Investigation of uncertainty quantification method for BE models using MCMC approach and application to assessment with FEBA data

Quantifying the uncertainty contributors of Best Estimate (BE) Thermal Hydraulic (TH) codes has been getting more and more attention in safety analysis of nuclear industry during recent decades. Yet for evaluation of intrinsic physical models which may not be readily measured, the quantification process is usually subjective and inaccurate. This paper investigates the statistical methodology in order to get the probability density function (pdf) of model parameters more objectively based on observed experimental responses. The simplification of mathematical model is described for the parameter estimation, and the solution using Markov Chain Monte Carlo (MCMC) algorithm is demonstrated. As the direct evaluations are computationally intensive, surrogate models using Radial Basis Function (RBF) are constructed to substitute the complex forward calculations. And to efficiently improve the accuracy of the surrogate model, an adaptive approach based on cross-entropy minimization to densify training samples at space of posterior pdf is applied. As an application, uncertainties of model parameters related to reflood phenomena implemented in RELAP5 code are quantified. It is indicated that the developed method which is independent of BE codes is feasible and efficient to apply. Through the check of uncertainty propagation, it proves that the uncertainty bands can envelope most of the experiment measurements with an advantage of accuracy. The model calibration by posterior mean value also presents a good improvement of calculations.