Investigation Of Uncertainty Quantification Method On Be Models And Application Of Reflood Model Evaluation

Best Estimate Plus Uncertainty(BEPU)method is the current trend in safety analysis and license application of Nuclear Power Plant(NPP)during recent years.So far,many methodologies of uncertainty assessment had been developed for NPP design and licensing.Distribution of uncertainty sources should be known for these methods to obtain the uncertainty of important output responses through propagation.Uncertainty methods have now a good maturity for the evaluation of uncertainty on accident transient.However,important efforts have to be done on compulsory step which is the quantification of input uncertainties.Most of the system codes don't provide enough uncertainty information on the internal physical models.Especially for some model parameters which could not be directly measured through experiment,the quantification is often performed by subjective engineering judgement with conservative estimation.Thus,it is indispensable to develop more objective and accurate methods for uncertainty quantification of physical models using experiment data and statistical theory.Considering the international trend and requirement of uncertainty analysis,this research carried out through three aspects of work:(1)developing uncertainty quantification methods for unobserved model parameters,(2)applying methods to assess physical models,(3)improving models according to the results of evaluation.In terms of the application example for model assessment,reflood phenomena happened during loss of coolant accident(LOCA)is chosen in this study.On the one hand,various flow regimes and heat transfer modes are related to reflood and the phenomena contain many unmeasurable parameters,thus difficulties are apparent for uncertainty quantification.On the other hand,the simulation of BE code for reflood under low inlet flow rate conditions is inaccurate,related models need to be improved.The main work accomplished in this paper includes:1.Developed new methods of uncertainty quantification to assessing unobservable physical model parameters of BE code.Considering the “error” between measurable experiment data and computation outputs,the uncertainty of input model parameters could be obtained through solving inverse problem.Deterministic analysis is used for the situation that relationship between input model parameters and observable output reponses is nearly linear.Expectation-Maximization(EM)based algorithm is applied to deduce the mean and variance of unknown model uncertainties.For nonlinear relationship or more general situation,stochastic analysis is considered.Combined with surrogate models which are built by radial basis function(RBF)neural network and algorithm of adaptively increasing training samples,MCMC based method is used to sample the model parameters in posterior space and obtain mean and variance.2.Six tests of FEBA facility are selected for the application of developed method to evaluate the uncertainty of related model parameters of typical thermal hydraulic code under reflood phenomena.Sensitivity analysis is performed to determine most influential models to important responses.Then,applications are carried out for both developed methods.Finally,comparisons and discussions are accomplished for these two evaluation methods.3.Models of film boiling heat transfer from wall to fluid,dispersed flow interfacial friction model are studied,improved and verified based on the evaluation results of uncertainty and sensitivity analysis.Considering the investigation of latest researches,the wall heat transfer model of film boiling is improved by taking the influence of droplet behavior into account.As for the interfacial friction model of dispersed flow which epistemic uncertainties are hard to eliminate,mean value got from MCMC based algorithm is applied to calibrate the interfacial friction.FEBA and FLECHT SEASET experiments are calculated for validation,and the modified models prove better predictions for both PCT and quench time.The uncertainty quantification methods studied in this paper has academic significance and application value for validation part of China's NPP software self-reliance and perfection of BEPU method.