A bearing fault image recognition method based on reliability analysis and RSO-optimized Alexnet
Shangbin Gao, Jiale Li, Zhun Zhang · 2023
Aiming at the problem of insufficient feature extraction ability and credibility loss in the recognition of bearing fault image, a bearing fault diagnosis method based on credibility analysis and RSO (Rat Swarm Optimizer) optimized Alexnet is proposed. Firstly, the collected bearing original vibration signal data is converted to a time-frequency graph by the method of signal conversion to image, and generate data sets. Secondly, the noise resistance of various lightweight network frames is tested, and the optimal network frame is obtained. Then, the parameters of the network framework are compared and adjusted by the progressive mesh method and the RSO algorithm. Finally, the interpretability and reliability characteristics of the network framework are illustrated by the Grad-CAM method. The experimental results of Case Western Reserve University bearing data set show that the proposed method can achieve close to 100%bearing fault image recognition accuracy and good anti-noise performance, and has excellent application prospects.