Application of the Modular Bayesian Approach for Inverse Uncertainty Quantification in Nuclear Thermal-Hydraulics Systems

Abstract

In the framework of BEPU (Best Estimate plus Uncertainty) methodology, the uncertainties involved in the simulations must be quantified to prove that the investigated design is acceptable. The output uncertainties are usually calculated by propagating input uncertainties through the simulation model, which requires knowledge of the model input uncertainties. However, in some best-estimate Thermal-Hydraulics (TH) codes such as TRACE, the physical model parameters used in empirical correlations may have large uncertainties, which are unknown to the code users. Therefore, obtaining uncertainty distributions of those parameters becomes crucial if we want to study the predictive uncertainty or output sensitivity. In this study, we present a Modular Bayesian approach that considers the presence model discrepancy during Bayesian calibration. Several TRACE physical model parameters are selected as calibration parameters in this work. Model discrepancy, also referred to as model inadequacy or model bias, accounts for the inaccuracy in computer simulation caused by underlying missing/insufficient physics, numerical approximation errors, and other errors of a computer code, even if all its parameters are fixed at their "true" values. Model discrepancy always exists in computer models because they are reduced representations of the reality. The consideration of model discrepancy is important because it can help avoid the "overfitting" problem in Bayesian calibration. This paper uses a set of steady-state experimental data from PSBT benchmark and it mainly aims at: (1) quantifying the uncertainties of TRACE physical model parameters based on experiment data; (2) quantifying the uncertainties in TRACE outputs based on inversely quantified physical model parameters uncertainties.