FRIB-led team earns U.S. Department of Energy machine-learning renewal grant

FRIB, in collaboration with Los Alamos National Laboratory (LANL), was awarded a renewal grant from the U.S. Department of Energy Office of Science (DOE-SC) Office of Nuclear Physics. The two-year renewal grant, titled "Online Autonomous Tuning of the FRIB Accelerator Using Machine Learning," secures funding for the application of machine learning (ML), a subset of artificial intelligence (AI), in pursuit of accelerator operational efficiency. The collaboration is led by Peter Ostroumov, associate director of the FRIB Accelerator Systems Division.

The grant will enable the team to continue the work carried out over the past two years and build on the successes already achieved. This work supports one of FRIB’s guiding paradigms—more time for science and implementation—by applying ML to FRIB’s beam-tuning process. Currently, tuning the superconducting linear accelerator (linac) consumes beamtime for nuclear physics experiments, and FRIB aims to reduce tuning time by about one-third. The results are expected to enable more efficient operations and scientific output. Reduced setup times and improved beam quality will free up more hours for scientific experiments, supporting discoveries in nuclear structure, astrophysics, and fundamental symmetries.

FRIB will reduce beam-tuning time through five main goals:

• improving its ML systems to collect and use data more efficiently;
• developing virtual tools to measure beam performance in real time without interrupting experiments;
• creating ML methods to fine-tune beams that contain multiple charge states;
• improving the stability of key equipment, such as the Electron Cyclotron Resonance (ECR) ion source; and
• building a digital twin—a virtual model of the beam—that can update automatically based on live data. 

This grant will support the development of new ML tools that can be adapted for use with other accelerators. Additionally, it will support workforce development by training MSU undergraduate and graduate students, fostering the next generation of experts in AI and accelerator technology. The grant marks a pivotal step forward toward enhancing FRIB’s capabilities and reinforcing U.S. leadership in high-power/high-beam intensity heavy-ion accelerators. 

FRIB will collaborate with SLAC National Accelerator Laboratory, Thomas Jefferson National Accelerator Facility, and Argonne National Laboratory to enhance open-source, modular software tools—originally developed by multiple researchers—to make advanced algorithms more accessible for online control at Nuclear Physics (NP) Scientific User Facilities.

The focus will be on extending these tools to support hybrid approaches combining reinforcement learning (RL), Bayesian optimization (BO), and system models. The primary application will be improving fast ion beam switching at the Argonne Tandem Linac Accelerator System (ATLAS) and FRIB. These tasks require methods that integrate RL, BO, and system modeling for coarse and fine tuning, as well as generalization to new beam parameters. The team will also use the software to advance hybrid tuning algorithms that enable faster transitions between beam setups. 

Michigan State University (MSU) operates the Facility for Rare Isotope Beams (FRIB) as a user facility for the U.S. Department of Energy Office of Science (DOE-SC), with financial support from and furthering the mission of the DOE-SC Office of Nuclear Physics.

The U.S. Department of Energy Office of Science is the single largest supporter of basic research in the physical sciences in the United States and is working to address some of today’s most pressing challenges. For more information, visit energy.gov/science.