Abstract: Aiming at the problems of large number of parameters, large size and poor anti-noise performance in traditional fault diagnosis models, a bearing fault diagnosis method based on MSDS-CNN( Multi-Scale Depth Separable Convolutional Neural Network) is proposed. The input signals are processed in parallel by using depth separable convolution of different scales to obtain multi-scale information and ensure the lightness of the model. The Dropout layer is introduced to improve the anti-jamming ability of the model, and the global average pooling layer is used to replace the full connection layer to reduce the number of model parameters. The experimental results show that the diagnostic accuracy of this method is up to 99. 6% , which is higher than other methods. The model has fewer parameters, smaller size and lighter weight. It also has good diagnostic accuracy under noise interference. 

Key words: depth separable convolution')">

depth separable convolution, multi-scale, fault diagnosis, lightweight, motor bearing