A CNN-MFKAN-Based Fault Diagnosis Method for Rolling Bearings Under Multiple Working Conditions

Fault diagnosis in rolling bearings significantly influences the dependability and performance of industrial equipment. Considering the shortcomings of traditional fault diagnosis methods, such as poor generalization performance and insufficient feature extraction capability, this article proposes a bearing fault diagnosis method by combining a convolutional neural network with a multiscale fast Kolmogorov–Arnold network (CNN-MFKAN), which integrates the feature extraction capability of CNN with the nonlinear processing superiority of MFKAN. In addition, in order to further enhance the fault feature extraction ability, the convolution block attention module (CBAM) is introduced in this article under multiple working conditions. Finally, the proposed CNN-MFKAN model is evaluated by utilizing Harbin Institute of Technology aircraft engine intershaft bearing system dataset. The experiments show that the presented method maintains a high fault identification rate and classification accuracy under multiple working conditions, and the proposed model is obviously superior to the current mainstream transfer learning methods in cross-working conditions, which further illustrates the superiority and effectiveness of this scheme.