Smith predictor
A dead-time compensation structure that removes long dead time from the loop by predicting the measurement.
What it means
The Smith predictor is a model-based structure for loops whose dead time is long relative to their time constant. The controller runs against a model of the process, and the difference between the model and the real measurement is used to correct the model, so the dead time appears to the controller to be gone.
It works because the controller is handed a prediction of what the process will do, instead of the delayed truth. Tuned properly it makes a dead-time-dominated loop behave like a dead-time-free one, a substantial improvement on what plain feedback can manage.
The price is the model. The predictor assumes the process is genuinely FOPDT, and where the real process differs from the model the loop can be less stable than plain feedback, not more. It belongs on commissionable processes with a model you trust — a study and engineering tool, not a trend-screen fix.
Related
Where this comes up
Last reviewed 2026-09-18.