Bearing Fault Diagnosis under Different Working Conditions Based on Generative Adversarial Networks

Aiming at the problem of different bearing data distributions caused by load and speed changes, the fault diagnosis model based on maximum mean discrepancy and Wasserstein generative adversarial network is proposed. The proposed model adopts the adversarial learning method with Wasserstein distance to realize the unlabeled feature transfer between the source domain and the target domain and minimizes the maximum mean difference between the source domain and the target domain. More specifically, the invariant features between the source domain and the target domain after minimizing the maximum mean difference are learned to achieve fault classification. The fault diagnosis of rolling bearing under different working conditions is carried out through the proposed transfer learning-based model and the purpose is to solve the fault diagnosis problem of unlabeled target domains. From the experimental results, the proposed fault diagnosis model can successfully diagnose rolling bearing faults under different working conditions.