PID tuning — Ziegler-Nichols, loop shaping: Question bank

3.5.39 · D5 · Physics › Guidance, Navigation & Control (GNC) › PID tuning — Ziegler-Nichols, loop shaping

Underlying machinery ke liye PID controller basics, Bode plot & frequency response, aur Stability margins (gain & phase margin) ko side mein khula rakho.

0. Symbols aur pictures jo traps se pehle chahiye

Neeche jo bhi hai wo ek choti si vocabulary reuse karta hai. Beech-question mein doosre notes pe bhejna avoid karne ke liye, yahan minimal self-contained dictionary hai, har symbol ek picture se tied hai.

Figure — PID tuning — Ziegler-Nichols, loop shaping

Yahan se sab kuch sirf unhi symbols aur pictures ka use karta hai.

True or false — justify karo

At the edge of sustained oscillation the open loop has gain 1 and phase −180°.

True. Ek signal jo inverted (−180°) feed back hoke aur unattenuated (gain 1) hoke exactly khud ko reinforce karta hai, deta hai ek constant-amplitude oscillation forever — yahi stability boundary ki definition hai (Figure s01, middle trace).

Increasing

K_{p}

always makes the closed loop faster.

False. Ye crossover frequency (speed) badhata hai ek point tak, lekin ultimate gain

K_{u}

se aage loop unstable ho jaata hai — "faster" ka matlab "diverging" ho jaata hai. Speed aur stability trade off karte hain.

Integral action improves both steady-state accuracy and phase margin.

False. Ye steady-state error khatam karta hai lekin

1/ s

pole crossover ke paas ek flat

- 90°

phase contribute karta hai (Figure s04), jo phase margin reduce karta hai — accuracy aur robustness yahaan opposite directions mein pull karte hain.

Derivative action adds damping, so more

K_{d}

is always safer.

False. Iska zero phase lead add karta hai (margin ke liye acha, Figure s04 green) lekin high-frequency noise ko

s

se multiply karta hai; bahut zyada

K_{d}

actuator ko sensor noise ke peeche bhaata hai aur amplified measurement jitter se destabilise kar sakta hai.

Ziegler-Nichols gains are conservative and safe to ship.

False. Classic ZN quarter-amplitude decay ko target karta hai, jiska matlab roughly

25%

overshoot — aggressive hai, conservative nahi. Ye ek starting point hai, final answer nahi.

The two ZN methods (ultimate-gain and reaction-curve) should give identical gains.

False. Ye plant ko alag-alag model karte hain (

K_{u}, T_{u}

critical point vs

L_{dt}, R

FOPDT fit) aur sirf approximately agree karte hain, kyunki

T_{u} \sim 4 L_{dt}

ek approximation hai. Similar ballpark expect karo, identical numbers nahi.

Gain margin and phase margin measure the same robustness.

False. GM batata hai instability se pehle gain kitna scale ho sakta hai (

ω_{180}

par); PM batata hai kitna extra lag add ho sakta hai (

ω_{c}

par). Ek loop ka GM acha aur PM kharab ho sakta hai ya vice versa (Figure s03).

A loop with high phase margin can still respond sluggishly.

True. Phase margin overshoot/damping bound karta hai, lekin speed crossover frequency

ω_{c}

se set hoti hai. Agar

ω_{c}

low ho to loop bahut damped lekin bahut slow ho sakta hai.

The standard form

T_{i} = K_{p} / K_{i}

and

T_{d} = K_{d} / K_{p}

, so a ZN table in

T_{i}, T_{d}

is interchangeable with one in

K_{i}, K_{d}

without care.

False. Ye convertible zaroor hain, lekin tables ek specific form mein stated hoti hain;

T_{i}

ko wahan laga do jahan

K_{i}

expected hai (ya

K_{p}

se multiply/divide karna bhool jaao) — ye sabse common tuning bug hai.

Loop shaping requires knowing the plant model; ZN closed-loop does not.

True. Ultimate-gain method real plant par ek experiment se

K_{u}, T_{u}

extract karta hai — koi model nahi chahiye. Loop shaping ek known

L (j ω) = C (s) G (s)

ko reshape karta hai, isliye use

G

chahiye.

Spot the error

"To hit steady-state accuracy fast, set

ω_{i} = K_{i} / K_{p}

near

ω_{c}

Error hai integral corner

ω_{i}

ko near crossover rakhna — isse integral ka

- 90°

lag exactly wahan dump ho jaata hai jahan phase margin chahiye.

ω_{i} \approx ω_{c} /10

rakho taaki integral lag crossover se pehle decay ho jaaye.

"For

G (s) = 1/ (s + 1)^{3}

∠ G (j ω) = - arctan (3 ω)

Galat: teen identical poles mein se har ek

- arctan ω

contribute karta hai, isliye phase

- 3 arctan ω

hai,

- arctan (3 ω)

nahi. Cascaded factors ke phases add hote hain (Figure s04); arguments ek arctan ke andar merge nahi hote.

"Sustained oscillation happens when

∠ G (j ω_{u}) = - 90°

Galat angle — stability boundary ke liye open loop phase

- 180°

par chahiye, kyunki sirf ek inverted feedback signal khud ko reinforce karta hai.

- 90°

par loop still comfortably stable hai.

"Use derivative on the error,

K_{d} d e / d t

, exactly as the formula says."

