New Approach for Bearing Fault Diagnosis Based on Fractional Spatio-Temporal Sparse Low Rank Matrix Under Multichannel Time-Varying Speed Condition

Rolling bearings are usually operated under variable speed conditions in industrial engineering; meanwhile, complex time-varying modulations and periodical transient impulses induced by localized failure of bearings are always accompanied by heavy background noise and other uncorrelated structure vibrations, resulting in challenges in fault diagnosis. To extract weak fault signatures from multichannel vibration data under time-varying speed conditions, in this article, a novel approach based on fractional spatio-temporal sparse low rank matrix (FST-SLRM) is proposed for the first time. First, the collected multichannel vibration data in time domain are resampled and then the corresponding angular-domain data will be obtained. Second, the periodical transient impulses related to localized failure in angular-domain are decomposed and isolated using the proposed FST-SLRM approach, in which the spatio-temporal feature of multiple channel signals can be captured using bitemporal matrices and bispatial matrices, and the sparsity and location of fault information in spatial and temporal domain can be determined. Last, bearing fault features can be extracted using order envelope spectrum (OES) method. Compared with three start-of-the-art benchmarks (i.e., the pure analysis prior (AP) algorithm, wavelet and total variation (WATV) sparse regularization algorithm, and nonconvex sparse regularization total variation (NSRTV) algorithm), the superiority and applicability of the proposed FST-SLRM approach are experimentally demonstrated by two multichannel bearing vibration datasets under variable speed conditions.