Dedicated AI/ML Accelerator Test Facility

Deployment of AI/ML in the Control Room and Operational Challenges for Heavy Ion Accelerators

Not scheduled

20m

48/1-112A/B/C/D - Redwood A/B/C/D (SLAC)

Operation of scientific systems: Accelerators and Tokamaks Operation of scientific systems

Speaker

Brahim Mustapha (Argonne National Laboratory)

Description

Good progress has been made in the online deployment of the AI-ML tools developed over the past few years at the ATLAS facility at Argonne. A dedicated AI-ML interface has been developed and integrated into the control system. It allows the user to simply select the beamline to tune, the task to perform, then hit the “execute” button and watch the results. In addition, a virtual accelerator (VA) model has been developed for the ATLAS linac. Using the same interface and online experience, the VA model allows operator training and offline testing before online deployment. Despite this significant progress, multiple operational challenges remain in the path of a fully autonomous system and digital twin applications. We will highlight some of these challenges; some are general while others are more specific to heavy-ion accelerators and radioactive ion beams.