Research on Fault Diagnosis of Rolling Bearings Based on Progressive Grid Optimization ResNet

The traditional rolling bearing fault diagnosis method requires a series of complex signal processing and relies heavily on the manual method of constructing algorithms and the empirical knowledge of experts, which leads to the introduction of some artificial random factors. Based on this situation, we searched the parameter settings of ResNet by the progressive grid method, to achieve the effect that no more artificial parameter settings are needed, eliminating the artificial random factors and making it more capable of feature extraction. The specific steps are as follows: first, a two-dimensional time-frequency variation of the bearing vibration data is used to construct the data set, followed by a network noise immunity test, then the data is input into the network framework, and a progressive grid method is used to search the parameters to obtain a better parameter setting; finally, an image is made by the variation of t-SNE and the feature coefficients of the sparse measure to show the feature extraction results of the network before and after optimization. The visualization comparison graph and it is found that the classification accuracy is higher in the case of good parameters by testing. By comparing different network architectures and adding the progressive grid method, we verify that: the framework obtained after optimizing the parameters has better noise immunity and adaptive feature extraction capability, and the bearing diagnosis accuracy is higher than that of the method before parameter optimization.