Principal Investigator

Jacobo Bielak

Status

Completed

Start Date

Jan. 1, 2016

End Date

Dec. 31, 2016

Project Type

Research Applied

Grant Program

MAP-21 TSET National (2013 - 2018)

Grant Cycle

2016 TSET UTC

Visibility

Public

Abstract

Globally, infrastructure is a vital asset for economic prosperity. However, condition assessments to ensure the continuing safe operation of these assets tend to be subjective and infrequent. We are developing a low-cost, objective method to continuously monitor rail systems from the vibrations recorded in a passing train as a complement to traditional inspection techniques. The concept of using passenger trains to probe infrastructure assets has been proposed previously, but most work has been limited to simulations, laboratory tests, and short-term experimentation. With previous support from the University Transportation Center, we have embarked on a long- term monitoring project with the Port Authority of Allegheny County. We have developed and deployed a robust automatic data acquisition and management system. Through this deployment we have been testing and refining our technique. We have instrumented two of Pittsburgh’s light rail vehicles and have been monitoring the rail system from the vibrations in the operational vehicle over the last two years. Our objective is to provide accurate, rapid, nearly continuous assessments of the tracks, track structures and bridges along the line as cost-efficiently as possible. We already have seen promising results, as we have been able to detect changes in the recorded signals after construction activity on both the track and ballast. As the next stage of our overall project, we propose to develop a systematic method of data analysis to fuse data from multiple instrumented vehicles, and detect changes in the condition of the rail/track system, and, thus, provide better insights for the Port Authority on high-priority maintenance issues. Ultimately, our goal is to develop an infrastructure asset management technology that can serve not only the Port Authority of Allegheny County but also infrastructure owners for a variety of transportation modes. Through this improved maintenance technology, we will enable safer more efficient transportation. 
    

Description

We propose to build upon our long term infrastructure monitoring project by (1) developing our ability to fuse and analyze data from multiple vehicles (2) integrating video of the track into our baseline data, and (3) expanding on the types of infrastructure changes we can detect. We have continued to work with the Port Authority over the last year, improving our data collection (instrumenting a second train), our data analysis (identifying tamping) and our understanding of the train’s behavior through simulations. We will leverage this partnership and our successes over the last year to achieve our goals in the next phase of the project.  
	Currently, the Port Authority maintains over 40 km of track, much of it built in the early 20th century. They monitor their system by sending inspectors to walk the track at least every quarter, and hire a vendor to perform ultrasonic inspection of the track on a yearly basis. Between these infrequent inspections, they rely primarily on the train operators (the drivers) to report anomalies or issues. 
To augment their knowledge of the system, we have instrumented two operational light-rail vehicles with accelerometers (Figure 1) to continuously collect information. The instrumentation of a second vehicle, completed this June, has more sensors, a better data acquisition system, and a more precise GPS antenna. While this second vehicle presents the opportunity of higher fidelity information, leveraging the data of more than one vehicle has been a problem from researchers in the past [1]. We will explore methods to learn from this heterogeneous data collected by several different types of sensors, placed a different locations within the train, and sampling at different rates. In addition, by comparing the information from the various sensors, we plan to evaluate the minimum amount of information needed to detect different types of changes in the infrastructure. In particular, we will look at the efficacy of different sensor resolutions, sampling frequencies, and noise levels, to see if cheaper sensors gathering less information could be used if the system were scaled up. 
Over the past year, we have used the data from the first train to successfully identify tamping work and we have simulated the response of the train to understand which features are most effective to detect this type of change. The simulation results are summarized in Figure 2, where we look at our ability to detect tamping using four different features for a variety of position uncertainty levels. This uncertainty is indicative of the error in the GPS location. Uncertainty level 1 means an error of up to 1m; the actual level of error we observe is around 6m, or level 6. From this figure, we see that a feature based on the energy of the signal is the most robust to position uncertainty, a similar feature to ones used successfully in the past [2]. 
We then used this energy feature to detect a tamping event as seen in Figure 3. Here we have plotted the energy we observe from the train, averaged over 8 passes, both before and after a tamping event. On the right, we show the correlation of these average energy signals, which indicates that the greatest change occurred where tamping occurred. Our ability to detect and localize tamping, a relatively minor change, over a large sections of track (we show 7km of track here) demonstrates the strength of this method for monitoring large infrastructure installations.

