An Unsupervised Anomaly Detection Method Based on Density Peak Clustering for Rail Vehicle Door System
Wenzhong Shi, Ningyun Lu, Bin Jiang, Youran Zhi, Zhixing Xu · 2019
Frequent failure of the Rail Vehicle Door System (RVDS) has strong negative impact on the safe operation of rail vehicles. It is an urgent need to detect abnormal doors in their early stage of failures for effective condition-based maintenance. This paper proposes an unsupervised anomaly detection method for RVDS based on Density Peak Clustering (DPC) algorithm. The method relies on the real-time information of door's position, motor's speed and current collected from the multiple doors of a railway vehicle. Features related to the doors' health status can be extracted from these data, and then the DPC algorithm is used to identify the abnormal ones. Verification on North Extension of Guangzhou Metro Line 3 shows that this method can detect unhealthy doors accurately and provide a valid technique for the remote intelligent maintenance system for rail vehicles.