Sparsity-Assisted Fault Feature Enhancement: Algorithm-Aware Versus Model-Aware

Abstract

Vibration signal analysis has become one of the important methods for machinery fault diagnosis. The extraction of weak fault features from vibration signals with heavy background noise remains a challenging problem. In this article, we first introduce the idea of algorithm-aware sparsity-assisted methods for fault feature enhancement, which extends model-aware sparsity-assisted fault diagnosis and allows a more flexible and convenient algorithm design. In the framework of algorithm-aware methods, we define the generalized structured shrinkage operators and construct the generalized structured shrinkage algorithm (GSSA) to overcome the disadvantages of $l_{1} $ -norm regularization-based fault feature enhancement methods. We then perform a series of simulation studies and two experimental cases to verify the effectiveness of the proposed method. In addition, comparisons with model-aware methods, including basis pursuit denoising and windowed-group-lasso, and fast kurtogram further verify the advantages of GSSA for weak fault feature enhancement.