AI Controller Uses Reinforcement Learning to Stabilize Fusion Plasma, Prevent Disruptions
-
Recent fusion experiments like NIF and tokamaks achieved milestones, but tearing instability remains an issue.
-
Tearing instability often leads to plasma disruption in conditions similar to ITER baseline.
-
We developed an AI controller using deep reinforcement learning to avoid tearing by controlling beam power and shape.
-
The AI controller maintains high pressure while keeping tearing possibility low by observing plasma profiles.
-
Experiments showed the controller can sustain higher performance plasma without disruption.