Fault diagnosis method of rolling bearing based on Group-Sparsity Learning

In bearing fault diagnosis, identifying fault features under strong background noise is a challenging task. As the weak fault signal of bearing is often interfered by noise, it's easy to cause the signal-noise ratio to be low. We combine the advantages of both variational modal decomposition (VMD) and sparse Bayesian learning (SBL) algorithms and propose a Group-Sparsity Learning Algorithm for Bearing Fault Diagnosis (VMD-SBGSL) based on the VMD and SBL. Compared with the traditional SBL, the proposed framework captures the group-sparsity structure of fault pulses, not only its own accuracy, but also the influence of adjacent accuracy is considered. The method firstly performs variational modal decomposition on the fault signal, selects the optimal modal number K and selects the optimal Intrinsic Mode Function (IMF) according to specific rules. Then use Variational Bayesian Inference (VBI) to solve the optimized SBGSL algorithm to recover the faulty pulse signal. No prior knowledge and no manual tuning of parameters is required in this process. Finally, envelope analysis is performed on the extracted results to identify bearing fault features. The method is applied to the actual fault data of rolling bearings and the comparison with two existing methods was carried out. According to the results, the method can effectively eliminate the interference of noise and can achieve accurate diagnosis of rolling bearing faults.