Multiple signal blind deconvolution for fault feature enhancement of rolling element bearings

Abstract

Transient repetitive impulses are a typical feature of localized damage to rolling element bearings. Enhancing the damage-induced impulse components hidden in collected vibration signals is an important issue for local defect recognition in bearings. Blind deconvolution (BD) is a typical approach for enhancing bearing impact impulses, which focuses on utilizing a filter to highlight the fault features in the filtered signal. Minimum entropy deconvolution is the most classic BD approach that solves the filter by maximizing the kurtosis value of the filtered signal. Inspired by this, various BD methods have been proposed. However, these methods are all applicable for extracting fault features from a single signal. Therefore, this article extends existing BD methods to multiple signals and proposes a multiple signal blind deconvolution (MSBD) guided by the weighted sum of multiple kurtosis for repetitive transient enhancement. The core of this new method is to utilize a filter to recover the fault impulse components in multiple signals simultaneously. An iterative solution strategy is devised to filter multiple signals by maximizing the weighted sum of their kurtosis values. The efficacy of MSBD was validated in comparison with existing BD methods using numerically synthesized and experimental bearing signals. The results showcase that, compared to typical BD methods, the proposed MSBD has more advantages in extracting bearing fault features by utilizing multiple signals.