ETD

• active learning
• gaussian process
• marine propulsion reactor
• multi-objective optimisation
• NSGA-II
• OpenMC
• small modular reactor
• surrogate model

This thesis applies surrogate-assisted multi-objective optimisation to a compact pressurised water reactor core fuelled with low-enriched UO2, modelled with the OpenMC Monte Carlo code. The reference is a 48 MWth land-based prototype of a maritime propulsion reactor, whose core data are not public, so the model was built from open data and declared assumptions. Enrichment, gadolinia loading, and reflector thickness were varied on a core of 32 assemblies. Each depletion evaluation takes about 20 minutes, so a Gaussian-process surrogate searched by the NSGA-II genetic algorithm in an active-learning loop selected the designs to simulate over successive optimisation campaigns. The early campaigns maximised cycle length and minimised radial peaking, and their designs exceeded the five-year refuelling requirement while relying on deeply inserted control rods. The problem was then reformulated with the cycle length as a constraint, minimising the peaking factor and the critical boron concentration. This produced a set of designs that meet the refuelling requirement with less soluble boron than the earlier designs, and showed that the boron demand falls as the gadolinia loading rises. The search used a two-dimensional model at fixed temperatures, and the designs of the final set are verified on a three-dimensional model of the core.