An adaptive fault detection framework based on graph attention network for locomotive diesel engines

Locomotive have been equipped with status monitoring devices that continuously record the operation status of locomotive. These devices use sensor data to detect the abnormal state of the locomotive, which is fundamental and critical for the locomotive operation. Sensor data is recorded in time series. In this study, a adaptive fault detection framework based on Graph Attention Network (GAT) is proposed, which is applicable to fault detection based on deep learning methods for the status monitoring data. The proposed method is applied to data on diesel engines recorded in operations. The results demonstrate that the proposed method enhances the anomaly score for anomalous data and enables early detection of anomalies in locomotive diesel engines prior to fault occurrence.