Bearing Fault Diagnosis in Noisy Environment Based on Continuous Wavelet Transform and Convolutional Attention Fusion Network

Accurate diagnosis of bearing faults in noisy environments is crucial for the safe operation of equipment under real working conditions. This paper introduces an innovative approach to bearing fault identification using continuous wavelet transform and a bidirectional fusion multi-scale convolutional attention network (BiMCAN). First, the original signal is processed using continuous wavelet transform to generate a timefrequency image. Then, the proposed multiscale convolutional SimMSC module and an improved Swin Transformer serve as the backbone network for feature extraction, with BiFPN used for multiscale feature fusion. Experiments conducted on the CWRU bearing dataset demonstrate that the proposed method outperforms other diagnostic methods in terms of diagnostic performance under strong noise and varying working conditions.