Journal of Propulsion Technology

In order to effectively extract the bearing fault information from the high frequency vibration signal generated by the bearing early fault excitation，the optimal wavelet packet basis selection method and the kurtosis value maximum screening principle are applied. A recognition method of the rolling bearing early fault is proposed based on the combination of improved wavelet packet decomposition（WPD），kurtosis value index（KVI）and Hilbert transform. Firstly，the optimal wavelet packet basis is calculated and the decomposition layer number is determined. Secondly，the WPD method is used to decompose the bearing fault vibration signals and several node components are obtained. Then，effective node components are selected for signal reconstruction based on the maximum principle of kurtosis index. Finally，the envelope demodulation analysis is carried out on the reconstructed signals，and the fault characteristic frequency is extracted to diagnose the bearing fault. The established method is used to analyze and diagnose the experimental data of outer ring and inner ring faults of ball bearing in Case Western Reserve University and the self-developed experimental data of outer ring and rolling body faults of roller bearing. The results show that the method can effectively improve the high-frequency resolution of fault signals and retain periodic impact components. In addition，this method can also accurately and effectively extract the 1~7 frequency doubling of the faults characteristic frequencies of ball bearing and roller bearing，and a series of their side band frequency modulated with the rotating frequency of shaft，which realizes accurate identification and effective diagnosis of rolling bearing faults.

Optimal wavelet packet base；Kurtosis value；Ball bearing；Roller bearing；Early fault diagnosis