Generalized Statistical Indicators-Guided Signal Blind Deconvolution for Fault Diagnosis of Railway Vehicle Axle-Box Bearings

Vibration impulses caused by surface defects of railway vehicle axle-box bearings are important features for fault diagnosis. To accurately diagnose axle-box bearing surface defects, robust blind deconvolution (BD) methods are proposed to extract defect-induced repetitive impulses from axle-box vibration measurements in this article. Specifically, a unified theoretical framework for the BD methods using time-domain and frequency-domain generalized statistical indicators as objective functions is established respectively by converting the deconvolution problem into a generalized Rayleigh quotient-based generalized eigenvalue problem. On this basis, two new BD methods for extracting repetitive impulse features are presented by employing the robust power function-based Gini index as the objective function. They obtain optimal deconvolution filters by adaptively evaluating repetitive transient features in the squared envelope and squared envelope spectrum of the filtered signal, respectively. The developed methods are verified on bench experimental signals of manufactured and naturally occurring axle-box bearing faults. The results show that the presented BD methods can effectively enhance repetitive impulse features and diagnose various defects of railway vehicle axle-box bearings.