The Plasma State Observer
The plasma state observer reconstructs the full plasma state from a sparse, noisy diagnostic set in real time -- the estimator that feeds the digital twin and the control stack.
You never measure everything
Control needs the full plasma state, but diagnostics only sample it: a limited set of sensors, some noisy, some dropping out mid-shot. The plasma state observer is the estimator that fills the gap, inferring the complete state from partial measurements fast enough to act on.
How it works
The observer is a data-assimilation problem: combine a model with incoming measurements to maintain a best estimate. Classical routes are Kalman-family filters and state estimation over the diagnostic set. Kronos's KFLOW code treats the diagnostics as a 76-node sensor graph and uses a message-passing graph neural network to impute dropped or noisy channels, beating a naive baseline by roughly two-to-one on held-out data.
Where it sits in the stack
- Consumes validated telemetry from the data fabric.
- Feeds the real-time state-estimation control layer and the predictive digital twin.
- Provides the state on which shape, position, and stability control act.
Honest scope
The observer is calibrated against the twin and against real device data where available; its uncertainty is carried forward, not hidden, and validation continues. It is the same estimator serving both machines -- the breeder (Hyperion) and the burner -- as a shared component of the control architecture.