Collaborative Double Sparse Period-Group Lasso for Bearing Fault Diagnosis

The localized faults of bearings can be diagnosed by extracting approximately periodic impulses from vibration signals. However, this feature may be deeply submerged in the high-level noise. In this article, a novel collaborative double sparse period-group lasso (CDSPGL) algorithm is proposed. The algorithm is based on two main priors of the fault bearing signal. The first is provided by the resonance frequency, and the second is provided by the fault characteristic frequency. Moreover, a novel collaborative period estimation strategy is developed to interact with the two priors according to the structural relationship between the two group-sparse models. Meanwhile, selection rules of regularization parameters are discussed in detail. Finally, the superiority of CDSPGL is verified through numerical simulation and diagnostic application.