Comparison of Realistic Large Break LOCA Analyses of a 3-Loop Westinghouse Plant Using Response Surface and Statistical Sampling Techniques

In 1988, the U.S. Nuclear Regulatory Commission (NRC) issued a revision to 10CFR50.46 which allows realistic calculations for loss-of-coolant accident/emergency core cooling system (ECCS) analysis. The revision also requires that uncertainties in the analysis method and inputs be considered, such that there is a high level of probability that the ECCS criteria would not be exceeded. Currently, two methodologies are approved by the U.S. Nuclear Regulatory Commission for such calculations. One technique relies on response surface generation and applying Monte-Carlo technique to the regression model to characterize the PCT distribution and quantify the uncertainty. In this method, uncertainty contributors are ranged individually to compute PCT response. Later, these results are combined to generate a response surface. There exists an additional uncertainty since PCT response to any one parameter or group of parameters is not independent from other contributors. The other technique uses non-parametric order statistics where uncertainty parameters are sampled simultaneously for each transient calculation. The required number of transient runs depends on the desired tolerance intervals and is independent of the number of contributing uncertainty parameters. Westinghouse has developed realistic LOCA methodologies based on both techniques. In this study a 3-loop Westinghouse plant is analyzed using each technique. Use of the same input model and same computational tool for thermal-hydraulic response of the system gives the direct comparison between the two uncertainty calculation techniques. For the example used in comparison, the calculated 95th percentile PCT with the response surface technique is 1171 °C (2140 °F) as opposed to 1045 °C (1913 °F) using the statistical sampling technique. The difference between the computed results is mainly associated with the elimination of superposition uncertainty. In addition, there are inherent differences between the two techniques, in their theory and as well as in their application. This study also compares their main features.