Artificial intelligence (AI) is a catalyst helping to accelerate scientific research and fast-track new scientific discoveries and breakthroughs. By identifying meaningful trends in large datasets, predicting outcomes based on data, and simulating complex scenarios, AI and its subset, machine learning (ML), which uses mathematical data models to help a computer learn without direct instruction, are transforming many fields, including particle accelerator science and applications.
Researchers in the Accelerator Technology & Applied Physics (ATAP) Division at Berkeley Lab are harnessing the game-changing potential of AI/ML as an accelerator design tool and for enhancing control systems to improve the performance of today’s accelerators and advance the development of future accelerator technologies.
For example, ATAP’s Advanced Modeling Program (AMP) develops and uses AL/ML techniques to dramatically reduce the time needed to model the intricate and ultrafast processes that accelerate particle beams to increasingly higher energies and intensities. To accomplish this, AMP uses high-fidelity simulation data to train so-called “ML-surrogate models” that can be used as flexible elements in future conventional-plasma hybrid accelerators—a crucial step toward improving the performance of today’s accelerators and colliders and supporting advanced accelerator modeling, design, and operation.
The AMP team is also working on new ML-based models that combine simulation and experimental data to provide real-time feedback for tuning and optimizing accelerator and beamline parameters during accelerator experiments. They are also actively training the next generation of particle accelerator physicists in AI/ML techniques; for example, AMP researchers, in partnership with colleagues from the SLAC National Accelerator Laboratory, teach a course on optimization and ML techniques for accelerators for the U.S. Particle Accelerator School. This national school provides graduate-level training and workforce development in particle beam science and associated accelerator technologies.
The Advanced Light Source Accelerator Physics Program in ATAP is also developing and applying AI/ML techniques to improve the performance of existing accelerators, designing future accelerators like the Advanced Light Source (ALS) Upgrade, and improving their operation.
Thorsten Hellert, a research scientist in the BACI Program at ATAP, adjusts the settings on one of the controls at Berkeley Lab’s Advanced Light Source. (Credit: Thor Swift/Berkeley Lab)
The program is a pioneer in developing and employing AI/ML in synchrotrons (like the ALS) for anomaly detection, online nonlinear optimization, and integrating various heterogeneous online data sources for better problem detection and resolution.
The ALS provides multiple, extremely bright sources of intense and coherent short-wavelength light for use in scientific experiments by researchers worldwide. Its Accelerator Physics Group was in the global accelerator community to deploy ML-based optimization on a synchrotron light source during routine user operations, significantly improving the performance of the ALS. The work led to Slavo Nemsak, a staff scientist in the ALs, winning the 2023 Klaus Halbach Instrumentation Award.
In another example of the innovative use of AI/ML, ATAP’s Berkeley Accelerator Controls and Instrumentation (BACI) Program employs AL/ML tools and techniques to control particle beams and lasers more efficiently. These AI/ML-based adaptive control algorithms automatically compensate for real-time changes to accelerator beams and other components, like magnets, outperforming contemporary beam control systems. This approach is better at understanding problems and using physics to create a response.
(l-r) Qiang Du, Dan Wang, and Tong Zhou make adjustments to the optics of a laser in the BACI laser lab at Berkeley Lab. (Credit: Thor Swift/Berkeley Lab)
BACI also uses ML techniques to develop novel fiber laser technologies for driving laser-plasma accelerators (LPAs), which promise more powerful and compact machines that are cheaper to build and operate than current technologies.
Researchers at ATAP’s BELLA Center are integrating AI/ML into the operation, optimization, and diagnosis of LPAs and their light source derivatives. The BELLA team, for example, employs AI/ML for rapid data analysis, driving active stabilization components, and mapping out key correlations between laser, plasma, and particle beam parameters.
To improve the performance of the superconducting magnets used to accelerate, guide, and shape the accelerator’s particle beam, the Superconducting Magnet Program (SMP) at ATAP is applying automated AI/ML analysis of microscopy images to detect early signs of sudden and unpredictable losses—a phenomenon referred to as quenching—in the magnets’ superconductivity. Quenching can generate temperatures high enough to destroy the magnets, costing millions of dollars in damage, and its early detection is essential to safeguarding the magnets.
By leveraging the transformative potential of AI/ML, researchers at ATAP are helping to extend the capabilities of today’s accelerators and develop next-generation accelerator technologies for scientific discoveries and breakthroughs.
To learn more …
“Open Science” Essential to Accelerator Modeling
Controlling Optical Phase Drift for Better Laser-Plasma Accelerators
AI/ML in the Superconducting Magnet Program
For more information on ATAP News articles, contact caw@lbl.gov.
