Estimation of system failure probability uncertainty including model success criteria

Uncertainty is an important part of formal decision making. Acknowledging this point, modem probabilistic risk assessments (PRAs) typically include uncertainty on parameters of the logic models used in the assessment. What is not addressed in many PRAs is the incorporation of uncertainty specific to the logic models themselves. Our paper investigates the uncertainty in the unreliability of a nuclear power plant auxiliary feedwater system. Herein, we focus on two types of uncertainty, namely "aleatory" and "epistemic." The parameter uncertainty (epistemic) for the model was determined from actual operational failure data from 1987 to 1995. Then, the system uncertainty was estimated and plotted for both aleatory and epistemic contributions. The results demonstrate that the epistemic uncertainty dominates the overall unreliability uncertainty for the auxiliary feedwater system. Factors such as the system success criteria (a type of epistemic uncertainty) can have dramatic impacts on the overall uncertainty outcome. As a result, PRA analysts may consider investing resources specifically focused on modelling uncertainty issues.