Real-Time Plasma Control
A fusion plasma changes on millisecond timescales, so control loops must sense, decide, and act faster than the instabilities they suppress.
The control problem
A confined plasma is a nonlinear system with many coupled states: position, shape, current profile, density, temperature, and impurity content. Left alone it drifts, and some drifts run away into events that end the discharge. Control keeps the plasma inside a safe operating envelope by continuously adjusting magnetic coils, heating, and fueling.
Sense, estimate, act
- Diagnostics sample magnetic fields, radiation, and density many thousands of times per second.
- A state estimator reconstructs the internal plasma state from those noisy signals.
- A controller computes coil and actuator commands to drive the state toward its target.
- Actuators respond within milliseconds, and the loop repeats.
Why computing changed this
Classical controllers use fixed gains tuned around one operating point. Modern control uses reduced-order models and learned policies that adapt across the whole scenario, handling the nonlinear coupling that fixed gains miss. Training happens offline against a digital twin; the deployed controller runs deterministically within its latency budget.
Both machines
The Hyperion breeder needs shape and current-profile control for a spherical tokamak with negative triangularity. The Aegis and MetroVolt burner is a tandem mirror, where control centers on the end-plug potentials that confine the central cell rather than on a toroidal current. The control philosophy is shared; the plant models differ.
The goal
Good control is not just stability; it is repeatable, high-performance operation and the foundation for disruption-free operation. Every controller is verified in simulation before it touches hardware, and its decisions are logged for later audit.