Bearing fault diagnosis based on Cluster-contraction Stage-wise Orthogonal-Matching-Pursuit

This paper presents a novel Cluster-contraction Stage-wise Orthogonal-Matching-Pursuit (CcStOMP) approach for bearing fault information extraction. This approach adds the cluster contraction mechanism to the Stage-wise Orthogonal-Matching-Pursuit (StOMP) algorithm and filter the selected atoms twice during atomic search, which makes the condition number of the support set more reasonable, thus realizing amelioration of the pathological equation in weight determination. It can improve the accuracy of sparse recovery while maintaining the rapid convergence characteristic, with good robustness. Both simulation and experimental studies have verified that the proposed CcStOMP approach can extract bearing fault feature component more precisely compared with StOMP, thus improving the accuracy of fault diagnosis.