Rolling bearing fault diagnosis method based on dynamic simulated model from source domain to target domain with improved alternating transfer learning

Currently, the source domain data for transfer learning in rolling bearings is predominantly derived from laboratory experiments, and the loss function of the network is single. To address these issues, this paper proposes Rolling bearing fault diagnosis method based on dynamic simulated model from source domain to target domain with Improved Alternating Transfer Learning. The paper considers factors such as the real-time position of the rolling elements, the size of fault defects, and the bearing speed to adjust the dynamic simulated model of rolling bearing mechanics, thus obtaining a source domain dataset with rich fault labels. An improved alternating transfer learning approach is introduced, which narrows the domain gap by alternating loss function calculations and harnesses the complementary strengths of different loss functions. To validate the proposed method, the Case Western Reserve University bearing dataset is used as the target domain, and three experiments are designed for verification, including scenarios with the same bearing model, cross-bearing models, and small sample datasets when transferring from the simulation domain to the target domain. The results indicate that, compared to other algorithms, this paper's algorithm achieves a higher accuracy in fault classification.