Nuclear data uncertainty quantification in fuel depletion calculations.

Spent nuclear fuel characterisation is a topic of major interest in these times of change, when sustainable energy scenarios are shaped. The burden of experimental assessment of spent fuel inventory is too large to allow the application of these techniques to all the fuel used in nuclear reactors. For this reason, fuel assembly models capable of predicting the discharged fuel inventory are needed. The validation of such models and of the capabilities of the codes used is often assessed through benchmark modelling. This consists in models of specific fuel assemblies designed following given specifications, which allows for comparison of the model prediction with the results of the experimental campaigns performed on samples from those assemblies, but also for comparison of different modelling codes and of nuclear data libraries. When it comes to uncertainty propagation, several uncertainty sources should be considered. In the following, after the description and validation of three benchmark models against experimental results, the uncertainty on the cross section data is propagated through such models to the spent fuel inventory. This is done via multivariate statistical sampling, comparing the uncertainty evaluation given by a number of nuclear data libraries. This thesis highlights the need of continuous improvement on the nuclear data covariance information as well as the relevance of identifying the mechanisms of uncertainty buildup. The uncertainty on the spent nuclear fuel inventory builds up through neutron-induced reactions linking some nuclides' concentrations and their uncertainty, but it also builds up during irradiation according to phenomena which relevance changes too. Moreover, the weight of those correlations and phenomena is shown to be dependent on the information stored in the nuclear data library considered in the simulation.