Abstract:There are lots of inband noise left behind when wavelet packet decomposition (WPD) is used to process signal with low SNR, which seriously affects the accuracy of rolling bearing fault diagnosis. Aimed at this issue, the frequencyweighted energy operator (FWEO) was proposed as a postprocessor of WPD to eliminate the inband noise and enhance the effect of fault feature extraction. The three layers WPD was carried out on the sampled fault data, and the coefficients of each frequency band were obtained. The kurtosis of each frequency band coefficient was calculated to obtain the optimal frequency band coefficient based on the principle of kurtosis maximization. The FWEO was used to track the instantaneous energy of the optimal frequency band coefficient, the inband noises was eliminated and fault impulse response information was further enhanced from the view point of signal energy. The final diagnosis results were obtained by analyzing the envelope spectrum. The effectiveness and practicability of the proposed method were verified through simulation data, laboratory data and engineering data.