LAUSR

Abstract Uncertainties of numerical predictions play important role in assessment of safety margins in nuclear reactors. Critical Flow models are used in the different types of Nuclear Safety Analysis but these are especially important in Loss of Coolant Accidents (LOCA) analyses. In this paper validation of the input decks is performed of all of the Marviken critical flow tests. Next uncertainties of the prediction of these models implemented into TRACE and MELCOR computer codes are analysed and discussed. Results of the global uncertainty analysis are shown in terms of trends of uncertainty values with respect to the pressure, nozzle length, water level, discharge coefficient and hydraulic diameter. Application of the global sensitivity analysis methods allowed ranking of the sources of uncertainties over large spectrum of experimental conditions. This was achieved thanks to usage of Bayesian Integrated Uncertainty and Sensitivity Analysis (BIGUSA) methodology. The analysis revealed that the uncertainties of critical mass flow predictions are predominantly resulting from uncertainties of the hydraulic diameter in both codes. In the TRACE code the hydraulic diameter, nozzle length and discharge coefficient were dominating factors. For MELCOR dominating factors were hydraulic diameter, discharge coefficient and pressure.