Computing Library › AI Plasma Control
AI Plasma Control

Plasma Control: An Overview

Why a burning plasma must be actively steered every millisecond, and how a layered control system keeps it inside a narrow operating window.

The plasma will not sit still

A magnetically confined plasma is an unstable, high-dimensional, partly observed physical system. Left alone it drifts vertically, changes shape, sheds heat unevenly, and can lose confinement in milliseconds. Control is not an accessory to a fusion device; it is a precondition for the device operating at all. Every discharge is the product of continuous feedback acting on coils, gas valves, heating systems, and pellet injectors.

What control must achieve

Kronos motion — operating point

Timescales set the architecture

The relevant timescales span roughly nine orders of magnitude, from microsecond magnetohydrodynamic (MHD) growth rates through millisecond feedback loops to second-scale scenario evolution and multi-second human decisions. No single controller spans that range. The result is a layered stack in which fast deterministic loops handle reflexes, slower model-based loops handle scenarios, and a predictive layer plus human oversight handle intent and exceptions.

Design context

At Kronos the breeder Hyperion is a D-T spherical tokamak and the burner is a D-3He tandem-mirror generator (in the Aegis and MetroVolt housings). Both are design and simulation efforts; the control concepts here describe method and modeling, not the behavior of built hardware. No hardware net-gain claim is made before first-of-a-kind first tritium.

The remaining pages in this section walk through the architecture layer by layer and then through the individual control problems: shape, position, current, density, disruptions, error fields, equilibrium reconstruction, actuator allocation, scenarios, and safe shutdown.