A Swin Transformer-based fault migration and diagnosis approach for gearboxes

Fault migration and diagnosis approaches are a long-term research focus among scholars and engineers. However, the fault samples of mechanical systems are insufficient, and traditional learning algorithms have limitations when trained with a limited number of labeled samples. This paper presents a method called STrans-MADA, which combines the Swin Transformer (STrans) and adversarial domain adaptation (ADA). During the training stage, STrans is embedded into the domains to improve image recognition ability and computational speed. During the adaptation stage, two GANs are utilized to augment the labeled fault sample information, align the features of both domains, and achieve fault migration and diagnosis. The STrans-MADA is validated using two datasets and compared with other fault diagnosis models. The average classification accuracy of STrans-MADA is 99.09% and 98.31%, respectively, demonstrating its high diagnostic accuracy. These experimental results confirm the effectiveness and advantages of our proposed method in terms of fault migration and diagnosis.