Nuclear data uncertainty propagation in the ARIANE GU3 burnup model using SANDY: Comparison between a FA and a pincell model

Abstract

Nuclear data uncertainties taken from the general-purpose evaluated libraries JEFF-3.3, ENDF/B-VIII.0 and JENDL-4.0u are propagated through a depletion model of the ARIANE GU3 sample using the SANDY stochastic sampling code combined with the Monte Carlo burnup code SERPENT-2. This approach enabled an accurate characterization of the uncertainty in many nuclide concentrations, for which measurements exist from post-irradiation experiments.

Stochastic sampling methods for uncertainty propagation in Monte Carlo burnup calculations are notoriously computationally expensive. To address this, the contribution of nuclear data uncertainties to the model response was assessed independently of Monte Carlo uncertainties using a methodology based on conditional estimators. Interestingly, unlike best-estimate values, uncertainty estimates were found to be rather independent of model simplifications. This was demonstrated by comparing uncertainty results for the GU3 fuel assembly model and for a simplified pincell model. The possibility to transpose uncertainties between such models suggests that high assay data accuracy is not strictly necessary for uncertainty analyses. Finally, the variance decomposition analysis revealed gaps in the uncertainty datasets of major nuclear data libraries, leading to an underestimation of total uncertainties in burnup calculations.