Speaker
Description
Precise orbit control in low- and medium-energy beam transport sections is essential for efficient beam delivery in heavy-ion linacs. In the RAON accelerator, the LEBT and MEBT sections are strongly affected by space charge effects, nonlinear beam dynamics, and variations in beam conditions, limiting the performance of conventional correction methods. In this work, we develop a reinforcement learning (RL)-based orbit correction framework supported by a machine learning surrogate model trained on large-scale beam dynamics simulations. The surrogate provides fast predictions of beam centroid responses under varying magnet settings, enabling efficient RL training. We investigate multiple continuous-control algorithms to learn optimal correction strategies in a high-dimensional action space. The trained policies are validated using high-fidelity simulations and demonstrate robust convergence and effective orbit centering across diverse conditions. The framework is applied to both LEBT and MEBT, confirming scalability to more complex beamline configurations. These results highlight the potential of RL-based methods for automated and adaptive orbit correction in heavy ion linac systems.
| I have read and accept the Privacy Policy Statement | Yes |
|---|