Design improvement of a minichannel beam dump wing through AI-driven genetic algorithms

At the FRIB (Facility for Rare Isotope Beams), a minichannel beam dump with 2 mm x 7mm water passages can dissipate about 20 kW of heat through high-velocity flow. Its central region features a 6°-tilted CuCrZr absorber plate with an explosion-bonded 2219 Al cooling channel for efficient convection. The wing-section surface, angled from 6° to 90° next to the absorber plate, experiences higher heat fluxes and maximum temperatures, becoming more severe as the beam size increases. Optimizing the beam dump’s geometry is therefore essential to accommodate variable beam sizes and absorb higher beam power. Even small angle changes in the surface orientation significantly affect the heat distribution, which raises the temperature of the CuCrZr plate and complicates global optimization in a limited space. To address this, a Genetic Algorithm (GA) explores a wide range of designs to prevent premature convergence, while a Soft Actor-Critic (SAC) method fine-tunes the details. A 2D finite volume method is employed to reduce the computational load. By merging the GA’s broad search with SAC’s adaptive refinement, the design can withstand higher heat loads under diverse beam conditions.