An Improved WGAN-Based Fault Diagnosis of Rolling Bearings

Abstract

The generative adversarial network (GAN) has been extensively applied in the field of fault diagnosis of rolling bearings under data imbalance. However, it still suffers from unstable training and poor quality of generated data, especially when training data is extremely scarce. To deal with these problems, an improved Wasserstein generative adversarial network (IWGAN)-based fault diagnosis method is put forward in this article. A classifier is introduced into the discriminator for gaining label information, thus the model will be trained in a supervised way to enhance stability. In addition, the matching mechanism of feature map is considered to ameliorate the quality of generated fault data. Then, by blending original data with generated data, a fault diagnosis method, by using stacked denoising autoencoder, is designed to realize fault diagnosis. Finally, the availability of proposed model is verified on the benchmark fault dataset from Case Western Reserve University. The results of the comparative experiments strongly indicate that IWGAN can not only effectively strengthen the balance of the original data but also enhance the diagnosing precision of rolling bearings.