Stochastic methods for the quantification of sensitivities and uncertainties in criticality analyses

This work describes statistical analyses based on Monte Carlo sampling methods for criticality safety analyses. The methods analyse a large number of calculations of a given problem with statistically varied model parameters to determine uncertainties and sensitivities of the computed results. The GRS development SUnCISTT (Sensitivities and Uncertainties in Criticality Inventory and Source Term Tool) is a modular, easily extensible abstract interface program, designed to perform such Monte Carlo sampling based uncertainty and sensitivity analyses in the field of criticality safety. It couples different criticality and depletion codes commonly used in nuclear criticality safety assessments to the well-established GRS tool SUSA for sensitivity and uncertainty analyses. For uncertainty analyses of criticality calculations, SunCISTT couples various SCALE sequences developed at Oak Ridge National Laboratory and the general Monte Carlo N-particle transport code MCNP from Los Alamos National Laboratory to SUSA. The impact of manufacturing tolerances of a fuel assembly configuration on the neutron multiplication factor for the various sequences is shown. Uncertainties in nuclear inventories, dose rates, or decay heat can be investigated via the coupling of the GRS depletion system OREST to SUSA. Some results for a simplified irradiated Pressurized Water Reactor (PWR) UO2 fuel assembly are shown. SUnCISTT also combines the two aforementioned modules for burnup credit criticality analysis of spent nuclear fuel to ensures an uncertainty and sensitivity analysis using the variations of manufacturing tolerances in the burn-up code and criticality code simultaneously. Calculations and results for a storage cask loaded with typical irradiated PWR UO2 fuel are shown, including Monte Carlo sampled axial burn-up profiles. The application of SUnCISTT in the field of code validation, specifically, how it is applied to compare a simulation model to available benchmark experiments is also discussed. SUnCISTT supports the selection of suitable experimental setups for a given application case by calculating and comparing correlation coefficients. We show an example of an experimental series consisting of 21 individual experiments and their cross correlations.