Imbalanced fault diagnosis of rolling bearing using a deep gradient improved generative adversarial network

Abstract

Intelligent fault diagnosis is extremely significant to maintain the healthy operation of rolling bearings. However, the fault signals from practical engineering are often seriously imbalanced, which limits the stability and accuracy of intelligent methods. The traditional diagnostic methods rely on manual feature extraction and large-scale labeled data, which will consume a lot of time and manpower. To improve the performance of imbalanced fault diagnosis of rolling bearing, an intelligent method for fault diagnostic based on deep gradient improved generative adversarial network is proposed. Firstly, a deep generative adversarial network is constructed, which integrates a gradient improvement technique to impose a gradient penalty on the objective function of the discriminator to avoid mode collapse. Then, a deep feature matching method is used to enhance the deep features in the generator to improve the quality of the generated data and eliminate the overfitting of network training. Finally, an automatic data evaluator is established to conduct the data quality enhancement, ensuring the accuracy and diversity of synthetic samples in time. The experimental verification is carried out on the electric locomotive bearing dataset, and the diagnosis results of the proposed method are compared with the advanced intelligent diagnosis method. The results denote that the proposed method shows great advantages and potential in imbalanced fault diagnosis.