Uncertainty Study of the In-Vessel Phase of a Severe Accident in a Pressurized Water Reactor

A comprehensive uncertainty analysis in the event of a severe accident in a two-loop pressurized water reactor is conducted using an uncertainty package integrated in the ASYST code. The plant model is based on the nuclear power plant (NPP) Krško, a Westinghouse-type power plant. The station blackout scenario with a small break loss of coolant accident is analyzed, and all processes of the in-vessel phase are covered. A best estimate plus uncertainty (BEPU) methodology with probabilistic propagation of input uncertainty is used. The uncertain parameters are selected based on their impact on the safety criteria, the operation of the NPP safety systems and to describe uncertainties in the initial and boundary conditions. The number of required calculations is determined by the Wilks formula from the desired percentile and confidence level, and the values of the uncertain parameters are randomly sampled according to appropriate distribution functions. Results showing the thermal hydraulic behaviour of the primary system and the propagation of core degradation are presented for 124 successful calculations, which is the minimum number of required calculations to estimate a 95/95 tolerance limit at the 3rd order of the Wilks formula application. A statistical analysis of the dispersion of results is performed afterwards. Calculation of the influence measures shows a strong correlation between the decay heat and the representative output quantities, which are the mass of hydrogen produced during the oxidation and the height of molten material in the lower head. As the decay heat increases, an increase in the production of hydrogen and the amount of molten material is clearly observed. The correlation is weak for other input uncertain parameters representing physical phenomena, initial and boundary conditions. The influence of the order of the Wilks formula is investigated and it is found that increasing the number of calculations does not significantly change the bounding values or the distribution of results for this particular application.