An Imbalanced Fault Diagnosis Method Based on Residual Wasserstein Generative Adversarial Network

In actual industrial processes, operations generally remain in a normal state for an extended period, with faults occurring less frequently. Since the amount of data representing faults is small, there is an imbalance between normal samples and fault samples, which affects the accuracy of fault diagnosis. To solve this problem, this paper proposes an imbalanced fault diagnosis method based on the Residual Wasserstein Generative Adversarial Network(ResWGAN). Firstly, shortcut connections are added to the hidden layers of both the generator and the discriminator to form a residual network, replacing the fully connected layers. Secondly, ResWGAN is trained to generate virtual fault samples and construct a balanced data set. Thirdly, a one-dimensional convolutional neural network (1D CNN) is used to build a classifier to achieve fault diagnosis. Finally, the experimental results on the Tennessee Eastman process show that the proposed method has a high diagnostic accuracy.