Abstract: In view of the disadvantage of the fault diagnosis method based on wavelet energy spectrum and energy spectrum entropy which requires wavelet decomposition coefficients to basically conform to Gaussian distribution, a new fault diagnosis method is proposed. This method analyzes the statistics features of the multi-scale wavelet coefficients of bearing vibration signals, uses a generalized Gaussian distribution (GGD) model to fit the histogram of wavelet decomposition coefficients of signals, adopts the maximum likelihood method to determine model parameters, and take these as signal characteristics to realize fault diagnosis. The proposed method is compared to methods based on wavelet energy spectrum, energy spectrum entropy and wavelet packet, and the results validate the correctness of design idea and the high-efficiency detectability of algorithm. The influence of wavelet base, window width and classifier on the diagnosis performance of the proposed method is analyzed, and the results show that the proposed method has very high stability and robustness.

Key words: Fault diagnosis, Generalized Gaussian distribution, Maximum likelihood method, Wavelet energy spectrum