Bearing Health Detection Method Based on Wavelet Time-Frequency Spectrum and Parametric Improved CNN

Abstract

In order to improve the accuracy of bearing fault diagnosis and the problem of slow training speed, this paper proposes a bearing health detection method based on wavelet time-frequency spectrum and Alexnet framework of CNN. Firstly, the bearing signals of different parts are continuously wavelet transformed, which are converted into two-dimensional time-frequency maps. Next, Gaussian white noise is added to the transformed results. We test the ability of anti-noise and verification accuracy under Alexnet, restnet and shufflenet networks respectively. Finally, the network with the best ability of anti-noise and accuracy is selected for a 10×10 grid search to determine its best parameter settings. This paper validates the method using the Case Western Reserve University bearing data center set, and the experimental results show that Alexnet based on wavelet time-frequency spectrum and grid method parameter optimization has high accuracy and ability of anti-noise.