Control-Oriented Modeling
Building models simple and fast enough for control yet faithful to the dynamics that matter - a different craft from physics modeling.
A different purpose
A physics model aims for the most accurate possible description; a control-oriented model aims for the simplest description that captures the dynamics a controller must handle, fast enough to use in design and in real time. The two goals pull in opposite directions, and control-oriented modeling is the craft of stopping at the right level of detail.
What to keep and what to drop
A good control model keeps the dynamics inside the loop's bandwidth and the couplings the controller must coordinate, and drops fast dynamics the loop cannot see and slow dynamics that appear as constants over the loop's horizon. Choosing this split correctly is what makes a model both usable and trustworthy for control.
Common reductions
- Linearization about an operating point for local loop design
- Rigid-body plasma models for position and current dynamics
- Reduced-order models fitted to full-simulation or experimental data
- Surrogate models that approximate expensive physics with fast functions
- Circuit models for the coil and power-supply dynamics
Validation against the real thing
A control model must be validated where it will be used: compared against higher-fidelity simulation and against experiment across the operating range the controller will visit. A model that is accurate at one operating point but wrong at another causes a gain-scheduled controller to fail exactly where a phase transitions. Validation covers the trajectory, not just a point.
Knowing the model's edges
Every control model is wrong outside its fitted range, and a disciplined design states where that range ends. Robust control explicitly bounds the model error and demands stability across it; gain scheduling switches models as the operating point moves. The mark of good control-oriented modeling is not pretending the model is exact, but knowing precisely where it is not.