Condition Monitoring of Railway Infrastructure a Case Study of a Light Rail System

doi:10.21203/rs.3.rs-4502922/v1

Download PDF

Case Report

Condition Monitoring of Railway Infrastructure a Case Study of a Light Rail System

https://doi.org/10.21203/rs.3.rs-4502922/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Railway infrastructure includes complex, integrated, multi-disciplinary systems requiring regular inspections and maintenance. Track and Overhead Line Equipment (OLE) are the main railway infrastructure systems. These systems deteriorate over time due to operations and environmental factors.

Monitoring technologies are used to replace manual measurements, to gather data relating to the condition of railway infrastructure assets. Engineers and experts review and make decisions. There is a collective interest to replace the current reactive and preventative maintenance with Predictive Maintenance (PdM) which means maintaining or renewing assets before they fail. This is achieved by condition monitoring of assets.

The research data is collected using a Remote Condition Monitoring (RCM) system mounted on a tram on the UK’s biggest light rail network. Data is processed by analytical software. The research aims to introduce Artificial Intelligence (AI), specifically a Machine learning (ML) algorithm, to inform PdM of railway infrastructure focusing on OLE and Track systems.

Condition Monitoring

Maintenance

Big Data

Railway

Asset degradation is a significant problem in the railway industry. Asset degradation leads to the unavailability of assets and related services if not monitored and addressed. To proactively address this problem, the concept of Preventative Maintenance (PM) came about, where the focus is to inspect and monitor the asset condition by gathering data periodically and using it to predict the life of assets and maintain them accordingly. This creates the need for equipment and methods to collect reliable data regularly and remotely, making this process informed and an opportunity to make this more intelligent by applying various data processing techniques. (Preprint, 2015) “Rail operators spend more than 80,000 euros on maintenance for every kilometre of the route”, Victor Borges calculates, “That’s a total of 15 to 25 billion euros a year. Imagine if they could save just one per cent of that a year? That would represent between 150 and 250 million euros of savings annually!” (Borges, 2019).

Network Rail (NR) is a major rail operator, maintainer and railway asset owner in the UK. It has awarded ESG Rail and DB System Technik (DBST) a contract to supply overhead line monitoring technology to increase the capability of its existing monitoring fleet. They introduced logic for collecting monitoring data using in-service trains instead of dedicated services for data collection. It reduces operational and planning challenges because no special arrangements are needed to accommodate the specialist trains running the already busy network (Burroughs David, 2018). It shows a genuine interest in the change and new data collection methods.

Corrective maintenance (CM), or reactive maintenance, is undertaken after a defect or failure occurs. This strategy leads to high maintenance costs due to sudden failure and system recovery; the Preventive maintenance (PM) strategy involves performing maintenance activities before equipment failure. PM can occur during the system downtime or while the system operates. The most significant advantage of PM is that it can be planned and performed when convenient (Budai, Dekker, & Nicolai, 2008). In Condition-Based Maintenance (CBM), the main objective is to optimise maintenance activities by estimating the component’s actual status using monitoring and inspection techniques. This results in discovering those components where maintenance is required to reduce the maintenance cost. Finally, predictive maintenance leverages the prediction of the failure time to proactively schedule maintenance activities, which could be an integral part of CBM since CBM involves more or less certain types of prediction.

Traditionally data collection on railway infrastructure has been through manual surveys and inspections. Still, the railway industry is moving towards automatic data collection methods due to the advancement, availability and relatively low cost of various sensors and data collectors. The industry is investing in this equipment and technologies. Data is collected by installing bespoke equipment with sensors, receivers, processors and storage devices. Typically, the data collection equipment remotely collects data without human intervention, which can be later processed. It creates an opportunity for further studies and research to make the output data more meaningful and can also be used to predict the behaviour and condition of assets. It removes or reduces the human element in the process, making it efficient and reducing the risk on site. The reporting is also much quicker (D'Agostino Antonio, 2016).

3.1. Case Study

Manchester Metrolink is the UK’s biggest light rail network serving the Greater Manchester area. In 2017 (KAM) Keolis Amey Metrolink (Current Operator and Maintainer of the Manchester Metrolink) was awarded a 10-year contract to operate and maintain the network. TFGM (Travel for Greater Manchester) contractual requirements included delivery of better asset condition information systems. The existing systems were manual or used aiding equipment that still required significant manual resources at the time of data collection. Various condition monitoring systems aimed at different asset groups were selected. (Pennington & Evstafyeva, 2020). The Selectra Vision(SelectraVision, computer vision system for measuring, inspection and diagnostics.) RCM (Remote Condition Monitoring) systems were selected to measure some condition parameters for Track (Track-V2 System) and OLE (CAT-VW System). This research focuses on these two systems.

3.2. OLE Measuring System (CAT-VW)

This contactless optical system is installed on the roof of the vehicle and measures OLE contact wire parameters such as wire height, stagger and wear using LED light and high-speed cameras at normal operational speeds. The system works by processing the reflected light from the underside of the wire. The light is projected on the wires from the unit by powerful LED lights. The reflected light from the area that comes in contact with the pantograph is brighter than the out-of-running profile of the wire and other OLE components. The in-contact site is more polished and shinier due to the constant sliding and rubbing with the pantograph carbon.

