Speaker
Description
Autonomous beam tuning of a tandem accelerator requires the coordinated control of multiple beamline devices and the continuous processing of beam diagnostic information. In the 2×6 MV tandem accelerator, the beam transport system consists of multiple electrostatic and magnetic elements distributed across several beamlines, resulting in a complex and highly coupled tuning problem. To support the future development of AI-assisted autonomous beam tuning, an integration approach based on the existing EPICS control system is proposed.
The existing EPICS control system provides access to beam diagnostic data, device parameters, and beamline control interfaces. Based on this infrastructure, an AI-assisted tuning framework is proposed as a high-level decision layer that can interact with accelerator devices through EPICS process variables. This approach enables intelligent tuning strategies to be integrated with the existing control infrastructure without requiring fundamental modifications to the underlying machine control system. Historical beam tuning data can provide a basis for developing data-driven models and intelligent tuning strategies. Safety considerations, including parameter validation, operational limits, beam-loss monitoring, and abnormal-operation handling, are incorporated into the proposed architecture to support safe interaction between the AI decision layer and the machine control system.
This work presents an approach for integrating AI-assisted beam tuning with an existing EPICS-based accelerator control system. The proposed framework provides an interface between accelerator physics, machine learning, and accelerator control, and establishes a foundation for future implementation and experimental validation of AI-assisted autonomous beam tuning on the 2×6 MV tandem accelerator.
