An Intelligent Method for Early Motor Bearing Fault Diagnosis Based on Wasserstein Distance Generati

The fault diagnosis method based on generative adversarial networks (GANs) has been successfully applied to the early fault detection of motor bearings, and it has effectively solved the problems of small samples, unlabeled sample features, and data imbalance in early faults. The existing methods, however, ignore the weak features of bearing vibration signals when solving the problem of missing early fault samples, which seriously affects the accuracy of diagnosis results. In addition, a large amount of human parameter adjustment work is required during the network model training process. This makes the whole diagnosis process lack objectivity and intelligence. To address these issues, this article proposes an intelligent diagnosis algorithm for early faults of motor bearings based on Wasserstein GAN meta learning (WGANML). First, the method performs feature extraction on the collected bearing vibration signals. Then, the generation of missing samples and the enhancement of weak features are completed by using the game between the generative and discriminative models. Second, to get rid of manual intervention, meta learning (ML) is applied to Wasserstein distance GANs (WGANs) parameter variation training for the first time. Finally, the feasibility and effectiveness of the proposed WGANML algorithm are proved by the open dataset of Case Western Reserve University (CWRU) and the data of the motor bearing fault experimental platform. In addition, compared with the existing advanced methods, the superiority of the proposed method in motor bearing early fault diagnosis is further verified.