Sparse Coding Shrinkage in Intrinsic Time-Scale Decomposition for Weak Fault Feature Extraction of Bearings

Abstract

Extraction of the weak fault features under strong background noise is crucial to early fault diagnosis in rolling bearings. However, these fault features (i.e., impulse components) in vibration signals are always submerged and distorted by heavy noise in the early fault phase. Thus, a new method called intrinsic time-scale decomposition (ITD)-based sparse coding shrinkage (SCS) (named ITD-SCS) is proposed as a sparse representation for impulse component extraction from bearing vibration signals. ITD can decompose the signal into a set of proper rotations (PRs) to enable impulse components as prominent as possible. Singular value decomposition (SVD) as the noise prefilter processing of SCS enables the preserved singular values to persist the sparsity of PRs. Finally, ITD-SCS uses SCS to extract the most important impulse components in the reconstruction signal. Thus, ITD-SCS integrates ITD, SVD, and SCS effectively based on their own characteristics for periodic impact extraction. The experimental results on the simulation signals and vibration signals collected from rolling bearings indicate that ITD-SCS is effective for extracting the weak fault features and performs well for bearing fault diagnosis.