AI Framework PACMAN Predicts and Controls Fusion Plasma in 20 Milliseconds

Researchers at the U.S. Department of Energy's Princeton Plasma Physics Laboratory and Princeton University have developed PACMAN, an artificial intelligence software framework that can predict and control fusion plasma in about 20 milliseconds, while operating under strict hardware safety limits. The findings were published in the journal Nuclear Fusion.

PACMAN, which stands for "prediction and control using machine learning," can simultaneously invoke multiple AI models to adjust heating systems, magnetic fields and other parameters based on predictions. The framework also checks AI-generated control actions and enforces hardware safety constraints to prevent operations beyond safe limits.

In five experiments on the tokamak at the DIII-D National Fusion Facility, PACMAN used AI to control the plasma heating system, predict edge energy bursts, detect and control plasma waves driven by fast particles, and adjust plasma density and rotation according to preset targets. The researchers also used PACMAN to simultaneously control six gyrotrons at DIII-D, coordinating power output and mirror positions to work toward the scientists' goals.

The team stressed that AI has not replaced human control of fusion experiments. Scientists still set the experimental goals and control parameters, and the system always enforces hardware safety limits.