224 / 2025-06-14 22:45:36

Intelligent fault diagnosis for train bogie based on multi-scale residual attention

Bogie fault diagnosis,Deep learning,Multi-scale feature fusion,Attention mechanism

As the key bearing part of the train, the health state of the high-speed train bogie is directly related to the traffic safety. In order to deal with the problems of relying on manual experience and insufficient accuracy in previous methods, this paper proposes a bogie fault diagnosis model (MSRA-Net) combining multi-scale convolution, residual connection and channel attention mechanism. Therein, multi-scale convolution is used to extract features at different time scales, the channel attention mechanism (SE Attention) is used to enhance the fault feature expression, and cross-scale feature fusion is used to realize multi-scale information fusion. An actual case of train bogie bearing shows that the proposed method achieves accurate identification on four types of typical faults, and the recognition effect on small sample data sets is significantly better than that of traditional methods.