Bearing Fault Prediction Based on Mixed Domain Features and GWO-SVM

The rotating machinery is composed of rolling bearing connection, so the fault identification of rolling bearing is a very critical task. We propose a bearing fault identification algorithm based on grey wolf optimizer (GWO) to address the common problems of high signal noise, inability of a single indicator to accurately reflect the true state of bearings, and optimization of support vector machine (SVM) prediction model parameters in bearing fault identification. First, the wavelet soft threshold is used to remove the noise of the original signal and then the empirical Fourier decomposition (EFD) is used in the decomposition and reconstruct signals. Second, in the aspect of feature extraction, the time and frequency domain features of the bearing data are selected to form the mixed domain features of the bearing signal. Finally, aiming at improving the bearing fault prediction accuracy, the GWO algorithm is used to optimize the parameters. Achievements: the signal-to-noise ratio can be effectively improved to 77.8 by using the wavelet denoising, and the parameter modeling optimized by the GWO algorithm can significantly improve the prediction accuracy, with an increase of about 3%–5%. It provides theoretical support for the optimization of bearing fault identification with this technology in the industrial field.