Optimal Demodulation Band Extraction Method for Bearing Faults Diagnosis Based on Weighted Geometric Cyclic Relative Entropy

Optimal demodulation band extraction is a significant step in rolling bearing fault analysis. However, existing methods, primarily based on global indexes and neglecting negative local outliers, cannot identify compound faults intense noise environments. To address this problem, novel method weighted geometric cyclic relative entropy (WGCRE) proposed. WGCRE defined the sub-bands model of logarithmic envelope spectrum (LES) to fully consider characteristic frequency pseudo-cyclostationarity. In detail, thresholds are separately set by white parameter harmonic-to-noise ratio exclude exogenous outliers. On basis, as geometrically index several different types avoid harmonic interference improve identification composite faults. WGCRE–gram, similar fast kurtogram (FK), then constructed replacing kurtosis with extract optimal band. Compared FK another LES-based method, logarithmic-cycligram, proposed more robust for accurately identifying single under external noise. The effectiveness verified through simulations actual tests. Simulation experiments kinds intensities preliminarily determine superior robustness face solid inner ring, outer further confirmed adaptability complex working conditions.