Bearing Fault Diagnosis Based on Pooling Weight Multi-scale Convolutional Neural Networks

The variable operating conditions of bearings bring challenges to the application of traditional bearing fault diagnosis methods, so it is necessary to explore more intelligent fault diagnosis methods. A novel fault diagnosis method based on Pooling Weight Multi-scale Convolutional Neural Networks (PWMCNN) is proposed in this paper to solve this problem. Firstly, for the problem that traditional CNN cannot extract multi-scale features, an improved multi-scale module is designed, which uses convolutional kernels of different sizes to extract features of bearing signals. Secondly, to improve the fault diagnosis accuracy under variable working conditions, the feature weights of each channel in the pooling layer are calculated to highlight the important features of each channel in the pooling layer that are more sensitive to variable working conditions. Finally, the performance of the PWMCNN model is compared with classical machine learning and deep learning methods. The results show that the proposed method has better average diagnosis accuracy in terms of single load, variable operating conditions.