Proportional & Integral Controllers

Proportional + Integral (PI) controllers were developed because of

the desirable property that systems with open loop transfer functions
of type 1 or above have zero steady state error with respect to a step
input.

U (s) K
The PI regulator is: = KP + I
E (s) s

But can be realised easily in the following form:

1
+ R2
1 C1 R2 U ( s) sC1
=−
sTI E (S ) R1
e(t) R1
_ u (t )
E(s) U(s)
KP + +
R 1 
= − 1 + 
 R2 sC1 R1 

Ex: assume we wish to apply PI regulator to a type 0 plant:

1
G p (s) =
s 2 + 3s + 2
Gc(s) Gp(s)
R(s) E(s) 1 U(s) 1 C(s)
+ K P (1 + )
sTI s + 3s + 2
2

Thus:
1 G p (s)
OLTF = K P ( s + )
TI s

C (s) K P ( s + 1 TI )
CLTF = =
R( s) s( s + 3s + 2) + sK P + K P TI
2

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Tuning PI Controllers
General approach to tuning:
1. Initially have no integral gain (TI large)
2. Increase KP until get satisfactory response
3. Start to add in integral (decreasing TI) until the steady state error
is removed in satisfactory time
(may need to reduce KP if the combination becomes oscillatory)

Anti-windup in I & PI controllers

Under some operating conditions non-linearities in the plant or
controller can stop an Integral controller from removing the steady
state error. If the Integrator output is not limited, then during this
time the total of the integrated (summed) error {≈ΣKIe(t)dt} will
continue to build. Once the restrictions are finally removed,
problems can arise because this built up “energy” must be removed
before the integral control can act normally… this can take a long
time. To avoid this, anti-windup circuits are added that place ±
limits on the integral total. These limits are usually placed on the
summed output of the P&I controller as well.

u(t)

5V

t
C1 R2

R(s) E(s) R1 U(s) C(s)

+ -1 _ Gp(s)
− +

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 PID (three term) Controllers
Proportional action: responds quickly to changes in error deviation.
Integral action: is slower but removes offsets between the plants
output and the reference.
Derivative action: Speeds up the system response by adding in
control action proportional to the rate of change of the feedback
error. Consequently this is susceptible to noise in the error signal,
which limits the derivative gain. When present this allows larger
values of KP and KI (smaller TI) to be used than possible in pure PI
regulators, but large values of derivative gain (KD) will cause
instability.
Error
{e(t)}

 1 de(t ) 
The PID regulator is given by: u (t ) = K P e(t ) + ∫ e(t )dt + TD 
 TI dt 
KP
Here: KI = K D = K P TD
TI

U (s) KI K D s 2 + K P s + K I N ( s)
= KP + + KDs = =
E (s) s s D( s)

NB: this transfer function is non-proper and is therefore difficult to

realise in practice.

Proper T.F.: Order N(s) ≤ Order D(s)

Strictly proper T.F.: Order N(s) < Order D(s)

These are generally easier to realise, and also reduce the

susceptibility of the derivative action to noise.

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Practical PID controllers

U ( s)  1 1 
= KP  + (1 + sTD ) 
E ( s)  sTI (1 + sαTD ) 

1. The 1/(1+sαTD) term acts as an effective low-pass filter on the

P+D regulator to attenuate noise in the derivative block.
2. If α = 0 the original PID form is obtained
3. Typically α = 0.1 to place the filter as far away from the
derivative action as possible. It is generally impractical to move
the filter further away as the control action (effort) becomes too
much and can-not be realised (supply rail or limiters saturate the
control effort)
4. PD controllers could be realised on their own if the plant does
not require an integrator.