While the detection of tamping has served as a validation of our system, there are other phenomena that are of great interest to the Port Authority. Over the next year we will work to identify other types of more pertinent track changes and track locations which may be causing wheel chipping. The latter is of particular interest to the Port Authority because when metal chips off a wheel, the entire car must be taken out of service until the wheel can be ground. Last year the grinding efforts could not keep pace with the number of chips that occurred. In July 2015 we recorded video footage of the interaction between the wheel and the rail over the entire network, in an attempt to detect regions that may be causing chipping. Next year we propose to integrate the data from recorded video into our baseline model of the track. Using this video, we will be able to more precisely explain the phenomenon we observe in the vibration data. 
 Furthermore, we plan to investigate two types of track issues pertinent to our system. First we will investigate the health of tracks which make up the switch gear. A damaged or worn out frog switch can have detrimental effect on the train, and several frog switches have had to be replaced in the Port Authority’s network over the last two years. We will analyze the data leading up to and following these replacements. Second, we will investigate where on the track the wheel rides on the flange. This occasionally occurs as the train passes through a curve, with the wheel on the outside of curve riding on the flange, as opposed to the tread of the wheel. Previous work has found particular features capable of detecting particular change rail changes, like a narrow band of energy which is indicative of rail squats [3]. Identifying these features for the proposed changes would be a significant contribution.
This year we propose to partner with a new organization, Analog Devices, Inc. They will provide us with new MEMS sensors and data acquisition systems that are more economical than our current sensors, but should have the capacity to detect the infrastructure changes of interest. If our deployment with their sensors is successful, we could more easily instrument additional vehicles, and draw on their expertise to ensure our system scales more effectively. Such a partnership with a private entity would complement our public partnership with the Port Authority.  
Our vehicle-based infrastructure monitoring technology has the potential to provide earlier indications of track and track structure anomalies, along with information about how the system ages. By expanding the number of defects we are able to detect and by leveraging data from more than one vehicle, we believe that our technique will be able to scale so that it becomes useful for numerous rail operators. This funding would allow us to further develop our technology, ensuring safer infrastructure through smarter maintenance. 

[1] J. Yang, A. Smyth, Y. Yang, D. Cavalcanti. “Road Surface Monitoring Via Multiple Sensor Equipped Vehicles.” IEEE INFOCON, May 26-June 1, 2015. Hong Kong, CN. 

[2] P. F. Westeon, C. S. Ling, C. Roberts, C. J. Goodman, P. Li, & R. M. Goodall, “Monitoring vertical track irregularity from in-service railway vehicles.” Proceedings of the institution of mechanical engineers, Part F: Journal of Rail and Rapid Transit, 221(1), 75-88 2007.

[3] M. Molodova, Z. Li, A. Nunez and R. Dollevoet, “Automatic Detection of Squats in Railway Infrastructure.” IEEE Transaction on Intelligent Transportation Systems. Vol 15, No 5, October 2014. 

Timeline

This project began with simulations in 2010 and laboratory experiments in 2011-2013. In September 2013, we instrumented the first train and began testing our work in the field. In 2014, we have completed important work on automating the transfer of data from the train into a database, and we have found features for detecting track and ballast changes when the location of the work is known. In 2015, we instrumented a second train, detected more subtle types of damage, and built simulations to understand which features are useful for analyzing our data. In 2016, we will work on fusing data from multiple trains, automatically detecting track changes, and localizing regions which may be causing chipping. 

Strategic Description / RD&T

Deployment Plan

We have already deployed our system on two light rail vehicles; over the next year will improve our analysis to offer the Port Authority the most pertinent information about the state of their network. 

Expected Outcomes/Impacts

In 2015 we demonstrated our ability to detect and localize changes on the track system. In 2016 we will improve on this capability and expand it to broader types of infrastructure changes. In addition, we will develop fusion techniques for multiple vehicles, a critical step for our technique to scale. Over the next calendar year, we will complete a journal paper on our novel instrumentation, and a second one to report on our ability to detect changes from our data. We will continue to incorporate improvements in the information we convey to the Port Authority, so they can directly benefit from our research. 

Expected Outputs

TRID

Individuals Involved

Email	Name	Affiliation	Role	Position
jbielak@cmu.edu	Bielak, Jacobo	CEE	PI	Faculty - Tenured
jelenak@cmu.edu	Kovacevic, Jelena	ECE	Co-PI	Faculty - Adjunct
lederman@cmu.edu	Lederman, George	CEE	Other	Student - PhD
noh@cmu.edu	Noh, Hae Young	CEE	Co-PI	Faculty - Adjunct

Budget

Amount of UTC Funds Awarded

$84819.00

Total Project Budget (from all funding sources)

$85000.00

Documents

Type	Name	Uploaded
Final Report	66_-_Final_Report.pdf	July 20, 2018, 4:27 a.m.
Publication	Dynamic responses, GPS positions and environmental conditions of two light rail vehicles in Pittsburgh	March 21, 2021, 7:44 p.m.
Publication	Detecting anomalies in longitudinal elevation of track geometry using train dynamic responses via a variational autoencoder	March 21, 2021, 7:45 p.m.

Match Sources

No match sources!

Partners

No partners!