Adaptive Maximum Second-Order Cyclostationarity Blind Deconvolution Based on Diagnostic Feature Spectrum for Rolling Bearing Fault Diagnosis

Abstract

Maximum cyclostationarity blind deconvolution effectively enhances the periodic components by maximizing the cyclostationary behavior associated with the fault-inducing source. However, the validity of maximum cyclostationarity blind deconvolution depends on prior knowledge of bearing characteristic frequency, which is influenced by the shaft rotation frequency and bearing elements. In addition, it tends to generate false cyclostationary components when the incorrect cyclic frequency is provided as input. To address the above problems, a diagnostic feature-based adaptive maximum cyclostationarity blind deconvolution (DFACYCBD) is proposed for identifying the incipient faults of bearings. A novel estimator known as the diagnostic feature spectrum (DFS) is introduced in this method, which is constructed based on a feature at each frequency in the enhanced envelope spectrum (EES). Specifically, the cyclostationary information of noisy signals is first extracted using the fast spectral correlation (Fast-SC) and then converted into the equal-frequency interval harmonic structure (EIHS) within EES. Subsequently, DFS is used to calculate the cyclic frequency, with the estimated result considered as the desired cyclic frequency. Even in conditions with heavy background noise, DFS is proven to yield precise estimated results as the cyclic frequency to input. Finally, the simulated signal and bearing vibration datasets are applied to validate the efficacy of DFACYCBD.