Rolling bearing fault feature extraction using non-convex periodic group sparse method

The detection and evaluation of early faults is currently an important issue in the condition assessment of machinery and has been a challenging problem. In this paper, a novel index and a periodic enhancement group sparse (PEGS) model are sequentially proposed for early fault feature extraction of rolling bearings. Firstly, an index is defined based on the energy operator Gini index to determine the early fault occurrence from the whole-lifecycle vibration signal of the rolling bearing. Second, PEGS is proposed by appending a period estimating approach and a novel non-convex penalty to the non-convex group sparse optimization. It can promote the sparsity of fault signals and its ability to extract fault features. Moreover, the adaptive selection strategy for regularization parameter and group size is discussed. Simulations and two real experimental cases verify that the proposed method can determine the early fault occurrence point and extract the fault features more accurately than other comparison methods.