Fault feature extraction and diagnosis of rolling bearings based on wavelet thresholding denoising with CEEMDAN energy entropy and PSO-LSSVM

In order to improve identification accuracy of rolling bearings with nonlinear and nonstationary vibration signals, a novel fault diagnosis method based on wavelet thresholding denoising, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) energy entropy, and particle swarm optimization least-squares support vector machine (PSO-LSSVM) is proposed. A wavelet thresholding denoising method is first applied to the vibration signals to reduce the noise-induced interference. Second, CEEMDAN decomposition is performed on the denoised signal to obtain multiple groups of intrinsic mode functions (IMFs), and the selection of feature vectors is carried out by combining the correlation coefficient and variance contribution rate to eliminate false feature components. The energy entropy of the selected IMF component is calculated, which is input into the PSO-LSSVM classifier as a feature vector for fault diagnosis and classification. The results show that the identification rate of various fault states of rolling bearings can reach 100%.