Bearing Fault Diagnosis Method Based on Multi-axis Vision Transformer and Weighted Subdomain Adaptive Adversarial Network

In order to tackle the challenges of intricate bearing fault labeling in industrial environments and improve diagnostic accuracy under variable operational conditions, the Multi-Axis Vision Transformer with Weighted Subdomain Adaptive Adversarial Networks (MaxViT-WSAAN) is proposed in this paper. It combines the Multi-Axis Vision Transformer with adversarial networks and applies local feature perception of CNNs and global insights from Transformers. Adversarial learning narrows the distribution gap between source and target domains, and a subdomain adaptive module with weighted loss aligns subclass distributions. The approach, tested on the CWRU dataset, delivers 100% accuracy in variable condition testing, surpassing other methods by 1.43 percentage points, demonstrating strong robustness in bearing fault diagnostics across different conditions.