Composition of the System:

The system is composed of the following elements:

Height/stagger/wear measuring box with sensors and receivers for the wire parameters measurement.
GPS antenna. (Shared with the Track measuring system)
Camera for OHL and pantograph video inspection
Camera for a panoramic view
Thermal camera
Compensation sensors to compensate for the vehicle rolling.

Recorded parameters:

The recording software records the following parameters:

Wear contact wires
Height and stagger of contact wires
Pictures of wires.
GPS position.
Vehicle speed.
Video of pantograph and overhead line

3.3. Track Monitoring System

The track measuring and inspection system is contactless and measures the track geometry and rail wear/profile. This system works on the principle that a laser blade creates the shape of the Track. The laser blade follows the Track’s surface while drawing the Track’s section with a beam of laser light while the measuring vehicle runs. High-speed digital cameras capture all the images drawn by the laser blade on the Track. The software processes all acquired images and gives the track profiles in all measured locations.

The main subsystems include,

GPS antenna to track the location of the captured data.
A camera for a panoramic view.
Odometer to track the metreage of the data captured for linear measurements.
Cameras for video of tracks.

3.4. Planning Data Collection and Retrieval

The data collection run involved planning the Light Rail Vehicle (LRV) out of service. This was requested through the LRV service planner. Next was to book a dedicated driver on a shift to go on a selected route for data collection. The LRV needed to be prepared for the data collection run on the day. This involved removing covers from the track system under the tram and wiping the glass windows on the casing that houses the laser. The covers for the OLE system open automatically once the system is turned on but still need cleaning from the accumulation of dust and carbon residue. The dome camera and infra-red camera also need to be wiped clean. All this preparation is done in one of the depots on a workshop bay with roof access and side pits.

Once the tram is prepared, operation control is informed and asked for authorisation to go on a selected route. Since the RCM tram runs between normal service trams during the day, priority is given to service trams carrying passengers. The equipment is only capable of recording measurements for one system at a time, i.e., OLE or Track. Hence, two runs are required on the same route to get both Track and OLE data. Data is downloaded off the onboard hard drive onto a portable hard drive for processing and data analysis. This data in raw form (CSV format) would be used for the proposed IMS.

The RCM system comprising the OLE and Track measuring subsystem can only measure one system, i.e., OLE or Track; hence, multiple runs are required to record data for each system. Table 1 (for OLE) and Table 2 (for Track) show a sample of data collected using the RCM equipment. Position and the GPS coordinates in the form of easting and northing are the common features of the two data sets. These features will be later used to integrate the two data sets to explore interdependence. The data is presented at every meter. The system resets to zero whenever the system is turned off. This makes the position feature only applicable to determine the distance covered in that particular run. Easting and northing, on the other hand, could be imported into Google Maps to relate the data to infrastructure along the route. These coordinates have an accuracy of up to 3 meters. The last two columns are user-introduced parameters where the middle of each station platform and OLE poles will be marked using their coordinates. The latter is important because the height and stagger values change by design at different poles depending upon the location requirement.

Table 1 shows the OLE data with CW (contact wire) height and stagger measured for up to two wires and diameter for up to four wires. Wear is measured every meter but presented as minimum or average over a meter length of contact wire. The minimum values are used to identify spot wear, i.e., localised wear of the wire due to an unusual hard spot or damage. The average wear value shows the usual wear of the wire over time. This value is used for inspection and maintenance purposes.

Table 2 shows track parameters. The system identifies the rail type by comparing it with the rail profiles. Gauge is the separation between the rails and is a controlled feature for the track. Cant is the difference of height between two rails. Twist is the cross-level variation between rails over a length of 2 meters. Twist faults can result in the derailment of the train; hence is a significant feature to control. Track curvature and the vibration given as vertical and transverse acceleration are helpful but not critical parameters. They affect the ride quality. Rail wear given as top, 45 degrees, and side are critical parameters that can lead to gauge widening and poor wheel-to-rail interface that can cause unwanted noise, rail fracture and derailment.

4.1. Data Analysis

The research focuses on the main parameters for both OLE and Track. These are the ones required to be maintained by the installation and maintenance manuals for railway tracks and OLE. It is important to note that OLE is based on the track design and generally follows the centerline of the Track.

4.2. Research Gap Analysis and Critical Review

(Tang et al., 2022) reviewed 141 scientific papers between 2010 and 2020. However, there were only 81 papers relating to the maintenance and inspection of railways. (C. J. Cho & H. Ko, 2015) studied the relationship between dynamic stagger and its effect on wear for trains travelling through a curve. The research focuses only on curves and uses images, not measured data. The paper has an inaccurate description of the contact wire in Fig. 1 (ibid., p.1), where the wire is shown outside the limits of the Track. It never happens as the zigzag profile of the wire is created by the curved track geometry. (Han et al., 2020) also looks at visual inspection methods to detect fractures in catenary systems clevises. Again, it does not cover the main maintenance parameters for OLE. (W. Sammouri et al., 2014) presented pattern recognition approach using floating train data, the paper briefly touches on monitoring railway infrastructure. Still, the main focus stays on rolling stock, specifically on “Tilting Pantograph Failure” The work mentions track defects and track inspection using probe vehicles. Still, there is no mention of OLE defects or inspection.

