Maximum cyclostationary characteristic energy index deconvolution and its application for bearing fault diagnosis

Blind deconvolution can effectively highlight the fault impulses submerged in vibration signals. However, the objective function of existing deconvolution methods deeply depends on prior knowledge about fault periods, which is often difficult to acquire in advance or may lack accuracy in practical applications. Additionally, its filter length selection remains an open problem, hindering the performance and generalization in industrial scenarios. To address the above issues, a maximum cyclostationary characteristic energy index deconvolution (MCCEID) is proposed to recover periodic impulses, where a novel cyclostationary characteristic energy index (CCEI) is established as the objective function. The CCEI captures local variation features at the fault characteristic frequency (FCF) to iteratively enhance periodic components, instead of focusing on aperiodic noise. Meanwhile, a periodic hierarchical assessment method is developed to sequentially identify resonance frequency slice and FCF, in which the interference from other frequencies is excluded and only the resonance frequency slice is applied to estimate FCF. In addition, an adaptive filter length determination framework is designed considering both the value of CCEI and time cost, thereby avoiding the manual determination of the filter length. The performance of MCCEID is demonstrated by simulated and experimental signals, and the results illustrate that MCCEID outperforms the other methods in fault feature extraction.