Uncertainty and sensitivity analysis of a PWR LOCA sequence using parametric and non-parametric methods

The Best Estimate Plus Uncertainty (BEPU) approach is being used worldwide for nuclear power plants licensing. This method relies on the use of best estimate models to simulate sequences, evaluating the uncertainties involved. To assess these uncertainties, several methodologies have been developed such as the non-parametric Wilks/Wald method, parametric methods that reconstruct a distribution from the data, or the binomial approach. Additionally, sensitivity analyses can be performed to obtain the correlation of the output-inputs. Finally, a variability analysis of the most influential parameters made to find a combination of parameters that can lead to damage is also useful. In this paper, all previous techniques are described, studied and applied by performing a large Monte Carlo set of simulations of a loss of coolant accident in a pressurized water reactor assessing two figures of merit. The comparison of the different methods show that the most conservative is the Wilks/Wald method; the least conservative is the parametric approach, and in between, the binomial one. The impact of the sample size is also studied for all methods, showing different behaviors for the different approaches.