Twin-in-the-Loop Control
Placing the twin inside the control loop gives the controller a live, corrected model to plan against, not a fixed design model.
The controller's window into the machine
A conventional controller acts on measurements alone. A twin-in-the-loop controller acts on the twin's continuously corrected estimate of the full internal state, and plans against the twin's model of how the machine will respond. The twin supplies both the present, an assimilated state, and the future, a fast model for prediction. This is the natural pairing of a digital twin with model-predictive control.
How the loop runs
- The twin assimilates the latest measurements into a state estimate with uncertainty
- The controller uses the twin's surrogate to predict responses to candidate actions
- It selects the action that best meets goals within safety constraints
- The action is applied, the machine responds, and the cycle repeats
Why it is powerful
Because the controller plans against a model that tracks the real, aging machine rather than an idealized design, its predictions stay accurate as the machine drifts. Because the twin reconstructs hidden states, the controller can regulate quantities no sensor measures directly, such as an internal plasma profile. And because the twin carries uncertainty, the controller can act cautiously when the estimate is poor.
Timing discipline
The whole loop must finish within the control period. This forces a strict separation between the fast surrogate-based path that runs in the loop and the slow full-physics path that runs alongside to check and retrain it. Any component that cannot meet the deadline stays out of the real-time path. See real-time computing.
In the Kronos program
Twin-in-the-loop control is developed today in simulation, controller and twin exercised against a high-fidelity model of the plant, a practice that will validate the loop before it ever touches hardware. It applies to both machines: shape and profile control for the Hyperion breeder, and plug-field management for the burner. No claim is made that these loops have run on hardware; the machines are not built. See hardware-in-the-loop testing.