Observability and Controllability

Observability asks whether sensors can reveal the internal state; controllability asks whether actuators can steer it.

Two questions before you build

Before a twin can estimate a state or a controller can regulate it, two structural questions must be answered. Observability: do the available measurements contain enough information to reconstruct the full internal state? Controllability: can the available actuators drive the state anywhere it needs to go? Both are properties of the system and its instrumentation, not of any particular algorithm.

Observability

A state is observable if distinct internal conditions always produce distinguishable measurement histories. If two different states would look identical to every sensor, no estimator, however clever, can tell them apart. Checking observability early tells the designers where diagnostics must be added so the twin can see what it needs to see.

Controllability

A state is controllable if some sequence of allowed inputs can move it from any starting point to any target. Uncontrollable directions are ones the actuators simply cannot influence; those must be handled by design margin rather than by control. For a fusion machine this shapes the choice and placement of coils, heating, and fueling systems.

Degrees, not absolutes

In practice observability and controllability are matters of degree. A state may be weakly observable, technically reconstructable but only with large uncertainty, or weakly controllable, steerable only slowly or with heavy actuator effort. Quantifying these degrees guides where to invest in better sensors or stronger actuators.

For the Kronos machines

Observability analysis of the Hyperion breeder guides diagnostic placement so the plasma equilibrium and internal profiles can be reconstructed from survivable external measurements. Controllability analysis of the burner's tandem-mirror plug fields determines whether the high-field plug and throat conditions can be steered by the planned coil and heating set. Both analyses are done today on the design models, before hardware exists, which is exactly when they are most useful. See state estimation and model-predictive control.