声振信号, 滚动轴承, 变分模态分解, Transformer 模型, 故障诊断

To address the limitations of single-sensor information and the low diagnostic accuracy of rolling bearing fault diagnosis in complex environments, a multimodal fusion method based on VMD(Variational Mode Decomposition) and a multi-head cross-attention mechanism is proposed. Acoustic and vibration signals are adaptively decomposed to extract key IMFs ( Intrinsic Mode Functions). A cross-attention mechanism is then employed to interactively fuse the features of acoustic and vibration signals, enabling deep multimodal feature extraction and noise suppression. Fault identification is performed using a Softmax classifier. Experimental results demonstrate that the proposed method effectively reduces noise interference and significantly improves diagnostic accuracy, exhibiting greater robustness and precision compared to traditional approaches.

acoustic vibration signal, rolling bearing, variational mode decomposition, Transformer model, fault diagnosis