TUNE THE MODEL.
COMPARE THE TRADE-OFF.
Rules are applied only to the model classes they were derived for. Results are candidates for simulation—not universal answers and never automatic hardware approval.
SELECT UP TO FOUR
Tuning methods
RESPONSE OVERLAY
No universal winner.
Compare speed, overshoot, integrated error, actuator activity, and robustness. A lower value is not automatically a better engineering choice.
| Method | Kp | Ki | Kd | Rise / s | Overshoot / % | IAE | TV(u) |
|---|---|---|---|---|---|---|---|
| Manual | 2 | 0.8 | 0.25 | 1.85 | 0 | 1.225 | 1.603 |
| Ziegler–Nichols | 6.857 | 9.796 | 1.2 | 0.21 | 87.89 | 5.364 | 268.9 |
| Cohen–Coon | 7.869 | 9.796 | 0.9706 | 0.18 | 93.32 | 10.93 | 542 |
| Skogestad SIMC | 1.29 | 0.6452 | 0 | 2.46 | 0 | 1.525 | 0.7742 |
ÅSTRÖM–HÄGGLUND STYLE
Relay feedback experiment
A symmetric relay of amplitude d drives a limit cycle. From process oscillation amplitude a and period Pu, the describing-function estimate gives Ku = 4d/(πa). This is simulated autotuning; do not transfer it to hardware without a safety review.
Controller gains shown in the simulation are parallel-form values: Ki = Kc/Ti and Kd = Kc·Td. Each rule’s source, assumptions, units, and limitations are documented in docs/tuning-methods.md and docs/method-validation.md.