Response Surface Modeling of a Large Break Loss of Coolant Accident With Comparisons to Order Statistics Approaches

NUREG/CR-5249 “Quantifying Reactor Safety Margins Application of Code Scaling, Applicability, and Uncertainty Evaluation Methodology to a Large Break, Loss-of-Coolant Accident” provides the general methodologies to be used in the development of realistic loss of coolant safety analyses. The objective of this paper is to start with NUREG/CR-5249 and develop a modified methodology. The modified approach will include a response surface, model adequacy checks, and development of the 95/95% confidence peak clad temperature cumulative distributions function. The response surface model will then be used to develop simulated results and conclusions about the order statistics best estimate approach. All work is conducted using a verified safety analysis input deck and RELAP as the thermal hydraulic best estimate analysis code. The objective of the order statistics comparison is to investigate the number of cases in which the maximum PCT, in a simulated order statistics approach, falls below the 95^th percentile value of the distribution and to assess the standard deviation in the maximum peak clad temperature of order statistics sets. Although order statistics may be a more economic approach to satisfying regulatory requirements, response surface models have several benefits that can complement the use of order statistics. The primary benefit is the insight gained into which parameters are most important in determining the peak clad temperature. This is of particular value to the licensee in convincing the regulator that its analysis is robust. The disadvantage is the number of runs required to develop the models. If we examine the main effects, the most significant input parameter is pipe break size. In support of a proposed modification to 10CFR50.46, the U.S. Nuclear Regulatory Commission undertook an expert elicitation to assess the change in frequency of pipe break accidents as a function of break size. The result of that elicitation was a probability density function that decreases approximately as (pipe diameter)^−1.5 in the region of large pipe diameters. Because break diameter is shown to be such a large contributor to PCT by the response surface, it is evident that calculated PCT could be substantially reduced if credit were given for this form of the uncertainty distribution rather than for the flat distribution used in the analysis (and industry).