Identifying Bearing Faults Using Multiscale Residual Attention and Multichannel Neural Network

To solve the problem of low signal-to-noise ratio and fault features can only be extracted from a single scale traditional convolutional neural network (CNN) in vibration-based bearing diagnosis, this paper proposes new multi-scale residual attention multi-channel (MSCNet), which effectively reduce noise fully extract meaningful different scales signal. The proposed method combines filtering methods to remove redundant parts signal, multiple filtered signals are input into CNN. CNN perform feature extraction on signal make focus valuable information through mechanism. Therefore, MSCNet achieves better performance. Experimental results two published datasets show that higher accuracy than five state-of-the-art (SOTA) networks strong environments.