Multisensor-Based Fault Detection Method for Air-Exchange Device in EMU Train

In this paper, we introduce and experimentally demonstrate a fault detection method for the air-exchange devices in the Electric Multiple Units (EMU) train, which utilizes the abnormal sound and vibration generated by the devices when the faults occur. The sound and vibration signals are fused, and the time-frequency matrix is extracted using a short-time Fourier transform (STFT). Fault recognition is performed using the trained Support Vector Machine (SVM) classifier. A sound detection system is built for experiments, in which the package of the sound sensor is designed to shield the sound from adjacent devices. The system includes an FPGA and an embedded system and can be used for fault detection in the future. The experiment shows that the accuracy of the fused signals is higher than the single sensor, up to 0.995. In addition, the performances of the algorithm are evaluated and the precision, recall, accuracy, and F1-score are all up to 0.99, which meets the actual fault detection requirements. Our method effectively improves the efficiency and accuracy of fault detection for the air-exchange device and can be widely used in the EMU train.