Eiffage Énergie Systèmes has created an AI solution to optimize predictive maintenance for railway infrastructure

BPL-HSL, which opened in 2017, connects Paris to Rennes in one hour and 25 minutes. This 182 km line is maintained by Ferlioz teams, who ensure the smooth running of 30,000 trains per year (with a punctuality rate of 99.2%). The infrastructure's performance and safety depend largely on rigorous maintenance. “Managers carry out conditional maintenance on infrastructure by regularly checking track geometry parameters, such as deformation, lateral alignment, cross levelling, and rail spacing. However, given the challenges of availability, cost and safety, predictive maintenance offers an innovative approach using AI to anticipate failures and optimize operations," summarized Digital Expertise Activities Manager Jean-Louis Haller.

Predictive maintenance is based on the cross-analysis of data collected from various sources: sensors embedded in SNCF Réseau's IRIS trains, maintenance data from Ferlioz's CMMS, and geometric track design and traffic data (number, speeds, tonnages).

Using a hybrid approach combining machine learning and deep learning, AI can then identify anomalies that could point to a future failure, thereby optimizing maintenance operations.

"To achieve a more robust prediction, our methodology is based on a three-step approach: clustering, which involves grouping track segments into families; mixed-effects models, which combine general and specific trends; and K-Nearest Neighbors (KNN), or learning from past operations to predict future effects," explained Jean-Louis Haller. The predictive model, currently in its test phase, has an average error rate of 4% at one month and 13% at twelve months.

Instrumentation and data collection
At the same time, our Group experts are studying the mechanical behaviour of track structures using sophisticated instrumentation comprising more than a hundred sensors installed on four representative sections. This meticulous data collection enables parameters such as temperature, humidity, deformation and acceleration to be measured. "This instrumentation along the tracks installed by Eiffage Infrastructures on the BPL-HSL allowed them to establish that a gravel-asphalt sub-ballast layer offers superior performance to a granular sub-base (increased protection against water infiltration, reduced vertical acceleration, and deformation stability). These data are being further analyzed through a partnership between Eiffage Infrastructures, Eiffage Concessions, Gustave Eiffel University, and Eiffage Énergie Systèmes, which is bringing its AI expertise to the project," said Jean-Louis Haller.

Driven by AI and human expertise, predictive maintenance is revolutionizing railway infrastructure management. On the BPL-HSL, it helps anticipate breakdowns, optimize costs, and ensure maximum safety. At a time when sustainable mobility is becoming a major issue, these innovative technologies are opening up new possibilities for other rail infrastructures and even other types of infrastructure.

www.eiffage.com