Abstract: Facilities providing bright thermal neutron beams are of primary importance for various research topics. At CEA-Saclay, a compact accelerator driven neutron source, SONATE, is investigated in taking advantage of the IPHI accelerator able to deliver a 3 MeV proton beam with an intensity up to 100 mA. To optimize the performances of such a neutron source, it is necessary to maximize the thermal neutron flux while minimizing the contribution of other particles. In this work, optimization has been performed using the Monte Carlo code TOUCANS, a neutron transport code based on Geant4 developed at CEA-Saclay. This latter has been coupled to PROMETHEE, a software allowing multi-objective optimization for many simulation software. In this work the Kriging metamodel based approach is used to optimize a neutron beamdump. To take into account the various constraints, noise on the detection system and radiation protection issues, several beamdump configurations are evaluated. The variation of beamdump parameters makes it possible to identify the set of optimal solutions, the Pareto front. It allows to focus on the set of best choices and to choose wisely the best configurations. After describing the validation of TOUCANS on experimental tests performed from 2016 to 2022, the capability of such an approach will be presented.
Keywords: Monte Carlo, Geant4 Simulation, black-box optimization, Kriging, active learning
