Quarter-decay tuning meets dead time: numbers
The lab compares the Ziegler–Nichols PI gains with half of them. This note asks what the rule was aiming at in the first place: its quarter-amplitude decay target already means about 50% overshoot. Then it checks the measured decay, a model-based alternative and the effect of the delay alone.
Companion to the lab experiment Tuning against dead time. Run it there; this note predicts what you will see and why.
The model in play
The motor with 200 ms of input dead time, a 30 rad/s target and no load. The PI gains are the reaction-curve suggestion of the previous experiment, kp = 0.9T/(KL) and Ti = L/0.3, rounded to the lab's slider steps: kp = 0.335 and ki = 0.41.
Simplifications that matter here
- Ziegler and Nichols fitted their rule to process recordings. The motor is close to, but not exactly, a first-order lag with a delay.
- The decay ratio and ζ below treat the dominant oscillation as a second-order mode.
Predict it by hand
Quarter-amplitude decay means each overshoot is a quarter of the one before. For a second-order mode the ratio of successive same-sign peaks is e^(−2πζ/√(1−ζ²)), so a ratio of 1/4 means ζ = 0.215, and the first overshoot of such a mode is the square root of the ratio: 50%. The rule's own target overshoots by half the step.
Check against the engine
The engine overshoots 63.13%: the first error peak is 18.94 rad/s at 0.76 s and the next 8.19 rad/s, a decay ratio of 0.43, so the motor is less damped than the rule's target. The PID-tuning calculator, which simulates the fitted model with the same rule, gives 56.19%. The fit is close but not the plant: the motor's second pole and the 20 ms clock are not in it.
What to change next
- Use the SIMC setting from the PID-tuning calculator: kp = 0.185, ki = 0.14.
SIMC with τc = L gives Kp = T/(K(τc + L)) = 0.187 and Ti = min(T, 4(τc + L)) = min(1.307, 1.96) = 1.307 s: a lower gain and a slower integral. On the fitted model the calculator predicts 4.05% overshoot. The engine gives 56.85° of margin at 2.35 rad/s, 10.39% overshoot and settling at 2.30 s.
- Keep the Z–N gains and cut the dead time to 100 ms.
The crossover stays at 4.33 rad/s and the delay returns 4.33 × 0.1 rad = 24.8° of phase, so predict 19.09° + 24.8° = 43.88°. The engine's workbench gives 43.88°; the run overshoots 28.88% and settles at 1.78 s.
Common mistakes when reading this run
- Calling the Z–N overshoot a malfunction. The rule's quarter-decay target already implies about 50%.
- Trusting a model-based prediction without checking the plant. The fitted model promised 4.05% for SIMC; the motor gives 10.39%.
- Ranking settings by rise time. Z–N rises in 0.20 s and SIMC in 0.38 s, yet SIMC settles first, at 2.30 s against 5.82 s.
Where to go next
Run this experiment in the lab, pin a run and try the two changes above.
Lectures
- LECTURE 32Tuning PID controllers: Ziegler–Nichols and what to do afterwards
- LECTURE 26Why −180° matters: Nyquist and margins
Terms used in this note
Calculator
PID tuning calculator: Open the fitted model in the PID-tuning calculator and compare the PI rules side by side: Z–N overshoots 56.19% and SIMC 4.05% on the model.