High-Speed Railway Bogie Fault Diagnosis Using LSTM Neural Network

Abstract

The bogie is a key component of the high-speed train, and its deterioration and damage will seriously threaten the safe operation. The timely fault detection and diagnosis of the bogie are crucial to the safety and comfort of the train. Multiple sensors are installed at different positions in the bogie, and vibration data containing rich fault information are obtained. Because the vibration signal is a complex and highly uncertain nonlinear signal, the traditional signal analysis method cannot effectively excavate the sensitive characteristics in the data. This paper presents a fault diagnosis method based on long-short-term memory (LSTM) recurrent neural network. LSTM can solve the problem of long-term dependence, thus the signal does not need to be preprocessed. This network can automatically extract effective features from multiple channels and allow to make the best of multiple channels by automatic combination. Experiments conducted on the dataset based on SIMPACK simulations have verified that the LSTM network can learn these fault features from the data. Fault classification accuracy of the network is achieved to 96.6%.