Abstract: In order to extract weak impulse signal in heavy noise, an improved method based on wavelet denoising using neighboring coefficients is proposed. The scaling factor of wavelet neighboring coefficients is computed with a different method to better extract the impulsive feature. The denoising experiments of simulated signals of bearing with different signal-noise-ratios indicate that the method can extract the impulse feature in heavy noise. The experimental results of early rolling bearing fault feature extraction show that this method is better than the traditional neighboring coefficients denoising method. The proposed method can effectively extract the early fault feature of roller bearing and can extract the fault frequency of the bearing.

Key words: Denoise, Fault diagnosis, Neighboring coefficients, Wavelet transform