Towards the automation of anomaly detection and integrated fault identification for railway switches in a real operational environment

Abstract

We report on the latest developments regarding automatic anomaly detection and fault diagnosis for railway switches achieved in the context of the EU Shift2Rail project In2Smart2. This paper presents improvements regarding the interpretability of the output of a data-driven anomaly detection routine, thereby increasing its usability for maintenance engineers. Warning messages are generated by a new module in case certain critical engine failure types are detected or additional information on detected anomalies is available. Further developments in the diagnostic model brought it to a state in which it is ready for a first plausibility check under supervision of maintenance experts. Taken together, these developments are an important step towards the integration of anomaly detection and diagnosis into a comprehensive condition monitoring system. The advances have recently been or are projected to be implemented in an upgrade of the workflow running in near real-time at the maintenance control center of Strukton Rail, in the Netherlands. This paper gives an overview of the workflow which includes the new warning module, the extended anomaly detection pipeline, and how the diagnostic model is to be conceptually embedded, i.e. its interactions with the other modules. Moreover, methodological improvements, e.g., in data preprocessing for feature extraction and warning generation, are also described.