An Fault Diagnostic Method Based on DRN-ACGAN for Data Imbalance

Abstract

Data imbalance, usually occurring in practical industrial engineering, restricts the effective application of intelligent fault diagnosis. To solve the data imbalance between faulty and healthy samples, an enhancement fault diagnosis method is proposed based on Deep Residual Network and Auxiliary Classifier Generative Adversarial Network (DRN-ACGAN). To improve the data enhancement effect, the ACGAN is optimized in two ways. Firstly, the generator uses DRN to prevent the gradient disappearing and over fitting problems caused by the deepening of network layers, improve the learning effect of useful features, and generate better quality samples. Secondly, Instance Normalization (IN) is incorporated into each layer of the generator network to avoid deviation of data. The validation experiments, as well as comparisons with the existing methods, are carried out for the bearing fault diagnosis under practical fault conditions. The results reveal that the proposed method can effectively improve the diagnostic performance for the imbalanced data.