etc. And thirdly, they often have high false detection rates as they lack methods to differentiate the signature of a train from other events, such as drilling and heavy machinery operation. With the objective of addressing all these issues, we have proposed a multi-sensor based system that can integrate the abilities of the different sensors so as to reliably predict all the essential characteristics of a train. In this paper we present the architecture of our multi-sensory system. We also report our findings and analysis of evaluating our sensor systems on tracks at the Transportation Technology Center, Inc. (TTCI) in Pueblo, CO. © 2014 by ASME.
@inproceedings{Shrestha2014,
   abstract = {Personnel safety is a cornerstone of railroad operations. However, various activities such as track maintenance and repair require workers to be physically present on or near railroad tracks that are potentially active. This creates a significant risk to the workers, especially at very high train speeds and in places where the geographical topography reduces direct line of sight. In fact, fatalities and injuries continue to be reported. A common approach to lessen this problem is to employ workers specifically to act as lookouts - to observe approaching trains and report them back to the work site. However, maintenance operations are often lengthy and it is common for the lookouts to lose concentration and thus the possibility arises that they fail to report incoming trains. Another possibility is the failure of the communication link between the observer and the workers. In an attempt to remove the human element and the associated risks, commercial vendors have developed automated systems to detect approaching trains and trigger alarms. However, the current commercial solutions available have several drawbacks. Firstly, deploying the available solutions requires procedures that are invasive and possibly destructive for the railroad tracks. They often involve digging underneath the tracks to install devices. This also requires significant installation effort and makes the solution often nonportable. This has serious operating concerns for the railroads since maintenance operations are temporary and installing permanent devices becomes inefficient. Also, since the solutions are not portable, deploying them on a large scale is not cost effective. Secondly, the solutions have limited sensing capabilities, employing only a single detection approach such as mechanical treadles, and do not provide comprehensive information about the trains being detected. For example, if the maintenance is occurring on parallel tracks, the sensors may fail to detect which of the tracks the train is currently operating on. Also, they cannot accurately capture all the characteristics of the approaching train, such as length, velocity, identifying signature, etc. And thirdly, they often have high false detection rates as they lack methods to differentiate the signature of a train from other events, such as drilling and heavy machinery operation. With the objective of addressing all these issues, we have proposed a multi-sensor based system that can integrate the abilities of the different sensors so as to reliably predict all the essential characteristics of a train. In this paper we present the architecture of our multi-sensory system. We also report our findings and analysis of evaluating our sensor systems on tracks at the Transportation Technology Center, Inc. (TTCI) in Pueblo, CO. © 2014 by ASME.},
   author = {P.L. Shrestha and M. Hempel and J. Santos and H. Sharif},
   doi = {10.1115/JRC2014-3729},
   isbn = {9780791845356},
   journal = {2014 Joint Rail Conference, JRC 2014},
   title = {A multi-sensor approach for early detection and notification of approaching trains},
   year = {2014},
}

Downloads: 0