A method based on clustering fast search for bearing performance degradation assessment

Abstract

Abstract To address the problem of relying on manual experience to determine the number of cluster centers, this paper introduces an approach for evaluating the performance degradation of rolling bearings using the clustering by fast search (CFS) algorithm. Few studies have explored the application of CFS for bearing performance degradation assessment (PDA). Unlike traditional cluster models, such as Fuzzy C-Means, Gustafson–Kessel, Gath–Geva, K-means, and K-medoids, CFS automatically selects the cluster centers according to local density and distance. First, the original vibration signals are processed using local mean decomposition to obtain several product functions (PFs). The top two PFs are selected based on the correlation coefficient and then analyzed using singular-value decomposition to extract the top two singular values (SV1 and SV2). Second, SV1 and SV2 are used as input to the CFS algorithm to identify the cluster centers automatically. Lastly, a confidence value is calculated based on the differences between the sample features and the identified clustering centers for bearing PDA. The results show that CFS outperforms other clustering methods and models, including root mean square, kurtosis, Shannon entropy, approximate entropy, and permutation entropy, detecting early-stage degradation more precisely.

Keywords

• Clustering by fast search
• Performance degradation assessment
• Bearings