Ensemble-based sensitivity analysis of a Best Estimate Thermal Hydraulics model: Application to a Passive Containment Cooling System of an AP1000 Nuclear Power Plant

Finite mixture models for sensitivity analysis of thermal hydraulic codes for passive safety systems analysis

“…It is worth pointing out that this TH model has been solely used for the purpose of demonstration of the effectiveness of the proposed UA and SA methods in reducing the computational burden of the analysis. Therefore, the intent is not the demonstration and qualification of the TH model itself, but rather the presentation of a systematic approach for SA that is here verified with respect to this benchmark case study, elsewhere analyzed with other approaches [Di Maio et al 2014 a;Di Maio et al 2014c;Yu et al,, 2015]. The input variables used for calculating the PCCS capability of condensing the steam produced are listed in Table 1, together with their distributions chosen based on expert judgment and literature review [Burgazzi 2004], [Zio et al 2008b], [Zio et al 2010b to "Upper value" are reported.…”

“…Best Estimate (BE) codes have been introduced more recently to provide more realistic results and avoid over-conservatism [Zio et al 2010] [10 CFR 50.46]; their use requires the identification and quantification of the uncertainties in the code outputs coming from simplifications, approximations, round-off-errors, numerical techniques and uncertainties in the input variable values [Pourgol-Mohammad 2009]. The quantification of the uncertainties in the output can be demanding in terms of computational cost, because it requires a large number of simulations of the BE-code [de Crécy et al 2008;Di Maio et al 2014a]. …”

A Bayesian ensemble of sensitivity measures for severe accident modeling

“…For this reason, one attempts to perform sensitivity analysis with minimal computational burden, for an effective dimensionality reduction of input variables, which can speed up calculations without diminishing the code capability of representing the real system response. Sensitivity analysis can, indeed, inform on the input variables most affecting the simulation results of the system results [Saltelli, 2002;Pourgol-Mohammad, 2009;Di Maio et al, 2014]. However, most sensitivity analysis methods can turn to be themselves computationally expensive [Secchi et al, 2008;Carlos et al, 2013;Di Maio et al, 2014;Pengfei et al, 2014].…”

“…Sensitivity analysis can, indeed, inform on the input variables most affecting the simulation results of the system results [Saltelli, 2002;Pourgol-Mohammad, 2009;Di Maio et al, 2014]. However, most sensitivity analysis methods can turn to be themselves computationally expensive [Secchi et al, 2008;Carlos et al, 2013;Di Maio et al, 2014;Pengfei et al, 2014].…”

“…The sensitivity results are drawn directly by analyzing Finite Mixture Models (FMMs) built on the results of an uncertainty propagation of the input variables uncertainties onto the simulation code. Ensemble of sensitivity measures can be used [Di Maio et al, 2014], where a set of alternative input variable rankings are aggregated to overcome possible misjudgments of the single sensitivity measure in case of a limited number of code runs. The aggregation methods commonly used are the "majority voting" and "sum" [Kukkonen et al, 2007], where the former consists in assigning the ranking position to each input variable which has been voted by the majority of the calculated sensitivity measures, whereas the latter aggregates by taking the sum of the ranking positions for each input variable provided by the individual sensitivity measures and, then, sorts them with respect to their scores.…”