Nonconvex Group Sparsity Signal Decomposition via Convex Optimization for Bearing Fault Diagnosis

Bearing fault diagnosis is critical for rotating machinery condition monitoring since it is a key component of rotating machines. One of the challenges for bearing fault diagnosis is to accurately realize fault feature extraction from original vibration signals. To tackle this problem, the novel group sparsity signal decomposition method is proposed in this article. For the sparsity within and across groups' property of the bearing vibration signals, the nonconvex group separable penalty is introduced to construct the objective function, leading to that the noise between the adjacent impulses can be eliminated and the impulses can be effectively extracted. Furthermore, since the penalty function is nonconvex, the convexity condition of the corresponding objective function to the global minimum is discussed. In addition, to improve the efficiency of parameter selection, this article presents an adaptive regularization parameter selection strategy. Simulation and experimental studies show that compared with the traditional method, the proposed method can better preserve the target components and reducing uncorrelated interference components for bearing fault diagnosis.