Fault detection, isolation, identification and recovery methods for perception systems in automated railway transport: A review

Significantly motivated by climate protection, not only Germany considers shifting the expanding transport sector towards more railway transport. The growing demand for train drivers stands in contrast to the demographic change, which causes a shortage of skilled workers. As a result, automatic train operation is gaining importance, because it offers perspectives regarding increased efficiency and safety. The publicly funded pilot project ‘AutomatedTrain’ aims to demonstrate unattended train operation during provisioning and stabling. To reach the highest grade of automation (GoA4), an extensive front monitor system for environmental perception is developed. Multiple LiDAR, RADAR, camera and ultrasonic sensors enable object detection, traffic guidance monitoring and distance estimation for continuous train control. Such sensors and related components underlie the risk of degradation and faults, often caused by environmental impacts like adverse weather, as seen in automotive applications. To avoid train control failures, a robust diagnostic system for fault detection, isolation, identification and recovery (FDIIR) is crucial.This publication first identifies typical perception system design approaches based on existing vehicle applications. Secondly, possible environmental impacts on sensors as well as resulting anomalies, faults and failure modes are determined. Thirdly, we present a systematic review of FDIIR methods with a view to future applications in public transport vehicles. The focus lies on operational reporting, in case of functional limitation, loss of redundancy or demand for maintenance. Unplanned corrective maintenance disrupts train operations, and planned preventive maintenance is inefficient. Thus, we emphasize predictive maintenance strategies towards fail-operational systems with increased availability, reliability and safety.