Sensitivity and uncertainty analysis of TRACE Physical Model Parameters based on PSBT benchmark using Gaussian process emulator

For best estimate thermal-hydraulics codes like TRACE and RELAP5, one major source of uncertainties is the inaccuracy of closure laws (correlations) which are used to describe transfer terms in balance equations. These closure laws were originally studied in separate-effect experiments but are implemented in thermal-hydraulics codes and used for different conditions. Thus, the model parameters involved in such closure laws may have significant influences on model outputs but are subject to considerable uncertainties. The aim of this paper is to perform sensitivity and uncertainty study with respect to selected physical model parameters in TRACE. The PSBT steady-state void fraction data is used as Quantity-of-Interest (QoI). Statistics of these physical model parameters are obtained by inverse Uncertainty Quantification (UQ) from a companion paper. Sensitivity Analysis (SA) aims at determining how uncertain input parameters contribute to the variations of outputs. In this paper, the Sobol' indices are used as a sensitivity measure, which quantify how the variation in a QoI can be apportioned to each input factor. Based on the results from SA, several input parameters that have significant effects on the QoI are selected for the UQ process. The prediction uncertainty is evaluated by considering the propagation of uncertainties in selected TRACE physical model parameters. As thousands of TRACE runs are needed for the sensitivity and uncertainty studies, the use of direct Monte Carlo sampling is not a viable solution. We developed Gaussian Process (GP) emulator as a surrogate model of TRACE to reduce the computation time substantially.