Rolling bearing fault diagnosis method based on MTF-MMCNN

During the operation of bearing, complex and variable working conditions and insufficient samples will lead to low accuracy of bearing fault classification using traditional methods. To solve these problems, a mixed multi-scale convolutional neural network (MMCNN) fault diagnosis model combined with Markov transition field (MTF) method is proposed. Firstly, the MTF is utilised to convert the one-dimensional vibration signal of the bearing into a two-dimensional image with more sufficient time information. Then, the multi-scale branch module and the multi-scale cascade module are fused to fully extract the image features, a Convolutional Fusion Attention Module (CFAM) is constructed to strengthen the important features and assign appropriate weight to each channel, and a dropout layer is designed to make the model more stable. Finally, the images are input into the MTF-MMCNN to achieve the classification of bearing fault, and the performance of the model is verified by using the CWRU bearing dataset and the MFS bearing dataset in our laboratory. The results show that MTF-MMCNN has higher classification accuracy and stronger robustness under different working conditions, which is beneficial for formulating maintenance strategies.