Reliability analysis for multi-component systems considering stochastic dependency based on factor analysis. (15th April 2022)

• Record Type: Journal Article
• Authors: Kong, Xuefeng; Yang, Jun; Li, Lei
• Is Part Of: Mechanical systems and signal processing. Volume 169(2022)
• Publication Date: 2022-04-15
• Subjects: Reliability analysis -- Multi-component systems -- Stochastic process -- Stochastic dependency -- Factor analysis -- Expectation-Maximization algorithm; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2021.108754
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: A reliability analysis framework is proposed for multi-component systems. The stochastic dependency among components is captured by the factor analysis. Explicit forms of system reliability functions with common structures are given. An efficient statistical inference method is developed for parameter estimation. Case studies show our method is better in describing system degradation process. Abstract: Reliability analysis for engineering systems with multiple components has gained increasing interest in recent. Most existing works assume that components follow identical degradation models and preset joint distributions or link functions to characterize the degradation interactions. However, distinct degradation characteristics of components are commonly observed in practice. Besides, the degradation interactions are usually complex and diverse, making the preset dependency structures not applicable. Confronted with the diverse degradation characteristics and complex degradation interactions, this paper offers a flexible reliability analysis framework for multi-component systems. First, a stochastic process-based general degradation model combining the Wiener process, Gamma process, and inverse Gaussian process is adopted to describe component degradation processes, and the factor analysis is employed to characterize the degradation interactions by seeking the latent common factors that dominate their interdependency. Thus the assumption of the identical degradation process and preset dependency structure can be relaxed, which enhances the robustness of the method. On this basis, we derive the explicit form of the system reliability function. An efficient Expectation-Maximization algorithm is then utilized for statistical inference to enable a fast computation. Finally, the superiority of the proposed method is demonstrated by two real case studies on lithium-ion battery packs.