Evolutionary Wavelet Denoising and Its Application to Ball Bearing Fault Diagnosis

Abstract

Threshold is the key factor in threshold-based wavelet denoising. Conventional threshold estimation methods fail to estimate the appropriate threshold for ball bearing fault signals denoising. To improve the denoising performance of threshold-based wavelet denoising with the conventional threshold estimation methods, an evolutionary wavelet denoising method is proposed. In the method, wavelet transform is used for the noise-contaminated signal decomposition and reconstruction, a function that approximates to the estimation error of hard thresholding is constructed and then the optimal threshold at each decomposition level is obtained by applying particle swarm optimization to the constructed function. Extensive numerical experiments on simulated signals and ball bearing fault signals are carried out to confirm the effectiveness of the method. The experimental results indicate that the method is highly effective in noise reduction and fault feature extraction. In comparison with the conventional threshold-based wavelet denoising methods, the method has better denoising performance in the sense of signal-to-noise ratio and mean square error.