UF researchers using machine learning to pursue fusion power

University of Florida researchers are using specialized machine learning algorithms running on the HiPerGator supercomputer to understand, predict and prevent off-normal plasma events inside tokamak reactors. The research targets the ionized gas plasma that fusion experiments seek to contain and control. Fusion aims to replicate the processes that power the sun by fusing atoms at temperatures above 100 million degrees Celsius. Christopher McDevitt, a plasma physicist and professor in UF's Nuclear Engineering program, leads efforts to improve plasma predictability. Two projects funded by the National Science Foundation and the U.S. Department of Energy National Nuclear Security Administration support the work. Researchers use the supercomputer to create machine learning models of complex plasma events instead of relying on trial and error. Unstable plasma can cause damage to reactor materials or generate energetic electrons unintentionally. Simulations on the recently upgraded HiPerGator allow conditions to be tested rapidly without risking equipment. Accurate prediction and prevention of anomalies could advance the development of fusion as a carbon-free energy source.