A Rotating Machinery Fault Diagnosis Method for High-Speed Trains Based on Improved Deep Learning Network

Rotating machinery is an important part of highspeed train system. Any failure of such components will have a serious impact on the service itself and even jeopardize the safe and reliable operation of the train. However, the limited diagnostic accuracy has become the bottleneck of fault diagnosis. In this paper, an adaptive fault diagnosis method based on deep sparse autoencoder (SAE) network is proposed. Firstly, the original time domain signals collected by sensor are preprocessed and used as input to the diagnostic network. Secondly, the SAE fault diagnosis network model is built. Thirdly, the loss function of SAE is modified to improve the diagnostic accuracy and efficiency of network. Finally, experiments verify the feasibility and advantages of the proposed method.