Challenge
Modern particle accelerators are complex, requiring extensive human intervention that leads to high operating costs, operational variability, suboptimal experiment optimization, and inadequate data integration. Further, the physical limitations of existing accelerator technologies slow progress in pushing the limits of resolution in space, time, and energy. Transforming accelerator-based facilities into highly efficient, autonomous, and more productive capabilities requires a tight integration of AI with design and operation.
AI Solution
Predicting chaotic beam dynamics, in which small perturbations cascade into major problems, could push AI to develop new capabilities in multi-scale temporal reasoning, physics-constrained learning, and robust uncertainty quantification. AI-driven digital twins that simulate complete beam dynamics in real time could dramatically reduce tuning time. Collectively, facility-based AI will become adaptive and self-updating, significantly boosting performance, efficiency, and scientific output.
Justification
DOE stewards one of the largest suites of accelerator-based experimental facilities in the world, with extensive operational data and a large, highly skilled workforce. The optimization of the nation’s large scale scientific infrastructure through AI-enabled design and the elimination of operational bottlenecks and cost inefficiencies will maximize the nation’s return on current and future infrastructure investments, revealing entirely new paradigms for scientific research through human-AI teaming and accelerating discovery.
National Impact
Accelerator-based facilities have been central to many of the most important discoveries of the 20th and 21st centuries. Integrating AI into accelerator design and optimization will increase the pace of future breakthroughs, enabling a deeper understanding of the universe, the development of new energy and computing technologies, and the creation of new techniques for the diagnosis and treatment of disease.
Aligned Actions from the National Laboratories
National Challenge: Enhancing Particle Accelerators for Discovery
Leading Lab: Lawrence Berkeley National LaboratoryParticle accelerators are strategic infrastructure for science, industry, and medicine—and in the age of AI, how we design and operate them must evolve. To meet this challenge, Lawrence Berkeley National Laboratory created Osprey, an agentic AI platform that streamlines access to accelerator controls and synthesizes information at unprecedented scale. Already installed at eight facilities through the Berkeley Lab-led Multi-Office Accelerator Team (MOAT) project, Osprey is now being co-developed as a community tool by seven national laboratories.
Once fully deployed, Osprey will unlock access to millions of machine parameters and deliver operator queries more than 100 times faster. It will support everything from AI-assisted to fully autonomous operations, including preventive fault detection and automated recovery.
Integrated into the American Science Cloud, Osprey will let researchers across the USA treat the nation's accelerators as a single, intelligent system—dramatically speeding discovery, industrial and medical applications, and the development of next-generation accelerators with higher energy and more precise beams.
National Challenge: Enhancing Particle Accelerators for Discovery
Leading Lab: Thomas Jefferson National Accelerator FacilityPowerful research particle accelerators provide insight into our visible universe, but these sophisticated machines are complex to operate and maintain at peak efficiency.
To meet this national challenge, Thomas Jefferson National Accelerator Facility is applying AI to identify and predict faults before interruptions occur, to reduce damaging radiation, and to balance system components to prevent anomalies. These advances are increasing the uptime and reliability of research particle accelerators, ensuring they operate at peak efficiency now and in the future.
This research is also helping to transform the design of next-generation machines for applications and improvements in medical care, materials science, and energy production.
The Linac Coherent Light Source (LCLS) X-ray laser delivers 93 kHz —almost 100,000 pulses per second. Experiments at higher pulse rates capture ultrafast processes with greater precision and collect data more efficiently. (Jacqueline Ramseyer Orrell/SLAC)National Challenge: Enhancing Particle Accelerators for Discovery
Leading Lab: SLAC National Accelerator LaboratoryParticle accelerators are large, complex systems that power scientific discovery, and AI has the potential to improve their performance, reliability, and scientific throughput. To address this national challenge, SLAC National Accelerator Laboratory is integrating agentic AI workflows into accelerator control systems, enabling progress in autonomous accelerator tuning and complex diagnostic analysis.
Alongside this, SLAC is using AI/ML system modeling, continual learning, and digital twin capabilities, leveraging real-time integration between high-performance computing and the Linac Coherent Light Source (LCLS) to enable faster prediction of beam behavior and, in turn, aid rapid beam customization. These AI-enabled capabilities have improved prediction speeds for digital twins by more than one million times, enabled orders-of-magnitude faster beam customization, and improved key experiment characteristics, such as beam emittance and acceleration efficiency.
These advances will enable higher-quality beams produced on-demand for scientific and commercial applications, spanning computer chip manufacturing to medicine, while accelerating discovery.
Explore Related Challenges
-
Automating at least some parts of the scientific experimental scheme will both increase the volume of data produced for improved AI models and improve the repeatability of experiments.August 21, 2026 -
The convergence of current and emerging AI technology with the growing availability of large, curated datasets may be a tipping point for materials discovery, design, and qualification.October 26, 2026 -
Novel AI could discover new quantum algorithms by automating and optimizing their design and translating them into applications without requiring prior domain knowledge.August 21, 2026