Computing Library › AI Plasma Control
AI Plasma Control

Controllability and Observability in Plasma Control

Two structural properties that decide whether a plasma quantity can be driven by the actuators and inferred from the sensors.

Two fundamental questions

Before designing any loop, two questions must be answered. Controllability: can the available actuators drive the plasma from any state to any desired state? Observability: can the available sensors determine the full state from their measurements? If a quantity is uncontrollable or unobservable, no controller, however clever, can regulate it.

The formal test

Kronos motion — control room

For a linear model with state matrix A, input matrix B, and output matrix C, controllability is tested by the rank of the controllability matrix [B, AB, A^2 B, ...] and observability by the rank of the observability matrix [C; CA; CA^2; ...]. Full rank means full controllability or observability. In practice, the useful question is not the binary rank but how strongly each mode is controllable or observable.

Degrees, not absolutes

Real systems are rarely perfectly controllable or hopelessly uncontrollable; they have modes that are strongly, weakly, or barely reachable. Singular-value analysis of the response matrices ranks the shape and profile combinations by how much actuator effort they need. Nearly uncontrollable directions demand huge coil currents for small effect and are best left un-targeted.

Consequences for design

Why it comes first

Controllability and observability are properties of the machine's design - its coils and sensors - not of the controller. They must be checked when the device is designed, because a control problem that is structurally impossible cannot be fixed in software. Good actuator and diagnostic layout is control design done in hardware.