A Novel Method for Enhancing Bearing Fault Feature Based on Sparse Group Lasso

Abstract

Bearing fault signal is covered by other signal components, making it difficult to diagnose the bearing fault and leading to accidents accurately. Thus, enhancing the fault features of the signal is significant to detect faults in bearings. A Group Sparse Lasso (SGL) model which can be solved using the Majorization-Minimization (MM) optimization algorithm is proposed to achieve the enhancement of vibration signal characteristics of faulty bearings. First, the sparse optimized objective function of the SGL is constructed, which remarkably preserves the signal's transient features while ensuring sparsity. Second, the linear programming problem is transformed into a convex optimization solution by the MM optimization algorithm, and an iterative solution can attain a series of sparse coefficients. Finally, the features of bearing faults are extracted using envelope analysis. The practicality of the method is proved by the simulated signals and two sets of experimental data. The results show that the technique can diminish the fault signal's redundant components and sparsely represent the fault signal while effectively enhancing and extracting the fault signal's features.