Accurate fault diagnosis of rolling bearings is critical yet challenging due to severe data imbalance in practical industrial scenarios. To address this issue, we propose a novel adaptive fusion-based multi-domain double-attention diffusion model for imbalanced fault diagnosis. The multi-domain double-attention diffusion leverages a dual-attention mechanism to generate high-quality fault signals, while the adaptive fusion module integrates multi-domain features through cross-attention to enhance representation learning. Extensive experiments on the CWRU dataset demonstrate significant performance improvements, achieving 99.57% accuracy under a severe imbalance ratio of 1:100, confirming the method's effectiveness and generalization capability.
 

Rolling bearing,Fault diagnosis,Data imbalance,Dual-attention mechanism,Multi-domain fusion