Two novel mixed effects models for prognostics of rolling element bearings. (15th January 2018)

• Record Type: Journal Article
• Authors: Wang, Dong; Tsui, Kwok-Leung
• Is Part Of: Mechanical systems and signal processing. Volume 99(2017)
• Page Start: 1
• Page End: 13
• Publication Date: 2018-01-15
• Subjects: Bearing degradation -- Mixed effects model -- Statistical modeling -- Remaining useful life; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2017.06.004
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Two novel mixed effects models are proposed for bearing prognostics. Each of the two models describes Phases I and II of bearing degradation simultaneously. The parameters B = ( α β ) ′ are not assumed to be independent in both their priori and posteriori. It is mathematically shown that the two models are more general than the previous works. Brownian motion error is better than normal random error to model bearing degradation. Brownian motion error results in positively skewed remaining useful life distributions. Abstract: Rolling element bearings are widely used in various machines to support rotating shafts. Due to harsh working environments, the health condition of a bearing degrades over time. A typical bearing degradation process includes two phases. In Phase I, the health condition of the bearing is in normal and it exhibits a stable trend. In Phase II, the health condition of the bearing degrades exponentially. To analytically model the bearing degradation process, two novel mixed effects models are proposed in this paper. Each of the two mixed effects models is able to simultaneously model Phases I and II of the bearing degradation process. The main difference between the two mixed effects models is that different error assumptions including multiplicative normal random error and multiplicative Brownian motion error are respectively considered in the two mixed effects models. Consequently, two different closed-form distributions of bearing remaining useful life are derived from the two mixed effects models via Bayes' theorem once real-time bearing condition monitoring data are available. 25 sets of bearing degradation data collected from an experimental machine are used to illustrate how the two mixed effects models work. Comparisons are conducted to show that the mixed effects model with multiplicative Brownian motion error results in lower prediction errors than the mixed effects model with multiplicative normal random error for bearing remaining useful life prediction.