Anomaly Detection for Railway Vehicle Equipment Using Condition Monitoring Data

Abstract

In recent years, some railway vehicles have been equipped with condition monitoring devices, which constantly record the operating condition of railway vehicle equipment. For more effective use of condition monitoring devices, we propose an anomaly detection method for railway vehicle equipment using Long Short-Term Memory (LSTM), which is a deep learning method suitable for learning time-series data. In this paper, we apply the proposed method to data on engines and air-conditioning units recorded on vehicles in operations. Results confirmed that the anomaly score for anomalous data increases by using the proposed method, and that anomalies are detected in railway vehicle equipment before faults appear.