Subscribe to our newsletter to receive the latest news and events from TWI:

Subscribe >
Skip to content

Risk Assessment Model for Rail Track Inspection

Railway infrastructure maintenance is key to maintaining the reliability, availability and safety of rail transportation systems, and maintenance activities are of particular concern to asset operators due to the rapid increase in train traffic, speed and load carried.

The work summarised here is part of TWI’s remit in the OptRail project which is funded by Innovate UK (IUK) and involves a variety of stakeholders including rail operators.  TWI has combined data provided by Network Rail with several UK and international open data sets, to assess the likelihood of defective tracks for a given line or track segment, according to factors such as traffic, speed and population density.  We have subsequently identified and analysed the damage factors for rail tracks using data driven methods, calculated the weight of each factor in terms of their contribution to failure, and developed a quantitative risk assessment approach.

Overview

OptRail is targeted towards Network Rail as the end user due to their interest in a predictive model for automated risk analysis for rail tracks.  Risk-based inspection (RBI) is a popular risk assessment methodology as it can improve the overall safety of critical plant, whilst reducing the duration and costs of inspection or maintenance, by focusing on components identified as ‘high risk’.  The model has a number of key tasks as objectives.

Objectives

  • Utilise a combination of end-user historical data with open datasets, including i) a traffic density dataset obtained from the Office of Rail and Road (ORR); ii) a population density dataset obtained from the Greater London Authority (GLA); iii) a track, maximum speed dataset obtained from Open Street Map (openrailwaymap.org)
  • Assess the likelihood of defective tracks on each line per mile using Gaussian process regression method
  • Develop a methodology for calculating consequence of broken rails

Apply expert inputs and Monte Carlo analysis to accomplish the risk assessment model.

Figure 1. Flow chart of rail track risk assessment
Figure 1. Flow chart of rail track risk assessment
Figure 2. The prediction of the likelihood of defective tracks per mile
Figure 2. The prediction of the likelihood of defective tracks per mile

Solution

Figure 1 shows a flow chart of rail track risk assessment.  The risk of the system can be calculated by a combination of the likelihood of failure (LOF) and the consequence of failure (COF).

In this case, the LOF is calculated by the likelihood of defective tracks.  Historical data, including failure types and value of damage factors, is essential to establishing the RBI model.  In order to obtain the consequence of failure, some additional parameters such as cost of maintenance and route information are required as well.  The RBI model provides a risk matrix to inform inspection or maintenance regimes.

The techniques that are shown in Figure 1 are described below:

1. Likelihood of defective tracks by Gaussian process regression

Based on a review, a nonparametric kernel-based probabilistic model, namely Gaussian process regression (GPR), is applied due to its advantages of easy automation, accuracy, and less stringent data requirement.  In this technique, the training data is composed of four predictors and one response variable.  The predictors include average annual traffic density, average population density, average maximum speed allowed on the tracks and season.  Rail age and traffic load data should be included if available.  The response variable is the likelihood of defective tracks.  A Hold-one-out validation method is applied to verify the mode, and repeated several times until each row has been used for the validation set.  The true and predicted failure data are presented in Figure

2. Consequence of broken rails

The COF is another necessary input for risk assessment models. In this case, three categories of costs associated with rail defect inspection and the corresponding maintenance actions are considered – dealing with:

  • a very seriously broken rail that causes derailment
  • a broken rail that does not cause derailment;
  • a small rail defect that does not grow to a broken rail

3. Risk assessment model

The risk assessment function can be written as: Ri = Pi X Ci

where

Pi = the likelihood of defective tracks, unit: defect/mile (or yards);

Ci = the cost consequence of failure for rail tracks, unit: pounds/defect;

Ri = the risk caused by defective tracks, unit: pounds/mile.

The risk matrix for rail track inspection is presented in Figure 3.

Different risk represents different suggested inspection time and interval.  The specific suggestions need expert inputs to the model.

The model is able to develop a risk profile for each line and each segment, see Figure 4.

Based on the risk assessment, the model can also assess the remaining useful life and optimise inspection using machine learning methods, i.e., Gaussian process regression, support vector regression, in real time if condition monitoring data and commissioning time data are available, see Figure 5.

Conclusion

This case study applies Gaussian process regression to develop a quantitative risk assessment approach, based on selected damage factors for broken rails.  The model can predict the risk for each line and segment, and suggest optimum inspection intervals.  The remaining useful life can also be estimated in real time, and inspection periods arranged based on condition monitoring data and historic rail maintenance data.

Acknowledgement

OptRail is a collaboration between the following organisations: RCM2 Ltd, Brunel Innovation Centre, TWI Ltd, Yeltech Ltd and Surrey Advanced Control Ltd.

For further information on the OptRail project, please visit OPTRAIL – RISK-BASED-INSPECTION FOR OPTIMISED MAINTENANCE

Figure 3. Risk matrix for rail track inspection
Figure 3. Risk matrix for rail track inspection
Figure 4. An example of risk prediction on a line
Figure 4. An example of risk prediction on a line
Figure 5. Remaining useful life prediction
Figure 5. Remaining useful life prediction
Avatar Xiaoxia Liang Project Leader, Asset Life Management

Xiaoxia joined TWI in May 2018. Her expertise is in data analysis with a view to enhancing risk-based and condition-based assessments, and her work includes using machine-learning techniques in respect to rail track inspection. Xiaoxia worked with the Maintenance team of Royal Dutch Shell plc whilst undertaking her PhD at London South Bank University, developing a health indicator for large rotating machines to conduct real time and effective maintenance. She has MSc and BSc degrees in mechanical design.

}