RMCA-DSCNN: a bearing fault diagnosis method based on residual mixed-channel attention mechanism

Abstract In the operation of industrial equipment, bearing fault diagnosis is of great significance to ensure the stability of the system and prevent serious accidents. However, the state change and noise interference during bearing operation will cause the fault signal features to be masked, which will affect the robustness and generalization of the deep learning model. The method combining attention mechanism and convolutional neural network has been widely used in the field of fault diagnosis, but its ability to extract key features still needs to be improved when dealing with complex bearing vibration signals. To solve the above problems, a new residual mixed-channel attention (RMCA) mechanism module is proposed. This attention mechanism can effectively alleviate the problem of gradient attenuation in the process of backpropagation, and significantly improve the problem of feature degradation and loss in the model. Combining RMCA with dual-scale convolutional neural networks (DSCNN), the optimal model structure of RMCA-DSCNN is obtained through a series of experimental verification. Through the visualization of the output weights by Grad-CAM, it is concluded that the RMCA module shows stronger discrimination ability in feature focusing. Finally, the proposed RMCA-DSCNN method is used to test various bearing signals. Experimental results on the CWRU dataset, XJTU dataset and PU dataset show that RMCA-DSCNN has an absolute advantage over other comparison network models.