Fault diagnosis of rolling bearings under time-varying speed based on the residual attention mechanism and subdomain adaptation

Aiming at the inconsistent distribution of rolling bearing vibration signal data characteristics under strong noise and time-varying speed and in view of the fact that the failure samples to be tested do not contain labels, a rolling bearing fault diagnosis method was proposed based on the residual attention mechanism and sub-domain adaptive unsupervised transfer learning technology. Firstly, in order to give full play to the image classification capabilities of the convolutional neural network (CNN), the one-dimensional time-domain fault vibration signal under time-varying speed was converted into a two-dimensional grayscale image by using continuous wavelet transform (CWT), which was then used as the input of the model in this article. Secondly, in order to better extract the common features of the source and target domains, a feature extractor was constructed by adopting the residual channel attention weak sharing network model proposed in this paper, in which the cross-layer connection method of the residual network and the channel attention mechanism, were used and thus the structural conditions of the traditional strong sharing network model were weakened. Thirdly, in order to match the condition distribution difference between the source domain and the target domain, the local maximum mean discrepancy (LMMD) measurement criterion was embedded in the network adaption layer. Finally, the public fault data sets of rolling bearings under time-varying speed were used for experimental verification and analysis. The results show that the method proposed achieves an average recognition accuracy of more than 93% under strong noise and time-varying speed, and has better properties of generalization and robustness than traditional convolutional neural network models. © 2022 Chinese Vibration Engineering Society. All rights reserved.