Conditional empirical wavelet transform with modified ratio of cyclic content for bearing fault diagnosis

Empirical wavelet transform (EWT) is usually employed to segment Fourier spectrum for fault diagnosis. However, the original empirical segmentation approach may be easily affected by noise. In this paper, several conditions and a modified ratio of cyclic content are then proposed to help establish proper spectrum segments and to improve fault diagnosis. The proposed conditions include a pre-whitening process to reduce discrete frequency noise, a threshold to avoid white frequency noise, an additional boundary for the last considered maximum, distance requirement for consecutive local maxima, as well as one iteration of finding local extremums. Finally, the proposed method is compared with EWT and fast kurtogram methods in three case studies. The results indicate that the proposed method can provide more favorable diagnosis results.