Probabilistic Fatigue Software Platform (ProbFat) for Optimizing Life-Cycle Decisions

Abstract

Fatigue is inherently uncertain. Thus, the use of technology for prediction of fatigue damage to ensure the long-term reliability of equipment must embrace such uncertainty. The proposed probabilistic fatigue software platform (ProbFat) accomplishes just that. The stochastic engine behind ProbFat is a collection of Bayesian Decision Networks (BDN) that assess risk to recommend the optimal life-cycle strategy for maximizing overall return (i.e. return = benefit – cost). Benefit is a measure of profitability (e.g. annualized revenue) and cost is a measure of expense (e.g. design, inspection, repair, replacement, failure, et cetera). To optimize the return, the BDN must incorporate all aspects of LCM including: (i) design and construction, (ii) monitoring, inspection, and maintenance scheduling, (iii) identification of damage, (iv) assessment of observed damage, and (v) decision strategies to either run, repair, rerate, replace, or retire the component(s) of interest at the best possible times. BDNs are structured as graphical cause-effect relationships, represented by probabilistic nodes with directional links. Populating conditional probabilities associated with nodes is accomplished through assessment of laboratory and site-specific data, contributions from subject matter experts and simulations from in-house deterministic and stochastic fatigue models. In this paper, two industry examples are used to demonstrate the effectiveness of ProbFat: (i) low cycle thermo-mechanical fatigue of a coke drum, and (ii) very-high cycle vibration fatigue of a piping system. For both applications the fatigue methods available in API 579-1/ASME FFS-1 are utilized for deterministic/probabilistic prediction of fatigue damage accumulation and remaining life. Methods available in API RBI are used for inspection planning and maintenance scheduling.