Rolling bearing fault diagnosis using impulse feature enhancement and nonconvex regularization

Abstract

In the past decade, sparse representation has received much attention in the field of fault diagnosis of rotating machinery. However, the effect of sparse representation largely depends on the signal-to-noise ratio (SNR) and constructed dictionary. To address these challenges, an impulsive feature enhancement method is proposed to improve the SNR of weak fault signal of rolling bearing firstly. Utilizing the structure characteristic of impulse response signal, that is, peaks and troughs appear alternately with the same intervals, a structure characteristic matrix is constructed for enhancing the weak impulse feature. Then, a Fused Moreau-enhanced Total Variation Denoising (FMTVD) penalty is developed to avoid the dictionary construction problem and induce the sparsity. The new cost function considers the sparsity of both the fault signal and its differential form, and its solution is derived according to the alternating direction method of multipliers (ADMM). By the two-step strategy, the weak fault features of rolling bearing that submerged in noise are extracted effectively. The performance of the presented method is verified using numerical simulation and practical rolling bearing data.