Enhanced Sparse Period-Group Lasso for Bearing Fault Diagnosis

Bearing faults are one of the most common inducements for machine failures. Therefore, it is very important to perform bearing fault diagnosis reliably and rapidly. However, it is fundamental but difficult to extract impulses buried in heavy background noise for bearing fault diagnosis. In this paper, a novel adaptive enhanced sparse period-group lasso (AdaESPGL) algorithm for bearing fault diagnosis is proposed. The algorithm is based on the proposed enhanced sparse group lasso penalty, which promotes the sparsity within and across groups of the impulsive feature of bearing faults. Moreover, a periodic prior is embedded and updated dynamically through each iteration of the optimization procedure. Additionally, we formed a deterministic rule about how to set the parameters adaptively. The main advantage over conventional sparse representation methods is that AdaESPGL is parameter free (forming a deterministic rule) and rapid (extracting the impulsive information directly from the time domain). Finally, the performance of AdaESPGL is verified through a series of numerical simulations and the diagnosis of a motor bearing. Results demonstrate its superiority in extracting periodic impulses in comparison to other state-of-the-art methods.