A Mission-Reliability-Oriented Health Prognosis Approach for Manufacturing Systems Considering Operational Uncertainty

• Published in: IEEE transactions on reliability Vol. 73; no. 1; pp. 650 - 663
• Main Authors: Cai, Yuqi, He, Yihai, Shi, Rui, Li, Jiayang
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.03.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Deep reinforcement learning; Deep reinforcement learning (DRL); Degradation; Ferrite; health prognosis; Machine learning; Manufacturing; manufacturing system; Manufacturing systems; mission reliability; operational uncertainty; Phase shifters; Production; Prognosis; Prognostics and health management; Reliability; Reliability theory; Risk management; uncertain production risk process; Uncertainty
• ISSN: 0018-9529; 1558-1721
• DOI: 10.1109/TR.2023.3311196

Considering the objective of production task accomplishment, the health state of manufacturing systems is contingent on three key operational factors: 1) task requirements, 2) equipment performance, and 3) product quality. Throughout the system lifecycle, these factors are subject to varying degrees of uncertainty resulting from the 5M1E (i.e., men, machine, material, method, measurement, and environment) fluctuations. Such operational uncertainties pose considerable challenges for decision-makers to obtain instructive results from conventional health prognosis methods. Consequently, based on a conceptual investigation of operational uncertainty and its implications on system health, this study proposes a novel health prognosis approach for multistate manufacturing systems from a mission reliability perspective. In this approach, the operational uncertainties are 1) categorized according to objective and subjective perspectives, 2) represented in fuzzy and probabilistic metrics, and subsequently 3) ensembled in a customized uncertain production risk process-based prognosis model for health prediction. To further eliminate the influence of subjective perturbations on the results, we adaptively assessed the boundaries of fuzzy parameters within this model by a deep reinforcement learning algorithm in lieu of expert elicitations. Finally, to validate the results, the proposed approach is applied to an industrial case study involving a ferrite phase shifter manufacturing system.