An alternative comprehensive framework using belief functions for parameter and model uncertainty analysis in nuclear probabilistic risk assessment applications
Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability
In nuclear power plants, probabilistic risk assessment insights contribute to achieve a safe design and operation. In this context, the decision-making process must be robust and uncertainties must be taken into account and controlled. In general, the uncertainties in a nuclear probabilistic risk assessment context can be categorized as either aleatory or epistemic. The epistemic uncertainty, which can be subdivided into parameter and model uncertainties, is recognized to have an important impact on actual results of probabilistic risk assessment. Traditionally, the approach of an epistemic uncertainty analysis in nuclear probabilistic risk assessment often relies on the probabilistic approach in which parameter uncertainty is treated by using an assigned probability distribution, e.g. the log-normal one, and model uncertainty can be taken into account through sensitivity studies. Such an approach has been recognized in several recent researchs to present limitations regarding the impacts of assigning probability distribution in case of rare operating feedback data. In order to overcome such a limitation, in this article we propose a comprehensive approach for uncertainty analysis from the parameter and model uncertainties modeling to the final step of the decision-making process using the Dempster-Shafer theory, which is recognized to be more general than the probabilistic approach. We also show that the traditional probabilistic approach, currently used in probabilistic risk assessment practice, can be totally integrated in this framework. Finally, the proposed framework is illustrated and compared with the traditional approach through a practical example from EDF Nuclear Power Plants probabilistic risk assessment application. Some discussions and conclusions for industrial probabilistic risk assessment contexts will be given.