Fault Diagnosis Method Based on Convolutional Attention Mecha-nism and ResNet34

Abstract

To solve the problem of network degradation caused by blind stacking of convolutional layers, and to further improve the feature extraction ability of deep learning models for gear small sample fault signals, a fault diagnosis method based on LW-DCGAN and C-ResNet34 models is proposed. Firstly, the LW-DCGAN model is used to enhance the PS-SDP image data set. Then, the convolutional attention mechanism is introduced into the residual block of the ResNet34 model, and the C-ResNet34 model is constructed to enhance the ability of fault feature extraction in PS-SDP images. Finally, experiments prove the effectiveness of the LW-DCGAN method under the C-ResNet34 model, and compared with other deep learning models, the C-ResNet34 model can more effectively extract the features of different fault data and improve the fault diagnosis accuracy.