AI Model Learns to Predict and Prevent Disruptions in Fusion Reactors, Paving Way for Improved Plasma Control
-
Researchers developed an AI model that predicts and avoids plasma from becoming unstable and escaping magnetic fields in fusion reactors. This allows fusion reactions to continue.
-
The model was tested on a real fusion reactor and was able to control the power and plasma shape to prevent instabilities.
-
The model works by being trained on real data rather than just physics models. It learns to maintain high power output while avoiding disruptive instabilities.
-
The system can predict tearing mode instabilities 300 ms before they happen, giving enough time to intervene and get the plasma under control.
-
Tearing modes are a major challenge for fusion reactors. Though still early stage, the AI proof-of-concept shows promise for optimization and control of fusion reactions.