Circuit realisation:
C2  1 1 
R2 R1 C1  R1 + R1 + 
= − 
U (s) sC sC1
e(t) 1
+
R3
_ u (t ) E ( S )  R 1 
 3 R2 + 
+  sC 2 
R 1 C (1 + sC1R1 ) 
= − 1 + + 2 
 R3 sC1 R3 C1 (1 + sC2 R2 ) 

If the filter term is disregarded for KP then:

U (s)  R C  1 sC1 R1   K sK PTD 

≈ −  1 + 2  + +  = − K P + P + 
E ( s)   R3 C1  sC1R3 (1 + sC2 R2 )   sTI 1 + sαTD 

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
A complete derivation results in:

   R  
  C2 R1 1 + 2   
 1 + s  R3   
  
  R1 C2   
  +  
U (s)  R C  1   R3 C1   
= − 1 + 2  + 
E (s)  3
R C1   R C  (1 + sC2 R2 ) 
sC R  1 + 2 
 1 3  R3 C1  
 
 
 
 

The Ziegler-Nichols tuning Technique for P, PI & PID controllers

(Golten & Verwer §7.7)

Proportional: Gc ( s ) = K P
 1 
PI: Gc ( s) = K p 1 + 
 sTI 
 1 1 + sTD 
PID: Gc ( s ) = K P  + 
 sTI 1 + s0.1TD 

1. Set TD = 0 & TI = ∞
2. Increase KP until the system just starts to oscillate (KP = KPO).
The frequency of oscillation here is ωc and the period is
TO=2π/ωc.

Set controller gains as:

Type KP TI TD

P
PI
PID

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Notes:
1. Even though first and second order systems cannot in principle
fully oscillate, practical systems always contain transport delays
and non-linearities that make them oscillate if the loop gain is
high.
2. A disadvantage with the Z-N technique is that it really needs to
be performed in real-time in the actual plant. In some cases it
may be undesirable to have the plant/process oscillate even for
tuning purposes.

Example of a practical PD Controller

Rate Feedback with the DC Servo Drive

If we place a tachometer on the servo drive, this will output a voltage

proportional to the mechanical speed ωm. This can be feed back to
the regulator.

R(s) = θref E(s) U(s) Km τ ωm 1

+ K C(s) = θ
(s + 1 τ ) s
−
−
Tacho Feedback Kω

As ωm = dθ/dt, feeding back some measure of the ωm term is like

feeding back the derivative (hence termed rate feedback)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
What effect does this have on our performance specs?

Eg: damping?
Consider the inner loop as a non-unity feedback control system then:

R(s) E(s) U(s) Km τ ωm

KK m τ
+ K
(s + 1 τ ) (s + 1 τ )
− =
KK m Kω τ
Tacho Feedback 1+
Kω
(s + 1 τ )

resulting in the following system:

R(s) = θref E(s) KK m τ ωm 1

+ C(s) = θ
s + (1 + KK m K ω ) τ s
−

C ( s) θ KK m τ
= = 2
R( s ) θ ref s + s (1 + KK m K ω ) τ + KK m τ

if Kω = 0 then the original transfer function arises.

1 + KK m K ω
Here: ω n = KK m τ and ζ =
2 τKK m

As opposed to simple proportional control, K can be increased to

give improved response times (larger ωn), while maintaining good
damping by increasing Kω.

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 ROOT LOCUS

Root locus: is a plot of the paths of the characteristic equation for a

closed loop system (CLTF poles) in the complex plane as a function
of various loop gains.

Plotting Techniques
1. Determine the OLTF of the system.
R(s) E(s)
+ Gc(s) GP(s) C(s)
−
H(s)
G ( s)
CLTF =
1 + G ( s) H ( s)

clearly the OLTF; G(s)H(s) affects the system poles.

In general form G(s)H(s) = KN ( s )

D(s )

N(s) is a polynomial in s of order "W"

D(s) is a polynomial in s of order "N"
N is also the order of the characteristic equation
K is called the loop sensitivity if the transfer function is
written so that the coefficients of the highest powers in s
in both the numerator and denominator are unity.

