From battery-sensor fault detection to automatic doors, predictive AI deployed across MI09 and MP05 trainsets
RATP: 10+ AI-powered predictive maintenance solutions
The Challenge
On RATP's RER A (MI09) and metro (MP05) trainsets, recurring malfunctions were causing frequent delays amid continuously rising ridership. The goal: anticipate failures rather than endure them, across critical components (sensors, doors, compressors, pantographs, safety and passenger information systems), while optimizing costs and the passenger experience.
The Solution
Talan developed more than 10 AI-powered predictive maintenance solutions for RATP, covering the full rolling-stock lifecycle: battery temperature sensor failure detection, speed sensor impedance analysis, air compressor failure prediction, pantograph monitoring, acoustic anomaly detection, prognostics for passenger information, braking/traction and the onboard IT system (SIE), and the SACEM safety system, plus a passenger load analysis tool to optimize the scheduling of long and short trainsets. Each solution relies on tailored models (statistical filters, Isolation Forest, Random Forest, One-Class SVM, Compact Prediction Trees) feeding reporting and alerting systems for the maintenance teams.
The Impact
Validated solutions reach high reliability levels: 92% on battery sensor fault detection, 83% recall on air compressors, 93% F1-score on the SACEM predictive model, and more than 95% on the onboard IT system (SIE). Several solutions are already in production, used daily by maintenance teams through reporting and alerting systems updated every 50 minutes, while others are being industrialized.