Improve the operational conditions of infrastructure and rolling stock and anticipate the flow of people transported to trains. These are two examples of the deployment of AI at Getlink. “Things are changing. We are facing a change in the way we work on a daily basis, warns Yann Leriche, general manager of Getlink (3,600 employees and 1.829 billion euros in turnover in 2023). But it is also a school of humility. PSometimes we have strong beliefs and it doesn’t work for the purpose we envisioned.”
The former Eurotunnel group, which has two main activities – infrastructure manager with the Channel Tunnel as well as two terminals (Coquelles and Folkestone) and Shuttle operator – is carrying out several projects around artificial intelligence. Its objective: to double the number of high-speed cross-Channel links in the next ten years. It targets concrete use cases for AI such as digital twins to improve traffic flow. They allow the modeling of the flow of people transported to the trains, the counting of the number of people per vehicle, the recognition of plates and even intelligent signage. Another use case: the availability of the tunnel with imaging analysis in order to be able to constantly observe the tunnel, analyze disparate signals and assist operators in their decision-making.
50% reduction in incidents related to applied brakes
The other crucial area is train availability. “It is the busiest rail highway in the world and, with Brexit, has become Europe's leading land border for freight and passengers” recalls Yann Leriche. The Eurostar transports 11 million passengers per year and the Shuttle, 1.4 million trucks (35% market share) and 2 million cars (50% market share). It is impossible in these conditions to have significant downtime. «The Shuttles are 800 meters long, the length of two TGVs placed end to end, recalls Denis Coutrot, head of data and artificial intelligence. We need to make locomotives communicating to collect data more quickly, detect anomalies and carry out engine diagnostics.”
For rolling stock, AI must make it possible to anticipate failures and optimize preventive maintenance plans. There are a dozen use cases to anticipate the failure of rolling stock and optimize a preventive maintenance plan, but also anticipate the repair of reusable parts.
A significant example is the phenomenon of applied brakes which damage wheels and rails. It is an abnormal heating of the brake or wheel band caused by an applied brake or a locked axle. The development of digital twins of bogies makes it possible to model the behavior of rail-road contact. “Our Shuttle trains go back and forth and we take measurements every hour when they arrive at the terminal, explains Denis Coutrot. With information from sensors placed on the rails, the algorithms will trigger alerts when the temperature rise is abnormal. In this case, this brake is neutralized until the next maintenance. » This isolates the brake, but does not prevent the train from continuing its journey by eliminating the risk of it bottoming on the wheel. This application was implemented at the beginning of the year, which has already made it possible to reduce the phenomenon of applied brakes by 50%.
Missing the target
Getlink also deploys a tool based on generative AI with its documentary corpus for searching information and documents based on a closed ChatGPT. “Guys don't use it, notes Denis Coutrot. They have all the regulations in mind and their binders. So this tool is not aimed at the right people. We put it on the intranet to open the floodgates. It will be operational in the 1st quarter of 2025”. An observation that Yann Leriche completes. “We are not sure that even in ten years these technicians will abandon the filing cabinet”indicates the latter. This application is more dedicated to non-regulation specialists. The time saving, in this case, is enormous since it avoids the multiplication of meetings and questions.
