Physics-Informed Multi-Objective Optimization of RFQ Cross-Sectional Shape Using a Genetic Algorithm

摘要: In the low-energy front end of modern linear accelerators, the radio-frequency quadrupole (RFQ) plays an indispensable role. RFQ design involves many multi-objective optimization problem. However, directly applying multi-objective optimization algorithms (MOGAs) to a full-model cavity is computationally impractical. RFQ internal electromagnetic-filed distribution mainly affected by the cross-sectional geometry, which provides a efficient means of significantly improving the cavity performance by optimizing the cross section. This paper presents a multi-objective genetic-algorithm framework that combines a physics-informed frequency model with CST electromagnetic simulations to optimize the RFQ cross-sectional geometry. The proposed framework is then applied to an existing medical heavy-ion RFQ. Compared with the initial design, the optimized cavity achieves an 8.7% increase in the Q value and a 16.7% reduction in RF power loss. A sensitivity analysis is subsequently performed using the simulation data generated during the optimization process to assess the effects of individual geometric variables on the objective functions. The sensitivity results demonstrate that the resonant frequency can be fine-tuned to the design value while keeping the other objective remain near the optimal solution. The full-model cavity simulation results show that the physics-informed combination of the multi-objective genetic algorithm with CST is an effective approach for RFQ cross-section optimization, and this method can be adapted to RFQ cavities operating at different frequencies.