This project tested signalling data to help identify overspeed events
Project number:
COF-UOH-85
Using signalling data to identify train overspeed events
'The AI Change Toolkit is a fundamental step forward concerning governance in a safety critical industry.'
The challenge
Overspeed can lead to major accidents. For example, overspeeding led to the 2013 high-speed train derailment in Santiago de Compostela (Spain) which resulted in 79 fatalities, and the 2016 Sandilands (UK) tram derailment, resulting in seven fatalities. Rail Accident Investigation Branch reports have shown several serious overspeed events.
The European Train Control System (ETCS) continuously monitors and controls train speed. However, it will not be widely used on the GB rail network for some time. Right now, the industry mostly bases overspeeding on:
- Train Protection and Warning System (TPWS) overspeed trips at signals and speed restrictions
- self-reporting
- manual checks on on-train data recorder (OTDR) traces as part of driver competence management.
We need a more comprehensive approach.
What we did
We developed a method for identifying overspeed events. It uses Train Describer data to find where average train speeds exceed line-speed limits. This works well for line sections that see steady train speeds.
Where line-speed varies, a travel time analysis can be used to spot unusual events. These can indicate instances of overspeed.
Machine learning clustering was used to find outliers due to speed restrictions. The models used travel time changes to predict where these were overspeed events.
The research was delivered through the RSSB and University of Huddersfield strategic partnership.
Benefits delivered
This study showed that we can predict overspeed events using existing signalling data.
We will use this as the basis to develop an online demonstrator for a specific line or route.
This will show whether a real-time tool could improve operational safety and efficiency across the network.