A change-point-based Wiener process degradation model for remaining useful life estimation. Issue 3 (1st October 2020)

• Record Type: Journal Article
• Authors: Prakash, Guru; Kaushik, Anshul
• Is Part Of: Safety & reliability. Volume 39:Issue 3/4(2020)
• Page Start: 253
• Page End: 279
• Publication Date: 2020-10-01
• Subjects: Condition-based maintenance -- degradation model -- Wiener process -- change-point -- remaining useful life -- Bayesian inference; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Risk assessment -- Periodicals; Industrial safety -- Periodicals; Industrial safety; Reliability (Engineering); Risk assessment; System safety; Electronic journals; Periodicals; 620.00452
• DOI: 10.1080/09617353.2020.1801165
• Languages: English
• ISSNs: 0961-7353

Abstract

Abstract: This paper presents a change-point-based Wiener process degradation model for remaining useful life estimation of a system. The advantage of using Weiner process for degradation modelling is twofold – first, degradation path needs not to be strictly monotonic, which often occurs when noise is present in the measurements or the degradation measure fluctuates itself; and secondly, a closed-form expression for failure time simplifies the Bayesian computations significantly. In this paper, a unified degradation modelling approach based on the Wiener process is presented, where both gradual degradation and sudden change due to shock events are identified in a unified formulation. The change detection in degradation path is posed as a hypothesis testing and novel test statistics based on log-likelihood function is proposed to accept or reject the null hypothesis. For a given degradation path, first, this test statistic is compared with the pre-set threshold, and an appropriate model (i.e. one or two phases) is selected. The parameters of the chosen model and remaining useful life are continuously updated, where the prior obtained from the historically available run-to-failure degradation paths is integrated with the data obtained from the in situ units in a Bayesian framework. A numerical example and a case study are presented to illustrate the proposed methodology.