Bearing Fault Image Classification Method Based on Interpretable Hyperparameter Optimization Model

With the refined development of industrial equipment, the health state of industrial parts such as bearing is particularly important. The analysis method of bearing fault images has also become an important issue in the direction of industrialized fault diagnosis. There are many difficulties in the analysis of fault diagnosis. In the face of strong background noise, the model is weak and the parameters have the problem of random factors. This paper is proposed to classify the bearing fault image classification method based on explanatory decision -making fusion and super-added model optimization models. This paper first conduct a two-dimensional waves change of the original one-dimensional data. Based on the wave analysis of the CMOR function, it is converted to a two-dimensional image with a variety of waves, REST NET18 and other networks for noise testing to get some network frameworks with strong robustness. Based on the network framework for super-added optimization, different group optimization algorithms (GWO, WOA, etc.) are used to compare Optimize algorithms, build a model with strong feature extraction capabilities, and use class activation mapping to make decision-making explanations. Finally, after public data verification, the model this paper obtained can cope with strong background noise, and can well overcome the random factors of setting the parameters. At the same time, the decision-making explanation of the model can be used in each problem and in the actual project.