Most of the research around railway infrastructure and maintenance uses image processing techniques, e.g., (X. Gibert, V. M. Patel, & R. Chellappa, 2015) applied computer vision techniques to inspect missing or broken parts of the railway track. (Tang et al., 2022) found that researchers mainly focused on track elements like track geometry, track deterioration and rail defects. At the same time, some studies included rolling stock as well in the category of maintenance planning, and papers incorporated AI applications focusing on the dynamic scheduling of maintenance activities by applying techniques such as expert decision rules.

The RCM equipment selected for the Metrolink network comes with its benefits and some shortcomings. The system automates data collection for some parameters very well, such as CW wear (for OLE) and rail wear (for Track). This capability reduces a huge amount of manual work. The video footage proved to be useful in some failure investigations. Data collection and recovery have become fast and multiple re-runs are now possible with reduced impact on possession times where parts of the network are unavailable for service.

The equipment does not record reference points, i.e., pole locations, automatically. These needed to be added manually to the software along with other significant points like station locations and track curves, which created a significant challenge in matching the collected data to the infrastructure. The Track system does not recognise rail types with a high accuracy, as a result, data cannot be evaluated in those sections. The OLE system cannot record data in tunnels and under bridges due to reflections from the tunnel roof and bridge soffits.

The collected data from these RCM systems still require a considerable amount of work and expensive engineering resources to identify, report and plan remedial work identified as a result of the condition monitoring system. Artificial Intelligence (AI) models will be used to bridge the gap between the current output and replace some of the human intelligence (HI). For height and staggers, historical data, including design and previously recorded values during an inspection, will be merged and compared with the RCM data. Intelligent decision-making algorithms will be applied to reduce manual data analysis and create autonomous decision-making capability.

Borges, V. (2019). How ‘predict and prevent’ equipment failure is now keeping trains on track. Retrieved from https://www.thalesgroup.com/en/worldwide/transport/magazine/how-predict-and-prevent-equipment-failure-now-keeping-trains-track
Budai, G., Dekker, R., & Nicolai, R. P. (2008) Maintenance and production: A review of planning models. Berlin ;: Springer.
Burroughs David. (2018). Network rail to test overhead line monitoring technology. International Railway Journal, Retrieved from https://www.railjournal.com/infrastructure/network-rail-to-test-overhead-line-monitoring-technology/
C. J. Cho, & H. Ko. (2015) Video-based dynamic stagger measurement of railway overhead power lines using rotation-invariant feature matching
D'Agostino Antonio, A. J., Carr Christopher. (2016). Big data in railways. ().
Han, Y., Liu, Z., Lyu, Y., Liu, K., Li, C., & Zhang, W. (2020). Deep learning-based visual ensemble method for high-speed railway catenary clevis fracture detection. Neurocomputing, 396, 556-568.
Pennington, C., & Evstafyeva, O. (2020, -01-14T12:17:09+00:00). Towards ‘intelligent infrastructure’. The International Light Rail Magazine, Retrieved from http://www.tautonline.com/towards-intelligent-infrastructure/
Preprint, V. F. A., Wickern Meyer. (2015). Challenges and reliability of predictive maintenance. Unpublished manuscript.
SelectraVision, computer vision system for measuring, inspection and diagnostics. Retrieved from http://www.selectravision.com/index.php
Tang, R., De Donato, L., Bes̆inović, N., Flammini, F., Goverde, R. M. P., Lin, Z., . . . Wang, Z. (2022). A literature review of artificial intelligence applications in railway systems. Transportation Research Part C: Emerging Technologies, 140, 103679.
W. Sammouri, E. Côme, L. Oukhellou, P. Aknin, & C. -E. Fonlladosa. (2014). (2014). Pattern recognition approach for the prediction of infrequent target events in floating train data sequences within a predictive maintenance framework. Paper presented at the - 17th International IEEE Conference on Intelligent Transportation Systems (ITSC), 918-923. doi:10.1109/ITSC.2014.6957806
X. Gibert, V. M. Patel, & R. Chellappa. (2015). (2015). Robust fastener detection for autonomous visual railway track inspection. Paper presented at the - 2015 IEEE Winter Conference on Applications of Computer Vision, 694-701. doi:10.1109/WACV.2015.98

Tables are available in the Supplementary Files section.

Tables.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Condition Monitoring of Railway Infrastructure a Case Study of a Light Rail System

Status:

Version 1

Abstract

Figures

Introduction

Methodology

3.1. Case Study

3.2. OLE Measuring System (CAT-VW)

3.3. Track Monitoring System

3.4. Planning Data Collection and Retrieval

Results and Discussion

4.1. Data Analysis

4.2. Research Gap Analysis and Critical Review

Conclusions and Future Work

References

Tables

Supplementary Files

Status:

Version 1