AI and machine learning applications in rail
Artificial intelligence (AI) and machine learning (ML) are analytical methods increasingly applied to maintenance, infrastructure inspection, and safety functions in European rail.
ML adds most value where the relevant patterns span many variables and enough recorded failures exist to train on. For rail, that puts wheelset and bearing monitoring and infrastructure inspection among the areas where deployment is furthest along.
Condition monitoring and dispatching support
Condition monitoring is the most widely deployed form of this. ML models analyse vibration, temperature, acoustic, and current data from onboard and wayside sensors, identifying bearing degradation, wheel defects, and pantograph wear ahead of scheduled inspection dates.
DB InfraGO is piloting ADA-PMB (Automatische Dispositionsassistenz auf Basis Produktionsmodell Betrieb), an automated dispatching assistant that proposes conflict-resolution options to human dispatchers; automatic conflict detection between train movements is already established practice, and the new element is the resolution proposal itself. As of late 2025, the system was being rolled out for integration directly into dispatchers’ screens, targeted for December 2026.
Automation and energy-optimised driving
Automatic Train Operation (ATO) at Grade of Automation 2 (GoA2) automates traction and braking commands within the parameters set by signalling and timetable, and is the automation level being rolled out on European mainline networks; like DAS below, it relies on deterministic control logic rather than machine learning. Driverless operation with an onboard attendant (GoA3) exists in European metro and urban transit systems, not on mainline rail.
Energy-optimised driving is a mature application area in European rail, though the best-known systems — SBB’s Adaptive Control (ADL, Adaptive Lenkung) and comparable Driver Advisory Systems (DAS) and connected DAS (C-DAS) used elsewhere in Europe — rely on deterministic optimisation rather than machine learning; SBB reports around 70 gigawatt hours in annual savings from ADL.
Infrastructure inspection
In infrastructure inspection, computer vision applied to train-mounted and lineside cameras identifies surface defects, vegetation intrusion, and geometry anomalies along the routes they cover.
Hitachi Rail’s acquisition of Omnicom — a rail monitoring technology company previously owned by Balfour Beatty — was announced in January 2025 and completed in August 2025, integrating Omnicom’s train-mounted track inspection systems, which use edge computing and machine learning for near real-time anomaly detection, into Hitachi’s HMAX digital asset management platform.
Regulatory constraints
AI applications in safety-critical contexts — maintenance decision support, ATO, anomaly detection for signalling — must be validated under the frameworks that already govern the systems they feed into.
Rail-related AI systems generally fall under the EU AI Act (Regulation (EU) 2024/1689) via Annex I, through their link to Directive (EU) 2016/797 on rail interoperability, rather than under the Annex III use-case list. For these Annex I systems, only a limited set of AI Act articles applies directly; the substantive conformity assessment, documentation, and oversight requirements are instead integrated into the existing rail interoperability framework rather than imposed as a separate parallel regime.
The “Digital Omnibus on AI” simplification package was published as Regulation (EU) 2026/1744 in the Official Journal on 24 July 2026 and entered into force on 27 July 2026, deferring high-risk compliance dates — to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems such as those covering rail