A Rolling Bearing Fault Diagnosis Method Based on Improved GAN and Multi-scale Feature Fusion SwinT Under Small Sample Conditions

To solve the problems existing in the fault diagnosis of rolling bearings, i.e., the category imbalance caused by the limited fault samples and numerous health samples, and the low accuracy of traditional diagnosis models, this paper proposes a new fault diagnosis method based on the improved Generative Adversarial Network(GAN) and the improved Swin Transformer framework. We first convert the one-dimensional vibration signal into a two-dimensional time-frequency image by the continuous wavelet transform. And then we use an enhanced GAN structure is utilized to generate synthetic fault samples from the images, which enriches the dataset and alleviates class imbalance. Thirdly, the enriched dataset is fed into the multi-scale feature fusion SwinT encoder, which learns the discriminative features and recognizes fault types. The experiment results show that the proposed SAWGAN CAAswinT method effectively alleviates the problems of data scarcity and class imbalance, and achieves high diagnostic accuracy at the same time. It is superior to traditional methods in rolling bearing fault diagnosis.