Fault diagnosis method for rolling bearings based on CWT-IDenseNet

Aiming at the problems of incomplete information contained in one-dimensional signals and overfitting of the DenseNet under variable working conditions, a rolling bearing fault diagnosis method based on continuous wavelet transform (CWT) time-frequency images and an improved densely connected convolutional network (IDenseNet) was proposed. Firstly, the one-dimensional vibration signal was converted into two-dimensional time-frequency images by CWT. Then, the DenseNet network was turned into IDenseNet, the ReLU activation function in the first convolutional block of DenseNet was replaced by the Swish activation function(which was smoother), and the style-based recalibration module (SRM) and the convolutional block attention module (CBAM) were introduced into the DenseNet network. The SRM focused on the weight of feature channels, while CBAM enhanced the feature expression ability from the two dimensions of channel and space. Finally, the two-dimensional time-frequency image was input into the IDenseNet model for feature extraction and fault diagnosis, and the fault diagnosis results were output through the Softmax layer of the model. The results show that the average fault recognition accuracy of the proposed method reaches 97.80% under constant and variable conditions, and the average fault recognition accuracy reaches 99.44% in the transfer learning model. The CWT-IDenseNet method can effectively improve the generalization ability of the model, which has significant advantages under constant and variable conditions, providing reference for improving the accuracy and reliability of rolling bearing fault diagnosis.