Uncertainty quantification and data assimilation framework towards the validation of MCNP for LFR applications

The qualification of computational tools for nuclear reactor design requires rigorous verification, validation, and uncertainty quantification. Validation and uncertainty quantification depend on the availability of experimental databases. While for traditional technologies the long operational experience has provided abundant and well-documented measurements, in the context of Lead-cooled Fast Reactors (LFRs), where such data remain limited, international benchmark projects play an essential role. This work focuses on the uncertainty quantification to support the qualification of the MCNP Monte Carlo code for LFR applications, with reference to the ALFRED demonstrator. During this PhD project, a set of experiments representative of ALFRED was selected from the International Handbook of Evaluated Reactor Physics Benchmark Experiments (IRPhE) maintained by OECD-NEA, and then simulated to evaluate the MCNP accuracy in computing criticality (keff) and reactivity effects, including control rod worth, sodium void reactivity, and isothermal temperature coefficient. A dedicated Python-based framework was developed to perform post-processing, uncertainty propagation, and data assimilation. The uncertainty quantification is based on the use of sensitivity coefficients, which were calculated by MCNP and subsequently combined with the covariance matrix to propagate nuclear data uncertainties. In addition, the Generalized Linear Least Squares (GLLS) data assimilation tool was implemented to evaluate the potential improvements of integrating experimental information to obtain a more accurate prediction for the ALFRED keff estimation. Both methodologies were verified against established codes. The results demonstrate good agreement with experimental data, in most cases falling within one standard deviation of total uncertainty. Nuclear data were identified as the dominant uncertainty source across all cases, while statistical contributions were generally negligible. The results of the GLLS analyses also revealed the potential for a significant improvement in the evaluation accuracy. This opportunity suggests further investigations to extend the scope and confirm the preliminary results here reported before inclusion in a validation dossier.