Bearing fault feature extraction method based on improved maximum correlated kurtosis deconvolution

Abstract Bearing condition monitoring is crucial for ensuring the safety and reliability of rotating machinery. In the early stages of bearing fault, the signal-to-noise ratio of fault signal is low due to strong noise interference and transmission path effects. Therefore, maximum correlated kurtosis deconvolution (MCKD) is performed to bearing fault diagnosis widely with its good performance on feature enhancement. However, the performance of MCKD is also restricted largely by its parameters setup. To enhance the filtered performance, an improved MCKD (IMCKD) method based on parameters optimization is proposed in this paper. In IMCKD, fault period T is corrected by multipoint kurtosis (MK) instead of selecting theoretical fault period values. Furthermore, the parameters L and M are set by Parrot optimizer (PO) based on its the global search capability, which avoids the limitations of manual empirical tuning. The proposed IMCKD is validated through simulation and experimental bearing fault data. Compared with other signal filtered methods, the filtered results demonstrate superior performance in extracting weak fault features.