Fault Diagnosis of Rolling Bearings Based on SSAE and Domain Adversarial

This paper proposes a domain adversarial neural network DASSM model based on Stacked Sparse Autoencoder (SSAE) and Maximum Mean Discrepancy (MMD) for the complex mapping relationship between vibration signals and health conditions under fluctuating working conditions of rolling bearings. By utilizing the characteristics of unsupervised learning in SSAE, the optimal values of sparse parameters in the proposed model are analyzed to construct an effective feature extraction network structure. This can extract effective features from the vibration signal sample data of fluctuating working conditions, providing a foundation for the transition from stable working conditions to fluctuating working conditions. By using domain distance measurement and domain adversarial alignment simultaneously, the generalization ability of the model is improved, effectively reducing the difference in feature samples between the source and target domains. The experimental results show that the DASSM model can achieve the transition from stable working conditions to fluctuating working conditions, with a bearing fault diagnosis rate of over 97% and good robustness of the model.