Simulated Data Driven GAN-Based for Fault Diagnosis Under Missing Fault Types

Generative adversarial networks (GANs) are often used to deal with the problem of sample data imbalance. However, traditional correlation algorithms may encounter problems regarding training stability or convergence difficulty. In addition, GANs can only generate existing fault samples, while it is likely that there is no corresponding fault sample data in practical engineering. To address this problem, this paper proposes a method based on a dynamical model with generative adversarial networks. A dynamics model is built to obtain the simulation signals of missing faults, and Wasserstein Auxiliary Classifier Generative Adversarial Network with Gradient Normalization (WACGAN-GN) is used to supplement and generate the imbalance data. To verify the validity of the simulated samples, the enhanced dataset is classified using a stacked autoencoder (SAE). The results show that the model performs well in fault diagnosis under missing sample conditions and outperforms other methods.