Learning a superficial correlated representation using a local mapping strategy for bearing performance degradation assessment

Abstract

As a prime technique for proactive maintenance, bearing performance degradation assessment (PDA), which aims to build a health index (HI) to assess the performance degradation process, has drawn more and more attention in recent years. To construct an HI of high quality, we propose a novel and robust fuzzy c-means (FCM) model, based on locally linear embedding (LLE), that aims to learn a superficial correlated representation using a local mapping strategy. First, a great mass of commonly used features from the time-domain, the frequency-domain, and the time–frequency domain are extracted from the original vibration signature. Features are then implemented to obtain the initial dimensions by maximum likelihood estimation (MLE). Subsequently, local mapping produced by LLE with the initial dimensions extracts the underlying manifold structure from all the remaining features, and a superficial correlated representation is obtained, generated from the space expanded by the features. Finally, an HI based on the subjection of the FCM model is used to assess the bearing degradation process. To validate the superiority of the proposed method, it is compared with three advanced PDA models through experiments on three public datasets. A comparison of the proposed method with those of the other studies confirms the potential of MLE-LLE as an effective feature-fusion tool for the PDA of bearings.