A two-stage blind deconvolution strategy for bearing fault vibration signals

Abstract

Abstract The presence of periodic sparse impulses in vibration signals often indicates the occurrence of machine faults. This study focuses on the detection and diagnosis of an exact fault in bearings. As a matter of fact, the measured bearing fault signal corresponds to the convolution between periodic impulses, caused by periodic shocks in a faulty bearing, and the impulse response function of the mechanical system. However, the measured periodic impulses are generally weak and dominated by noise and other interferences. In this scenario, this paper introduces a novel approach to identify and restore periodic transients due to bearing faults through a deconvolution process based on sparsity. The proposed two-stage deconvolution strategy is based on an adapted Continuous Single Best Replacement algorithm. The major benefit of this blind deconvolution technique is the ability to estimate the impulses by exploiting a bearing fault model. The application on simulated and experimental data shows the effectiveness of this method in recovering the periodic impulses.