Reducing low adhesion problems by using AI
Moisture on the rail mixing with contamination such as 'leaves on the line,’ rust, or grease can cause low adhesion between the rail and wheel. This happens throughout the year but especially in autumn.
Low adhesion can often disrupt passenger journeys. It can also cause safety risks, such as signals passed at danger and station overruns. Stopping trains reliably and predictably in a range of conditions is key to the safe running of a busy railway.
We know AI can be used to provide a quantitative estimate of friction between the wheel and rail. In this project we tested the approach using real-world conditions, including images captured onboard trains
The University of Sheffield and RSSB improved an existing model and tested it with relevant data. An online demonstrator was also made available for industry users.
The model, delivered as part of an earlier project, was re-trained with data representing a range of low adhesion conditions. Bespoke hardware was developed to capture images and environmental data from passenger trains.
To test the tool, industry users were given access to an online demonstrator and guidance on how to use it. This meant Mobile Operations Managers, who visit sites to assess low adhesion conditions, could generate friction estimates in situ.
This project showed the AI tool can help staff make evidence-based decisions about low adhesion. This helps to reduce costs of around £355m each autumn.
In the future, the tool could be incorporated into an app. This could improve the process for capturing data and generating estimates. It would mean that tactical decisions could be taken based on more objective and granular data.
Alternatively, train-borne equipment could be used to capture low adhesion information at a line or route level. This could reveal long-term trends that can enhance autumn planning.