Critical dead time of a PI loop, calculated
The lab raises the dead time step by step and watches the margin shrink. This note goes straight to the edge: one division gives the delay at which the loop stops settling, a second the period it rings at there, and two single-gain changes show which gain to lower when a delay appears.
Companion to the lab experiment Dead time limits speed. Run it there; this note predicts what you will see and why.
The model in play
PI on the default motor (kp = 0.22, ki = 0.6, kd = 0), a 50 rad/s target and no load, updated every 20 ms. The opt-in input dead time θ delays the drive voltage in whole 5 ms steps, and the frequency workbench includes the same delay in its sampled loop.
Simplifications that matter here
- The phase rule treats the loop as linear. Nothing clips until the loop is past the edge; at the lab's 300 ms setting the voltage reaches 12 V.
- The lab's menu offers 0, 20, 50, 100, 200, 300 and 500 ms; the edge lies between two entries, so it was checked in the engine at 5 ms resolution.
Predict it by hand
A delay leaves the gain crossover where it is, 3.49 rad/s, and removes ωθ of phase there, so the margin falls by 20.0° per 100 ms. The loop loses its whole 53.08° margin when ω_c θ equals it in radians. At that edge the loop rings at the crossover frequency.
Check against the engine
At 260 ms the engine's oscillation still shrinks: its largest error is 20.98 rad/s between 4 and 8 s and 18.65 rad/s between 8 and 12 s. At 270 ms it does not (30.59, then 30.95 rad/s). The workbench margin changes sign between them, 1.06° at 260 ms and −0.94° at 270 ms. At the lab's 300 ms the speed oscillates with a period of 1.85 s, between the 1.80 s estimate and the 2.04 s at which the phase now crosses −180°; with the voltage clipping, it is a limit cycle rather than a linear mode.
What to change next
- At 200 ms, halve only the integral gain (ki = 0.3).
The PI zero ki/kp moves from 2.73 to 1.36 rad/s, further below crossover, so the controller adds less lag where it matters: predict a larger margin. The engine gives 36.14° at 2.99 rad/s, 20.29% overshoot and settling at 2.16 s, against 35.47% and 6.28 s with the original gains.
- At 200 ms, halve only P instead (kp = 0.11).
Now the PI zero rises to 5.45 rad/s, above crossover, and the integrator's lag dominates there: predict a smaller margin, not a larger one. The engine gives 1.58° at 2.90 rad/s, 68.16% overshoot, and the speed never settles.
Common mistakes when reading this run
- Lowering P to cope with a delay in a PI loop. It moves the PI zero up toward crossover and adds lag, as the second variation shows.
- Looking for the delay in the Bode magnitude. Dead time changes only the phase, so the magnitude plot and the crossover stay put.
- Treating the 300 ms menu entry as the limit. The edge is at 0.265 s; a design should keep a margin, not stop just short of it.
Where to go next
Run this experiment in the lab, pin a run and try the two changes above.
Lectures
- LECTURE 26Why −180° matters: Nyquist and margins
- LECTURE 25Drawing Bode plots by hand and computing margins
Terms used in this note
Calculator
Bode plot calculator: Open this PI loop without the delay or the 20 ms hold in the Bode calculator: 54.30° of phase margin at 3.45 rad/s. Subtract ωθ from the phase by hand for any delay.