2. Factorise G(s)H(s) into the form (s+a) where a maybe complex.

K ( s + z1 ) K
G ( s) H ( s) =
( s + p1 )( s + p2 ) K

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
3. The characteristic equation is: 1 + G (s) H ( s) = 0
so that the roots of this equation must satisfy the expression:
K ( s + z1 ) K
G( s) H ( s) = = −1
( s + p1 )( s + p2 ) K
Because G(s)H(s) is a complex quantity, the above condition
{G(s)H(s) = −1} is in fact a composite of two conditions which
both must be valid.

i. A magnitude condition

G (s) H (s) = 1 …(i)

ii. An angle condition
∠G ( s ) H ( s ) = (1 + 2h)π …(ii)
where h=0, ±1, ±2,…

These conditions can be investigated graphically by plotting all of

the poles and zeros of the OLTF in the complex plane and noting
that each of the terms (s+z1),(s+z2),...,(s+p1),(s+p2)... are vectors
terminating at an arbitary point s in the complex plane.

For s to be a point on the root loci it must satisfy both conditions:

s
Ex: jω

K ( s + z1 )
G (s) H (s) =
s ( s + p2 )( s + p3 )

-p3 -z1 -p2 -p1 σ

similar to Ex fig 7.6

(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 using (i) we have:
K s + z1 s s + p2 s + p3
=1 K=
s s + p2 s + p3 ⇒ s + z1
using (ii) we have:
± 180 o = ϕ 1 − φ1 − φ 2 − φ 3
Further examples:

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Qualitative analysis

How do zeros and higher order poles in the OLTF (G(s)H(s)) affect
the root locus plot?

In order to answer this we'll first try to obtain a conceptual feel for
the effects of additional zeros and poles on simple and more complex
systems by looking at the following root locus plots.

Observations
1. The addition of a zero to the OLTF - pulls the root locus to the
left. This tends to enable poles to be selected that result in stable
faster response times (ie, the settling time is smaller).

2. The addition of a pole to the OLTF - pulls the root locus to the
right so that slower response times are possible, but the new root
locus paths may cross over into the RHP so that K must be
bounded to avoid instability.

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 Ex fig 7.4
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 Ex fig 7.5
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 Root Locus Plotting Rules

The following eight rules enable fast sketches of root locus to be

accomplished that:
- are a useful check for computer generated plots.
- enable a qualitative idea of the system performance.

1. Number of branches of the Root Locus

“The number of branches (root loci paths) equals the number of
poles of the OLTF (N)”.

2. Real Axis Locus Values

Applying the angle condition to any search path along the real axis:
“If the total number of real poles and zeros (in the OLTF) to the
right of a search point s on the real axis is odd, this point lies on the
root locus”.
Note:
(i) The angular contribution of complex conjugate pairs (poles and zeros) to
such a point is 360°, and can therefore be ignored.
(ii) A point satisfying this condition may be part of more than one branch.

3. Root Locus Start and End Points

N

∏ s− p c
K= c =1
W

∏ s−z
h =1
h

when: s = pc, K = 0, and when: s = zh, K = ∞.

also since for proper systems: N ≥ W, then: (N−W) poles must exist
such that their root locus branches do not finish on a zero, but rather
go to infinity (s→∞) as K→∞, (this is equivalent to the effect of a
zero).

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
“The starting points (K=0) are the poles of the OLTF locus path.”
“The ending points (K=∞) are either zeros of the OLTF or points at
s=∞ that are considered equivalent zeros of multiplicity (N-W)).

4. Asymptotes of Locus as s approaches Infinity

“N-W branches of the root locus have straight line asymptotes (as
s → ∞ )emulating from a single point whose angles λ (w.r.t the real
axis) are given by:
(1 + 2h)π
λ=
(N − W ) h=0,±1, ±2 …”

Proof:
K
lim G ( s ) H ( s ) = lim N −W
= −1
s →∞ s →∞ s

must satisfy the magnitude and angle condition, hence:

∠G ( s ) H ( s ) = ∠ K s N −W = (1 + 2h)π
( N − W )∠s = (1 + 2h)π
where ∠s = λ
As N and W are constants for a given OLTF then as s → ∞
(irrespective of magnitude) the angle λ → constant. Thus the
branches are asymptotic to straight lines.

