An effective health indicator for rolling elements bearing based on data space occupancy. (January 2018)

• Record Type: Journal Article
• Authors: Zhang, Shuai; Zhang, Yongxiang; Li, Lin; Wang, Shengjie; Xiao, Yujie
• Is Part Of: Structural health monitoring. Volume 17:Number 1(2018)
• Page Start: 3
• Page End: 14
• Publication Date: 2018-01
• Subjects: Health indicator -- rolling elements bearing -- multiple general features -- K-means algorithm -- local outlier coefficient; Structural health monitoring -- Periodicals; Structural stability -- Periodicals; Strength of materials -- Periodicals; Nondestructive testing -- Periodicals; Constructions -- Stabilité -- Périodiques; Résistance des matériaux -- Périodiques; Contrôle non destructif -- Périodiques; Electronic journals; 624.17
• DOI: 10.1177/1475921716682432
• Languages: English
• ISSNs: 1475-9217

Abstract

An effective degradation indicator created from the general features is still a hotspot for the condition monitoring of bearing. Since the single general feature–based indicator cannot fully describe the degradation performance, some new indicators based on the characteristics of multiple general features are proposed using pattern recognition and dimensional reduction algorithms. Although most of these indicators can show the incipient fault and good degradation tendency only across the failure stage, also with oscillation, seldom of them can give the whole health condition of the entire useful life. Hence, further research is still needed to reduce the oscillation impact on the sensitivity of the indicator and make remarkable tendency of the degradation. For that purpose, a new indicator is proposed. For the capability of preserving the full structure of data manifold, Orthogonal Neighbourhood Preserving Projections is used for the data preprocessing to obtain the low-dimensional features with the main information remained. Since the preprocessed data can describe different health conditions, K-means clustering algorithm is used to classify the data into different sets according to different health conditions without human intervention. Then, the summation of mean maximum radius of the data set can be used as the health indicator. In order to further decrease the disturbance of the background noise, local outlier coefficient algorithm is used to eliminate the outliers' influence on the effectiveness of the health indicator. The experimental results illustrate that the proposed indicator can show distinct tendency of different health conditions with less influence of the randomness and more sensitive to the incipient fault.