Scientists are using artificial intelligence to tackle one of the biggest challenges in nuclear fusion: keeping extremely hot plasma stable long enough for fusion reactions to continue.
Researchers from Princeton University and the U.S. Department of Energy’s Princeton Plasma Physics Laboratory have developed an AI framework called PACMAN, designed to monitor and control fusion plasma in real time. The system has been tested on the DIII-D National Fusion Facility tokamak in San Diego.
AI Makes Decisions in Milliseconds
Fusion plasma can become unstable extremely quickly, leaving human operators little time to react. PACMAN addresses this problem by combining several machine-learning models into a single control system.
The framework typically completes its control cycle in around 20 milliseconds, allowing it to repeatedly analyse plasma conditions and adjust the system far faster than a human operator could.
Predicting Problems Before They Happen
One of the most promising results came from an experiment involving a plasma instability known as a tearing mode.
Researchers found that an AI model could predict the instability about 200 milliseconds before it developed. Instead of waiting for the problem to appear and then trying to suppress it, the system could modify the plasma conditions in advance.
The experiments also demonstrated AI control of heating systems, plasma density and rotation, as well as waves generated by fast particles.
Humans Still Set the Rules
Despite the advanced automation, PACMAN is not designed to remove humans from fusion experiments. Hardware safety limits remain in place, while researchers continue to set the objectives and examine the results after experiments.
The modular design also means scientists can add or replace AI models more easily, potentially helping researchers test new control strategies much faster.
What It Could Mean for Fusion Energy
Nuclear fusion is being studied as a potential source of abundant electricity, but maintaining a stable plasma remains a major scientific challenge.
The researchers believe PACMAN could eventually be adapted for different tokamaks and future fusion machines. If AI can reliably predict and prevent plasma instabilities in increasingly complex experiments, it could become an important tool in the wider effort to make fusion energy practical.