Examples:
(i) (N-W) = 1
jω

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 (ii) (N-W) = 2
jω

(iii) (N-W) = 3 jω

jω
(iv) (N-W) = 4

5. Real Axis Intercept of the Asymptotes

“The asymptotes intersect at a point σo on the real axis”
N W

σ 0 = ∑ ℜ( pc ) −∑ ℜ( z h ) ( N − W ) jω
 c =1 h =1 

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
6. Breakaway Point on the Real Axis
In many instances root loci must breakaway from the real axis (ie,
become complex) in order to reach the end points (K= ∞). If these
loci return to the real axis at some later stage there will also be a
break-in point.

Plotting the loop sensitivity (K) against the real axis σ demonstrates
when these conditions arise.

Examples: If a locus exists on the real axis between:

(i) two poles: the locus starts at K=0 and as K increases must
breakaway in order to finish on either a zero or a point of
infinity (both with K=∞).
K

Ex fig 7.11 σ
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

(ii) two zeros or a zero & a point of infinity: 2 branches must

break-in from the complex region between them so as to
terminate at these end points.

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
(iii)a pole and a zero: either pairs of breakaway and break-in points
occur, or none at all.
K

σ
Breakaway and break-in points can be identified as follows.
For any point on the root locus:
G(s)H(s) = KN(s)/D(s) = -1
hence
K = -D(s)/N(s)
so that
d(K)/ds = 0
gives the minimum & maximum value of K with varying s.
Only answers with pure real values are of interest (s=σ) and these
must satisfy rule 2.
Ex:

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
7. Complex Pole (or Zero) Angle of Departure
The angle condition can be used to determine the direction that a
locus branch leaves a complex pole, or enters a complex zero.
Recall that:
W N

∑ ∠z − ∑ p
c =1
c
h =1
h = ϕ 1 + ϕ 2 + K − φ1 − φ 2 − φ 3 − K = (1 + 2 h)π

A test point is chosen very close to, but not exactly on, the complex
root of interest (“r”) so that all angles between the remaining roots
and this test point are equivalent to the angles measured between
these roots and “r”.
( s + z1 )
Ex: Given G( s) H ( s) =
s( s + p 2 )(s + a − jb)( s + a + jb)

Thus to find the angle of departure for complex pole P3 (-a+jb):

ϕ 1 − (φ1 + φ 2 + φ 3 + φ 4 ) = −180 o

jω
-p3

-z1 -p2 -p1 σ

-p4
8. The Imaginary Axis Crossing Point
“Place s = jωc in the characteristic equation and solve for ωc.”

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 Control in the Frequency Domain
Bode and Polar Plots of Type 0-2 Transfer Functions
Type 0: Type 1:
Bode:
K K (1 + jωTa 2π )
G ( jω ) H ( j ω ) = G ( jω ) H ( jω ) =
(1 + jωT1 2π )(1 + jωT2 2π ) jω (1 + jωT1 2π )

K=10, T1=2π/10, T2=2π/100 K=100, Ta=2π/100, T1=2π/10

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Polar/Nyquist: 90° 90°

−1 0° −1 0°
±180° ±180°

−90° −90°

Type 2:
Bode: (Example)
K2
G ( jω ) H ( jω ) =
( jω ) 2 (1 + jωTa )

Ex Fig 8.10
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Polar: (Additional examples of typical type 2 polar plots)

(stable) (unstable)

Ex Fig 8.16, 8.18

(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
 Nyquist Stability in the Frequency domain
The Nyquist criterion applied to a CLTF states that:
“For stability; none of the roots of the characteristic equation must
lie on or to the right of the imaginary axis in the s-plane.”
This rule can be translated into the frequency domain using
Cauchies Theorem with a closed loop contour “c” incorporating the
whole RHP ( ∫ f ( z)dz = j 2π ∑ residues ).
c

In practice this rule must applied to the solution of the CLTF’s

characteristic equation. Thus if:
B( s) = 1 + G ( s) H (s) = 0
Any zero of B(s) inside the RHP will create a +2π rotation, while
any pole of B(s) will create a −2π rotation. The total number of
rotations of the poles and zeros are respectively: PRb, ZRb.
There will be a net number of rotations related to the number of
poles and zeros of B(s) inside the RHP. NR= ZRb+ PRb.

