Speaker
Description
Achieving reliable, fast, and reproducible cyclotron tuning remains a key operational challenge as accelerators move towards increasingly complex beam configurations and higher intensities. To address this, we conducted a two-week experimental campaign at PSI Injector 2 to evaluate the feasibility of applying reinforcement learning (RL) for real-time beam optimization. These experiments represent an important first step towards automated and reliable cyclotron control, demonstrating the potential of RL-based approaches to improve tuning efficiency and operational stability. We will present the experimental setup, methodology, and safety strategies, highlight key results, and discuss lessons learned for future deployment at high-current HIPA operations.
