Detecting Anomalies in Traction Motor Bearing Using Multi-scale Segmented Autoregressive Network

Traction motors significantly impact locomotive efficiency, and their bearings are pivotal components. Timely detection and identification of abnormalities in these bearings are essential for the normal operation of rail transit. In this paper, we present a novel approach for detecting anomalies within the bearings of traction motors in rail transit systems. Our method is underpinned by the analysis of data from vibration sensors to detect common irregularities. To meet the demands of real-time and accurate anomaly detection, we introduce an innovative Multi-Scale Segmented Autoregressive Network (MSSAN). Specifically, the MSSAN adopts a segmented autoregressive structure, with each segment using multi-scale feature analysis on vibration sensor data over a certain time range and passing the information to the next segment in an autoregressive manner. This design incorporates contextual information, thereby balancing inference speed and accuracy, ensuring the timeliness of anomaly detection. Using a well-established test bench, we conducted simulation experiments on the proposed MSSAN. The results demonstrate the effectiveness of our method in detecting and identifying anomalies occurring in the bearings of traction motors, indicating its practical utility for ensuring normal rail transit operation.