Machine learning is helping solve long-standing magnetic stability problems in Tokomak reactors.
Achieving the promise of nuclear fusion as a limitless, safe, and clean energy source is a major challenge for science and engineering. The simplest closed magnetic geometry used to confine ultra-hot plasmas is the donut-shaped rotationally symmetric tokamak. These configurations are susceptible to instabilities, or tearing modes (TM), that can rearrange the tokamak’s magnetic field lines, disrupt the tokamak’s symmetry, and disrupt the behavior of the plasma. Much work is being done to resolve this issue.
Cristina Rea and Stuart Benjamin investigated the growing role of machine learning (ML) in achieving this goal.
“The mechanism by which TM appears in tokamaks remains nonlinear, coupled, and chaotic,” said author Benjamin. “The balance of stabilizing and destabilizing effects at the sensitive ‘rational plane’ where the TM forms is tipped by intermittent, rapid instabilities elsewhere in the rotating plasma column. However, the final state of the unrelaxed tearing mode is simple: a giant magnetic bubble grows like a slug within the plasma, and the rotation stops before the plasma disperses to the wall.”
The researchers focused on a variety of areas, including ML prediction of tear mode initiation, use of ML to aid in interpretation of tear initiation data, and plasma controllers leveraging an artificial intelligence-based tear stability predictor.
“TM remains incredibly difficult to predict with physical models, but its probabilistic complexity is attractive to scientists leveraging ML,” said Benjamin. “That’s why we wanted to explain how recent research applies AI to large-scale experimental tokamak datasets and provides new insights into the physics and control mechanisms of TM that must be perfected to ensure that TM does not endanger future tokamak power plants.”
sauce: “A review of machine learning-driven studies of tearing modes in tokamaks,” C. Rea and S. Benjamin, Physics of Plasmas (2026). This article can be accessed from: https://doi.org/10.1063/5.0325461 .
This paper is part of the 2nd European Conference on Magnetic Reconnection in Plasma Collection. Click here for details here .
