Fault Diagnosis Method Based on CWGAN-GP-1DCNN

Abstract

In the actual industrial process, the fault data collection is difficult, and the fault sample is insufficient. The Imbalanced datasets is the main problem that is faced at present. However, the fault diagnosis method based on model optimization has over-fitting phenomenon in the training process. Therefore, using data enhancement methods to provide effective and sufficient fault samples for fault detection and diagnosis is a research hotspot to deal the data imbalance problem. To solve this problem, in this paper, a Conditional Wasserstein Generative Adversarial Network (CWGAN-GP1DCNN) with gradient penalty based on one dimensional Convolutional Neural Network is proposed to enhance the data of real fault samples to detect all kinds of bearing faults. Experimental results show that the proposed method can effectively enhance the sample data, improve the diagnosis accuracy under the condition of unbalanced fault samples, and has good robustness and effectiveness.