Genetic algorithm for multilayer shield optimization with a custom parallel computing architecture

This paper introduces a novel architecture for optimizing radiation shielding using a genetic algorithm with dynamic penalties and a custom parallel computing architecture. A practical example focuses on minimizing the Total Ionizing Dose for a silicon slab, considering only the layer number and the total thickness (additional constraints, e.g., cost and density, can be easily added). Genetic algorithm coupled with Geant4 simulations in a custom parallel computing architecture demonstrates convergence for the Total Ionizing Dose values. To address genetic algorithm issues (premature convergence, not perfectly fitted search parameters), a Total Ionizing Dose Database Vault object was introduced to enhance search speed (data persistence) and to preserve all solutions' details independently. The Total Ionizing Dose Database Vault analysis highlights boron carbide as the best material for the first layer for neutron shielding and high-Z material (e.g., Tungsten) for the last layers to stop secondary gammas. A validation point between Geant4 and MCNP was conducted for specific simulation conditions. The advantages of the custom parallel computing architecture introduced here, are discussed in terms of resilience, scalability, autonomy, flexibility, and efficiency, with the benefit of saving computational time. The proposed genetic algorithm-based approach optimizes radiation shielding materials and configurations efficiently benefiting space exploration, medical devices, nuclear facilities, radioactive sources, and radiogenic devices.