Formula sahi hai lekin implementation nahi: reference

r

mein step se

d e / d t

ek huge spike ban jaata hai ("derivative kick", Figure s05). Balki derivative measurement par use karo

- K_{d} d y / d t

— see Integral windup and anti-windup integral side ke analogous clamping fix ke liye.

"Since GM

\geq 6

dB is the target, a loop with GM = 20 dB is even better tuned."

Zaroor nahi — bahut bada gain margin usually matlab loop over-detuned hai (low

ω_{c}

), isliye sluggish hai aur disturbance rejection kharab hai. Margins ek window hain jisme rehna hai, "bigger is better" score nahi.

"Anti-windup changes the tuning gains

K_{p}, K_{i}, K_{d}

Nahi — anti-windup sirf integrator ko accumulate hone se rokta hai jab actuator saturated ho. Gains unchanged rehte hain; ye nonlinear saturation behaviour fix karta hai, linear tuning nahi.

Why questions

Why does pure P control leave a steady-state offset?

Zero error par P term zero output karta hai, lekin load/disturbance ke against hold karne ke liye nonzero output chahiye — isliye loop us chhote se error par settle ho jaata hai jo wo holding output produce karta hai. Sirf integral action hi zero error par output de sakta hai.

Why does the integral term specifically eliminate steady-state error?

Integrator tab tak accumulate karta rehta hai jab tak koi bhi error exist kare, isliye uska output tab hi change karna band hoga jab error exactly zero ho — ye steady state mein

e = 0

force karta hai.

Why place the derivative zero below the crossover frequency?

Uska phase lead dheere-dheere

0° \to + 90°

badhta hai jab

ω

zero se upar jaata hai (Figure s04, green); us lead ka zyaataar hissa

ω_{c}

par already active rehne ke liye (jahan margin chahiye), zero ek lower frequency par hona chahiye taaki phase crossover tak aa jaaye.

Why does dead time

L_{dt}

force smaller controller gain in the reaction-curve method?

Dead time pure phase lag

- ω L_{dt}

add karta hai jo frequency ke saath badhta hai bina gain touch kiye, isliye phase margin tezi se khatam karta hai. Iska sirf ek hi defence hai:

ω_{c}

lower karo (hence lower gain), isliye

K_{p} \propto 1/ L_{dt}

hai.

Why can't a proportional-only loop achieve both fast response and small overshoot on an oscillatory plant?

K_{p}

badhane se closed-loop poles root locus par imaginary axis ki taraf jaate hain, speed badhti hai lekin damping ghatti hai. D-action ke added lead ke bina, speed aur damping ek saath nahi kharid sakte.

Why is high open-loop gain wanted at low frequency but not high frequency?

Low-frequency gain slow disturbances aur tracking error suppress karta hai (loop unhe "dekhta" aur correct karta hai); high-frequency gain sensor noise amplify karta aur robustness violate karta jahan plant model sabse kam trustworthy hai.

Why does phase margin usually grow as

ω_{c} \to 0

ω_{c}

lower karna Bode plot par crossover ko leftward slide karta hai us region mein jahan plant ke poles abhi apna pura

- 90°

donate nahi kar paaye (phase abhi

0°

ke paas hai, Figure s04). Kyunki PM

= 180° + ∠ L (ω_{c})

, lower

ω_{c}

par kam-negative phase matlab bada margin — lekin slower, disturbance-blind loop.

Edge cases

What does closed-loop ZN do if the plant never sustains oscillation as

K_{p}

rises?

\leq 2

poles aur koi dead time nahi wale plants ke liye phase kabhi

- 180°

nahi pahunchti (har pole

- 90°

par cap hai, Figure s04), isliye koi finite

K_{u}

nahi (equivalently koi

ω_{u}

nahi) — method simply apply nahi hoti, aur reaction-curve ya model-based method use karna padega.

For a pure integrator plant

G = 1/ s

, what is its phase, and what does that mean for ZN?

Iska phase constant

- 90°

hai, kabhi

- 180°

nahi pahunchta, isliye koi ultimate gain exist nahi karta. Aisa plant already steps ke liye zero steady-state error rakhta hai, isliye heavy integral action aksar unnecessary hai.

What happens to phase margin as

ω_{c} \to 0

(very detuned loop)?

PM typically ek large stable value ki taraf badhta hai (upar "why" dekho), lekin loop uselessly slow ho jaata hai aur disturbances reject nahi kar pata — ek "stable but deaf" controller. Robustness without performance.

If the sensor is very noisy, why might you set

K_{d} = 0

entirely?

Derivative noise ko

s

se multiply karta hai (gain frequency ke saath badhti hai), isliye noisy channel par ye actuator command dominate kar sakta hai; D drop karna (ya isse

K_{d} s / (1 + τ s)

se filter karna) thodi si damping trade karta hai ek bahut quieter, safer output ke liye.

For an FOPDT plant with

L_{dt} \to 0

(no dead time), what does the reaction-curve formula predict?

K_{p} = 1.2/ (R L_{dt}) \to \infty

— recipe infinite gain demand karta hai, jo nonsense hai. Ye signal karta hai ki delay-free plant us tarah limited nahi hai jaise FOPDT model assume karta hai; alag design use karo.

What is the danger of tuning at one operating point for a nonlinear plant?

Jo gains ek setpoint ke paas acha margin dete hain wo doosri jagah tiny ya negative margins de sakte hain (changed local slope/gain), isliye loop test par stable aur service mein unstable ho sakta hai — gain-scheduling ya state-space/LQR methods isko address karte hain.