Since B(s) cannot have any zeros (poles of CLTF) in the RHP for
system stability, the requirement for stability is:
Z Rb = PRb − N R = 0

However, the poles of B(s) are simply the poles of the OLTF, while
the zeros of B(s) are the poles of the CLTF.
Proof by example:
K K
G ( s) H (s) = CLTF =
s 2 (1 + sT ) s 2 (1 + sT ) + K

s 2 (1 + sT ) + K
B( s) = 1 + G ( s ) H ( s ) = =0
s 2 (1 + sT )

So that our new stability rule can be rewritten as:

PR ( CLTF ) = PR (OLTF ) − N R = 0

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
In the frequency domain this test for stability only requires a plot of
B(jω) as "−∞ < ω < +∞" and we need to determine the net number of
rotations about the origin (B(jω)=0).

Alternatively since B(jωω)−−1 = G(jω

ω)H(jω ω), we can plot G(jω
ω)H(jω
ω)
and look at the net rotations about the −1 point:

Type 0 system stability.

k
Given: G ( jω ) H ( jω ) = ... (1)
(1 + jωT1 )(1 + jωT2 )(1 + jωT3 )
plotting this results in:
90°

±180° −1 0°

NR =
from (1): PR(OLTF)= 0,
thus: PR(CLTF)= 0 −
−90°

N is determined by noting the direction of increasing ω at intersections of an

arbitrary radial line from −1 with the plot of G(jω)H(jω).

A Condition on the Nyquist Stability Theorem

A simple condition, necessary for valid application of the Nyquist
Theorem requires that:
“The contour integral “c” must never pass through any poles or
zeros of B(s)”.
In systems ≥ Type 1, stable poles lie at the origin. Thus the path of
the contour in the s-plane must be modified so that it does not pass
through the origin. This is accomplished by taking an infinitely
small detour around the origin in the s-plane from –j0 → +j0.
Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
The effect of this in the frequency domain on a type n system is to
add n infinitely large cw semi-circles to the Nyquist plot connecting
the –j0 and +j0 points.
Proof:
The modification to the contour s=εejθ (-90<θ<90) is a ccw semi-circle
with infinitely small radius ε. For a type n system near the origin as
s→0:
K K K − jnθ
lim G ( s ) H ( s ) = = = e
jω → 0 ( jω ) n (εe jθ ) n ε n
A type 1 example
k
G ( jω ) H ( jω ) =
jω (1 + jωT1 )(1 + jωT2 )

Ex Fig 8.27, 8.29, 8.30

(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3
Type 2 examples
k (1 + jωTa )
G ( jω ) H ( jω ) =
( jω ) (1 + jωT1 )(1 + jωT2 )
2

where T a > T1 + T2 + T3

Ex Fig 8.33
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Conditionally stable systems can also arise where both increases and
decreases in k will cause instability (as shown below)

Ex Fig 8.34
(D’Azzo & Houpis “Linear Control System Analysis & Design”)

Final notes:
Only the positive part of G(jω)H(jω) for 0<ω<+∞ need be plotted
since the polar plots are symmetrical about the real axis.
If the curve of G(jω)H(jω) passes through -1, Nyquist fails to tell us
anything since this is equivalent to zeros on the Imaginary axis,
however this also means the CL system will have a term which
produces sustained oscillations. This condition is considered
unstable.

Created by Dr Grant Covic for paper: Systems and Control 1998: Document 2 of 3