A Fast bearing Fault diagnosis method based on lightweight Neural Network RepVGG

Abstract

In view of the shortcomings of existing deep learning methods in rolling bearing fault diagnosis, such as large number of training parameters and complex network, a fast rolling bearing fault diagnosis method based on lightweight neural network RepVGG was proposed. Firstly, the vibration signal is converted into three-channel time-frequency image by the combination of short-time Fourier transform (STFT) and pseudo-color processing technology, then the time-frequency image is inputted into the RepVGG network model for training. and the experiment is carried out on the case Western Reserve University (CWRU) data set. The accuracy is 99.62% and the training time is obviously lower than other popular fault diagnosis algorithm models based on deep learning. Finally, using the open source framework ncnn to deploy the RepVGG network model to the edge computing node Raspberry Pi, the average test accuracy is 95%, and the running efficiency is good.