PACMAN AI framework for controlling fusion systems safely makes key decisions in milliseconds

An innovative software framework, known as PACMAN (Prediction And Control using MAChiNe learning), has been developed by researchers at the U.S. Department of Energy's Princeton Plasma Physics Laboratory (PPPL) and Princeton University. The AI framework can make life-saving decisions in milliseconds, keeping humans firmly in charge of the goals while managing the complex process of fusion energy systems.

Fusion energy, which could serve as a virtually unlimited source of electricity, is being perfected on Earth using tokamaks that use powerful magnetic fields to hold a plasma, or electrically charged gas. The plasma's stability requires constant adjustments, including heating systems, magnets, and gas injectors. Instabilities, or small disturbances, can grow in milliseconds and threaten the fusion reaction.

Traditional computer simulations take days or months, far too slow for real-time control during an experiment lasting only minutes. PACMAN addresses this issue by using machine learning models that can describe plasma behavior on the millisecond time scale. The AI framework operates in four stages: gathering real-time measurements, checking for errors, predicting plasma behavior, and calculating commands for the tokamak.

PACMAN demonstrates its effectiveness by predicting tearing modes (a type of instability) 200 milliseconds in advance, allowing the plasma to be adjusted to avoid the issue. It also simultaneously steers all six gyrotrons (systems that heat the plasma) in real-time, achieving complex goals set by researchers. The framework allows for easy integration of new models and rapid testing, making it a valuable tool for fusion research.