Rolling Bearing Fault Feature Extraction Method Using Adaptive Maximum Cyclostationarity Blind Deconvolution

Abstract

Maximum cyclostationarity blind deconvolution (CYCBD) is an effective method for extracting bearing weak fault impulse. However, the CYCBD-based bearing fault diagnosis method has the following problems: setting the cyclic frequency and filter length requires artificial experience guidance, and improper settings lead to incorrect diagnostic results. To this end, a novel method of parameter adaptive maximum cyclostationarity blind deconvolution (ACYCBD) is proposed for bearing fault diagnosis. This method analyzes the cyclostationarity of the signal by the fast spectral correlation (Fast-SC) algorithm and obtains the enhanced envelope spectrum (EES) of the signal. The cyclic frequency is accurately estimated using envelope harmonic product spectrum (EHPS) based on the harmonic-related spectral structure (HRSS) in EES. Finally, the filter length is determined by envelope entropy efficiency assessment (EEEA) index. The proposed ACYCBD-based bearing fault diagnosis method is validated by simulated signal and experimental data to effectively extract weak fault impulses from bearing vibration observation signal without prior knowledge.