PID tuning — Ziegler-Nichols, loop shaping: Foundations

This page assumes you know nothing. Every symbol the parent note used is unpacked here, in the order that lets each one lean on the last.

1. The feedback loop — the picture everything lives in

Before any symbol, look at the loop. A machine we want to control is called the plant. We tell it where to go; it reports where it actually is; we compare; we correct. Round and round.

Figure — PID tuning — Ziegler-Nichols, loop shaping

The little $(t)$ after each letter means "this is a value that changes with time $t$ ." At $t = 0$ we start; as seconds tick by, each of these numbers moves.

2. The three pieces of information hiding in the error

The controller only ever sees the error signal $e (t)$ . But that one wiggling line secretly contains three different stories, and the letters P, I, D are just names for reading each story.

Now the three tools. Each is a piece of maths chosen to answer exactly one of those questions.

Proportional — "react to the present height"

Definition The proportional term

$K_{p} e (t)$ Plain words: push in proportion to how wrong you are right now — twice the error, twice the push. Picture: the orange arrow measuring the height of the error curve at the present instant. Why this piece: it is the simplest possible reaction — no new symbol, just the raw error $e (t)$ scaled by a number. Why the topic needs it: something must react to the present gap, and P is that something; it is the backbone every loop starts from. What problem it solves and its limits: P alone gets you most of the way to target fast, but two flaws appear. (1) It leaves a leftover steady-state error — because to keep pushing at all, P needs a non-zero error to feed on, so it settles just short of target. (2) If you scale it too hard (big $K_{p}$ ) the loop over-reacts and starts to oscillate, threatening stability. Those two flaws are exactly why the I and D pieces below have to exist.

Integral — "add up the past"

Derivative — "read the slope, predict the future"

3. The gains — the three tuning knobs

Why "tuning" exists at all: too much $K_{p}$ → wobble; too much $K_{i}$ → sluggish overshoot; too much $K_{d}$ → jittery from noise. Choosing the trio is the entire subject of the parent note.

4. The Laplace domain — why $s$ appears everywhere

The parent note suddenly writes $C (s) = K_{p} + \frac{K _{i}}{s} + K_{d} s$ . Where did $t$ go, and what is $s$ ?

Definition The Laplace transform itself

The transform of a time signal $f (t)$ is defined by an integral: $L {f (t)} = F (s) = \int_{0}^{\infty} f (t) e^{- s t} d t$ Plain words: multiply your signal by a decaying probe $e^{- s t}$ and add it all up over all future time. Each value of $s$ picks out one "flavour" of the signal, packing the whole time history into a single function $F (s)$ . Why $e^{- s t}$ : that decaying factor is what magically turns a derivative into " $\times s$ " (integration-by-parts does it) and an integral into " $\div s$ ". The fine print (region of convergence): the integral only adds up to a finite answer when $s$ is "big enough" that $e^{- s t}$ shrinks faster than $f (t)$ grows — the set of such $s$ is the region of convergence. For the well-behaved, bounded signals in control this is always satisfied near $s = j ω$ , which is the only region we actually evaluate, so we may treat the transform as valid everywhere we need it.

5. Size and angle of a wave — magnitude and phase

When we plug $s = j ω$ , the result is a complex number. It has two readable parts, and the entire loop-shaping method lives on these two.

Common mistake "The phase is just

$arctan (up-part / right-part)$ ." Why it feels right: on the arrow picture, $arctan$ of (vertical over horizontal) is the slope-angle, and for an arrow in the top-right corner it is correct. The fault: plain $arctan$ only ever returns an angle between $- 90°$ and $+ 90°$ , so it cannot tell a bottom-left arrow ( $real < 0, imag < 0$ , true angle near $- 135°$ ) from a top-right one — it collapses opposite quadrants together. But control lives out near $- 180°$ , exactly where this fails. Fix: use the two-argument $atan2 (imag, real)$ , which looks at the signs of both parts to place the arrow in the correct quadrant and returns the full signed angle over the whole $- 180° \dots + 180°$ range. In the figure, the bottom-left arrow gets its true $\approx - 143°$ only because we checked both signs. See Bode plot & frequency response, which plots this quadrant-correct phase versus $ω$ .

6. The special frequencies and margins

Now the last cluster of symbols the parent uses.

7. How the foundations feed the topic

The prerequisite map below is drawn as a real figure — colour-coded by which of the three "stories" each idea belongs to — so you can see the dependency flow rather than read it.

Read it bottom-to-top: the raw loop signals feed the error; the error splits into the present/past/future pieces; those attach to gains and combine into the control law; the control law is carried into the $s$ -domain, read off as magnitude and phase, and finally distilled into the margins that ZN and loop-shaping tune.

Equipment checklist

Hide the right side and test yourself before moving on.

What does

e (t)

mean and what is its formula?

The error,

e (t) = r (t) - y (t)

— where we want to be minus where we are.

What problem does pure proportional control leave unsolved?

A leftover steady-state offset (it needs a non-zero error to keep pushing), and it can oscillate if

K_{p}

is too big.

What is the picture of the integral

\int_{0}^{t} e d τ

The accumulated shaded area under the error curve from start to now.

What does the derivative

d e / d t

represent visually?

The slope (steepness) of the error curve at the present instant.

Which knob scales the past, and how do you undo integration in the

s

-domain?

K_{i}

scales the past; integrating becomes dividing by

s

Write the defining integral of the Laplace transform.

F (s) = \int_{0}^{\infty} f (t) e^{- s t} d t

, valid on its region of convergence.

Why do we replace

t

with

s

Laplace turns calculus into algebra — integrate ⇒ ÷

s

, differentiate ⇒ ×

s

What are

j

and

ω

j = - 1

, the imaginary unit;

ω

is angular frequency in rad/s, with

ω = 2 π / T

Why can't plain

arctan

give the phase, and what fixes it?

It collapses opposite quadrants (

\pm 90°

only);

atan2 (imag, real)

uses both signs for the full

- 180° \dots + 180°

range.

What do magnitude and phase of

L (j ω)

tell you?

Magnitude = size change of the wave; phase = timing shift in degrees.

Why is phase

= - 180°

with magnitude

= 1

the instability edge?

The wave returns flipped and full-size, so feedback reinforces it forever.

Define phase margin.

PM = 180° + ∠ L (ω_{c})

, the spare lag before instability; aim

30

–

60°

What are

K_{u}

and

T_{u}

Gain at sustained oscillation, and the period of that oscillation.

The letter

L

has two meanings on this page — what are they?

L (s)

= open-loop transfer function (§4);

L

= dead-time delay in seconds in the FOPDT model (§6).