Frequency Adaptive Maximum Second-Order Cyclostationarity Blind Deconvolution for Bearing Fault Diagnosis

Maximum second-order cyclostationarity blind deconvolution (CYCBD) has been demonstrated to exhibit excellent performance in revealing bearing fault impulses. However, its application heavily relies on prior knowledge of fault characteristic frequency (FCF), which cannot be obtained ahead in practical engineering. To address this problem, a novel frequency adaptive maximum second-order cyclostationarity blind deconvolution (FACYCBD) is proposed. In FACYCBD, a periodic hierarchical assessment method is proposed to facilitate the accurate estimation of cyclic frequency, where the unique merit of spectral coherence (SCoh) to simultaneously unveil both resonance frequency and FCF is fully exploited. Specifically, the fault information of all cyclic frequency slices (CFSs) is converted into numerical representations to select the optimal slice containing abundant fault information, and then the cyclic frequency is determined using the trend curve of the optimal slice. Especially, to avoid frequency interference independent of faults, only the optimal slice rather than all slices is considered to estimate cyclic frequency. Subsequently, the filtered envelope spectrum (ES) is constructed to extract fault features by employing estimated cyclic frequency as a replacement for FCF. The FACYCBD can adaptively perform without any prior knowledge about fault, making it better aligned with real-world industrial scenarios. Finally, the advantage of FACYCBD is validated through both simulation and experimental data captured from different test rigs and compared with other deconvolution methods.