Princeton researchers report that a new AI model has solved one of the major roadblocks to generating fusion energy.
The Princeton Plasma Physics Laboratory has developed an AI model that predicts, and then figures out how to avoid fusuin power plasma becoming unstable and escaping the strong magnetic fields that hold it inside certain donut-shaped reactors. They published their findings Wednesday in Nature
Donut-shaped tokamak reactors rely on magnets to squeeze plasma particles close together and keep them constantly spinning around a ring, creating a lasting fusion reaction. They’re one of the front-runners in designs for a practical fusion reactor. But if there’s one little disruption to the magnetic field lines running through the plasma, the delicate balance keeping it all contained gets out of whack: The plasma escapes the magnets’ clutches and the reaction ends.
Chijin Xiao, a plasma physicist at the University of Saskatchewan who wasn’t involved in the study, explained that these instabilities can lead to catastrophic consequences. “When the plasma stops operating, there are several risks: one is that all the energy stored in the plasma is going to be released as thermal energy and may damage the wall of the reactor," she said. "More importantly, a sudden change in the [magnetic] current can introduce a great deal of force on the reactor that can really destroy the device. Xiao added that one of the biggest tokamak reactors around today, ITER in France, is only designed to withstand a few of these plasma disruptions before the whole machine has to be repaired—a huge expense. The goal is to catch instabilities while they’re small and intervene.
The Princeton lab’s model can predict so-called tearing mode instabilities 300 milliseconds before they happen. It doesn’t sound like a lot of heads-up, but it’s enough time to get the plasma under control, their study shows.
Researchers tested the algorithm on a real reactor, the DIII-D National Fusion Facility in San Diego. They saw that their AI-based system could control the power being pumped into the reactor and the shape of the plasma to keep the swirling particles in check.
Co-author Azarakhsh Jalalvand said in a statement that the success of the AI model comes from the fact that it was trained on real data from previous fusion experiments, rather than theoretical physics models.
“We don’t teach the reinforcement learning model all of the complex physics of a fusion reaction,” Jalalvand said. “We tell it what the goal is—to maintain a high-powered reaction—what to avoid—a tearing mode instability—and the knobs it can turn to achieve those outcomes. Over time, it learns the optimal pathway for achieving the goal of high power while avoiding the punishment of an instability.
”The study is significant, said co-author Jaemin Seo, because previous models studied have only been able to suppress tearing instabilities after they happen. “Our approach allows us to predict and avoid those instabilities before they ever appear.”
But tearing mode instabilities are just one of the ways plasma can become unhinged. There are dozens of ways a glob of plasma can wobble, bend, or break apart: like a kinked garden hose, a fan, or even a sausage.
Nevertheless, tearing instabilities are one of the biggest challenges on the way to boundless clean fusion energy. “Tearing mode instabilities are one of the major causes of plasma disruption, and they will become even more prominent as we try to run fusion reactions at the high powers required to produce enough energy,” said Seo. “They are an important challenge for us to solve.”
