Speaker
Description
Improvements in beam power and current for large-scale accelerators place increasingly strict demands on operational stability. Traditional operation-and-maintenance strategies are becoming insufficient to meet these high-availability requirements. Rapid advances in artificial intelligence offer a new technical paradigm for delivering efficient, reliable accelerator operation. In this study, we combine nonlinear dynamics with modern machine-learning algorithms to develop a robust beam-tuning method. The method is trained in simulation and has been successfully transferred to a real accelerator system. Building on this result, we developed a flexible, AI-driven beam-tuning platform that significantly improves tuning efficiency and operational flexibility. Future work will focus on enhancing algorithm generalization and on advancing an intelligent operation-and-maintenance framework for accelerators.
