Multi-objective evolutionary optimization of a graphite subcritical pile for central thermal neutron flux enhancement

Subcritical assemblies inherently exhibit complex, high-dimensional, and non-convex configuration spaces due to the coupled placement of fuel, moderator, and neutron sources, making traditional optimization approaches inefficient. In this work, we present an optimization framework for improving the thermal neutron flux characteristics of subcritical graphite piles operating in mixed neutron spectra and systematically explore candidate configurations that enhance thermal flux quality while suppressing epithermal and fast components. We employ a (μ + λ) evolutionary strategy over 250 generations with a population of 100 individuals per generation, incorporating custom operators and symmetry constraints to reduce dimensionality and ensure physically meaningful neutron distributions. Fitness evaluation was performed using the OpenMC Monte Carlo neutron transport code, tightly coupled to the optimization workflow through its Python API and implemented using the DEAP evolutionary computing framework. The algorithm explored more than 50,000 feasible configurations of the Purdue University Subcritical Pile and identified multiple Pareto-optimal solutions balancing two competing optimization objectives: thermal neutron flux magnitude and spectral purity, defined as the ratio of thermal-to-total neutron flux. We were able to identify candidate configurations that increased the thermal-to-total neutron flux ratio up to 256% while enhancing the thermal neutron flux by as much as 93%. Although demonstrated using the Purdue pile, the proposed framework is generalizable to other subcritical graphite systems and provides a scalable approach for multi-objective configuration optimization in nuclear research and irradiation environments.