MOWES: An SCoh Weight-Optimized Framework for Enhanced Rolling Bearing Fault Diagnosis

Spectral coherence (SCoh) is a useful technique for uncovering the second-order cyclostationarity of rolling bearing fault signals. Recently, some methods incorporating the weighting idea into SCoh have been devised. They aim to extract bearing fault information in relatively discrete frequency band. However, for spectral frequency slices (SFSs) of SCoh, the weighting functions designed by most current methods are still rough, potentially causing inaccurate diagnostic results. Moreover, the tolerances used to determine fault characteristic frequencies (FCFs) are not robust enough in many studies. These methods tend to fail when the deviation between the nominal and actual values of FCFs is large. To address such issues, a new SCoh-based bearing fault diagnosis method named multiple optimization weighted envelope spectrum (MOWES) is proposed. The method devises the fault abundance (FA) as the initial weight function of SFSs, followed by a three-stage optimization: preliminary fault content evaluation, clarity quantification of slice fault characteristic frequencies (SFCFs), and similarity measurement among SFCF groups (SFCFGs). The derived final weight function can exactly recognize SFSs that contain rich fault infor-mation. Ultimately, MOWESs are constructed from the weighted average of the SFSs. The simulated and experimental results demonstrate that the proposed MOWES has superior diagnostic performance compared with four advanced methods.