Error-Weighted Collaborative Dictionary Learning for Rolling Bearings Fault Diagnosis

Abstract

The fluctuating operational environments in rotating machinery systems lead to temporal variations in signal patterns, thereby significantly increasing the complexity of constructing an accurate and robust dictionary in the sparse representation (SR) method. To address this issue, this article proposes a new error-weighted collaborative dictionary learning (EWCDL) method for fault detection of rolling bearings. The approach introduces a data fidelity term that incorporates the local features of the signal, aiming to overcome the inherent assumption of uniform weighting in K-singular value decomposition (K-SVD). Then, a specialized dictionary learning model is developed to achieve collaborative enhancement of the performance of a superior dictionary in conjunction with an inferior one. In addition, to reduce the influence of outliers on the extraction of local features, the density-based spatial clustering of applications with noise (DBSCAN) method was utilized to identify and eliminate prominent outliers. The validity and effectiveness of this approach are verified by simulation analysis and case studies.