Data augmentation in fault diagnosis based on the Wasserstein generative adversarial network with gradient penalty

Abstract

Fault detection and diagnosis in industrial process is an extremely essential part to keep away from undesired events and ensure the safety of operators and facilities. In the last few decades various data based machine learning algorithms have been widely studied to monitor machine condition and detect process faults. However, the faulty datasets in industrial process are hard to acquire. Thus low-data of faulty data or imbalanced data distributions are common to see in industrial processes, resulting in the difficulty to accurately identify different faults for many algorithms. Therefore, in this paper, Wasserstein generative adversarial network with gradient penalty (WGAN-GP) based data augmentation approaches are researched to generate data samples to supplement low-data input set in fault diagnosis field and help improve the fault diagnosis accuracies. To verify its efficient, various classifiers are used and three industrial benchmark datasets are involved to evaluate the performance of GAN based data augmentation ability. The results show the fault diagnosis accuracies for classifiers are increased in all datasets after employing the GAN-based data augmentation techniques.