Computing Library › AI Plasma Control
AI Plasma Control

Verification and Validation of Control

Establishing that a control system does what it should before it is trusted with a machine - and keeping that trust over time.

Two distinct questions

Verification asks whether the control system was built correctly - does the code implement the intended design, meet its timing budget, and behave as specified? Validation asks whether the intended design is the right one - will it actually keep this plasma inside its limits? Both are needed; passing one without the other is not enough to trust a control system.

Testing before the plasma

Kronos motion — lego machine

Control is tested long before it meets a plasma. Software-in-the-loop runs the controller against a simulation model; hardware-in-the-loop runs the real control hardware against a real-time plasma simulator, exercising the actual timing and interfaces. These tests probe normal operation, limit approaches, sensor failures, and fault responses in conditions too dangerous to try first on a machine.

What must be verified

Validation against reality

Ultimately a control approach is validated against experiment - discharges where the model's predictions are checked against what the plasma did, and the controller's performance is measured against its goals. Because no existing machine reaches a reactor's regime, validation is layered: models validated against present experiments, and their extrapolation to new regimes treated as uncertain and flagged, never assumed.

Continuous, not one-time

Verification and validation are not a gate passed once. Every change to code, models, or scenarios re-opens them, and every discharge is fresh evidence for or against the models. For Kronos this is design and simulation work: the control approaches for the breeder Hyperion and the burner are validated in modeling, with hardware validation awaiting construction planned from Q2 2027.