Fermilab is leading a collaboration with six other national laboratories to incorporate artificial intelligence into the design, construction, and operation of particle accelerators, a project conducted as part of the Department of Energy’s Genesis mission. These highly complex machines, made up of tens of thousands of interdependent devices, are essential to advances ranging from cancer treatment to nuclear fusion research, and most importantly, to eliminating “forever chemicals” from our water supplies. The multi-office particle accelerator team, known as MOAT, aims to create integrated AI systems to increase efficiency and accelerate scientific discovery. “MOAT’s ultimate vision is to fully integrate AI into accelerator design, construction, and operation, fundamentally changing the pace of discovery and resulting innovation,” said Jonathan Jarvis, MOAT collaborator and director of Fermilab’s Accelerator Research Division.
Genesis Mission powers multi-lab AI collaboration
The scale of this effort reflects a broader national commitment to AI-powered scientific discovery. MOAT is part of the Transformational AI Models Consortium (ModCon), which leverages DOE resources to deploy self-improving AI models. Researchers at Berkeley, Argonne, Jefferson, Oak Ridge, SLAC, and Brookhaven are all contributing to this unified approach, moving away from traditional silos of AI prototype development within individual labs. “Typically, each of our labs would develop their own standalone prototypes,” said Torsten Herert of Berkeley Lab. “The Genesis mission really forced our community to come together to jointly develop and deploy this new AI software.” Fermilab’s Accelerator Technology Test Facility, FAST/IOTA, serves as the primary demonstrator of these AI tools, enabling testing across a variety of accelerator types and particle beams.
Initial demonstrations of MOAT’s research, including Osprey tools that accelerate the task by a factor of 100, have already been submitted to the DOE Office of Science, and the team is developing a digital twin of the accelerator complex for virtual diagnostics and beam conditioning. Jean-Luc Bey, director of the Advanced Modeling Program at Lawrence Berkeley National Laboratory, explained that the goal is to accelerate discovery and expand knowledge faster than would otherwise be possible.
MOAT integrates AI throughout the accelerator lifecycle
The pursuit of increasingly sophisticated particle accelerators, essential tools for advances across multiple scientific fields, is benefiting from a concerted effort to integrate artificial intelligence across the lifecycle. In addition to optimizing existing systems, the Multi-Office Particle Accelerator Team (MOAT) is establishing a unified AI framework aimed at fundamentally changing the speed of scientific discovery. This approach is necessary due to the sheer size of these machines. State-of-the-art particle accelerators are comprised of tens or hundreds of thousands of interdependent devices, requiring complex management. MOAT’s first demonstration showcased the Osprey AI tool, which achieved 100x speedups on certain tasks through the use of AI agents, autonomous systems capable of reasoning and independent action.
MOAT’s ultimate vision is to fully integrate AI into the design, construction, and operation of accelerators, fundamentally changing the pace of discovery and resulting innovation.
Jonathan Jarvis, Fermilab
Osprey AI agent achieves 100x task acceleration
Beyond basic physics, these advances promise to address pressing real-world challenges, including the removal of “forever chemicals” from water supplies, and will benefit from enhanced capabilities of these powerful machines. A recent demonstration demonstrated the initial success of MOAT’s Osprey AI tool, achieving 100x speedups on certain tasks. This is a feat made possible by the use of autonomous AI agents. These agents are capable of reasoning and independent action and represent a key component of the Multi-Office Particle Accelerator team’s long-term vision. AI systems are trained on decades of operational knowledge, including documented solutions to accelerator failures, enabling rapid problem resolution that is traceable to cause. MOAT is developing an interconnected “digital twin” of the accelerator complex, enabling virtual diagnostics and testing before implementing changes to the physical hardware.
MOAT’s goal is to speed up the way we discover and extend knowledge in fundamental physics, chemistry, biology, materials science, and more than any other method. ”
Vai
FAST/IOTA Facility Validates AI-Driven Accelerator R&D
FAST/IOTA, Fermilab’s accelerator technology test facility, currently serves as a key validation point for the Multi-Office Accelerator Team (MOAT)’s ambitious integration of artificial intelligence into particle accelerator research and development. MOAT’s long-term vision extends to leveraging decades of operational knowledge accumulated from facilities like Fermilab. This approach, combined with the development of an interconnected “digital twin” of the accelerator complex, is expected to save billions of dollars and years of effort while dramatically improving performance and expanding the scope of research faster than previously possible, from producing medical isotopes to removing “permanent chemicals” in water.
MOAT’s ultimate vision is to fully integrate AI into the design, construction, and operation of accelerators, fundamentally changing the pace of discovery and resulting innovation.
