Bearing Fault Feature Extraction Based on Adaptive OMP and Improved K-SVD

Abstract

The condition of the bearing is closely related to the condition and remaining life of the rotating machine. Targeting the problem of the large number of harmonic signals and noise signals during the operation of rolling bearings, and given that it is difficult to identify the fault in time, an adaptive orthogonal matching pursuit algorithm (OMP) and an improved K-singular value decomposition (K-SVD) for bearing fault feature extraction are proposed. An adaptive OMP algorithm is applied, which uses the Fourier dictionary to improve the solution method of the OMP algorithm so that it can separate the harmonic components in the signal faster and more accurately. At the same time, the stopping criterion of the adaptive sparsity is improved in dictionary learning. There is no need to manually set the sparsity in the algorithm initialization process, which avoids the problem of algorithm performance degradation due to improper sparsity settings, and improves the efficiency of the K-SVD algorithm. As shown by theoretical verification, algorithm comparison, and experimental comparisons, the algorithm has certain advantages in fault feature extraction during rolling bearing operation, and the algorithm still has considerable practical value in long-duration and strong noise environments.