A Bearing Fault Diagnosis Method Based on Multimodal Attention Feature Fusion

This paper proposes a fault diagnosis method based on Multi-modal Attention Feature Fusion (MMAFF). Firstly, a multi-modal processing strategy is introduced, converting one-dimensional vibration signals into two-dimensional image modalities, and extracting features from both signal and image channels. Secondly, the attention mechanism is introduced to integrate the dual path features. Then, the diagnosis framework is obtained through an improved residual block. Finally, the method is tested on the Case Western Reserve University (CWRU) rolling bearing dataset, and cross-condition comparison experiments demonstrate that the proposed method (MMAFF) exhibits good recognition capabilities under different working conditions.