LSTM-based Anomaly Detection for Railway Vehicle Air-conditioning Unit using Monitoring Data
Toshihide Yokouchi, Minoru Kondo · 2021
Railway vehicles have a lot of equipment, and their faults possibly have significant impacts on reliability and safety of railway operations. Some vehicles these days have Train Control and Monitoring System (TCMS), and constantly monitor and collect data of the equipment conditions during operations. In this paper, utilizing the monitoring data, we propose a method to continuously evaluate abnormalities of vehicle equipment using Neural Network with Long Short Term Memory (LSTM), which effectively and flexibly learn time series data. In the method, to realize anomaly detection, we introduce "anomaly score", which is an indicator to express the degree of abnormality of vehicle equipment. In this paper, we show that the anomaly score increases as a fault of an air-conditioning unit progresses from 1.5 month before its fault became apparent. Consequently, the reliability of the railway operation will be improved by evaluating abnormalities of vehicle air-conditioning unit or identifying signs of faults from monitoring data recorded with TCMS.