An Adaptive Weighted Adjacent Difference Sparse Representation for Bearing Fault Diagnosis

Abstract

Vibration monitoring is the effective and convenient method for machine fault diagnosis. However, the fault feature extraction (i.e., impulse components) is difficult due to strong noise. In this article, a novel adaptive weighted adjacent difference sparse representation (AWAD-SR) is proposed for bearing fault diagnosis, which promotes the sparsity while reducing noise interference. First, the adjacent difference sparse regularization term is proposed for impulsive feature extraction from vibration signal. Second, an adaptive weighting method is proposed for solving the challenging problem on setup of regularization parameter that affects the feature extraction performance of those sparse representation (SR) methods. AWAD-SR setups adaptively the regularization parameter of each sampling point in a vibration signal according to their amplitude, which improves not only the feature extraction performance but also the applicability in practical engineering application cases. Finally, an effective majorization-minimization (MM)-based optimization algorithm is proposed for solving the sparse optimization problem of the weighted SR, which has the advantages of fast convergence speed. The effectiveness and superiority of AWAD-SR are validated on the simulation signal and two bearing vibration signals. The comparison results show that feature extraction of AWAD-SR from vibration signals is superior to other state of-the-art approach.