Chapter 4: Feedback Control System

Characteristics
• Open vs. Closed-loop control
• Error Signal Analysis
• Sensitivity of feedback
• Propagation of disturbances in the
feedback loop
• Robustness and Sensitivity
• Steady-State Error
1
Open-Loop Control
An open-loop (direct) system operates without feedback and
directly generates the output in response to an input signal.

Gc (s ) G(s)
R(s) Y(s)
U(s)
Actual output
response

• Perfect servo control would be achieved if Y(s)=R(s) , i.e. if

1
Gc ( s )G ( s ) = 1 or Gc ( s ) =
G( s)
Perfect control requires the controller to be the inversion of the plant
(if the model is perfect and there are no disturbances)
2
Why Prefect Control Is Not Practical
The plant inversion
• Relies on very accurate model
• Requires the plant and its inverse to be stable
1 1 D( s)
G=
c (s ) = =
G (s ) N ( s ) / D( s ) N ( s )
• Poor at rejecting disturbances

3
The Closed-Loop Control
A closed-loop system uses a measurement of the output
signal and a comparison with the desired output to
generate an error signal that is applied to the actuator.

Input Output

Y (s) G (s)
T=
(s) =
R( s ) 1 + H ( s )G ( s )

4
Error Signal Analysis Y ( s) G( s)
T ( s) = =
R( s) 1 + G ( s) H ( s)
E ( s ) = R ( s ) − Y ( s ) = (1 − T ( s )) R ( s )
( E ( s ) → 0 as T (s ) → 1)
?

Now Consider general case:

External disturbance
Set-point
Output
Input
Measurement noise

T(s)=? E ( s ) = ?
5
Error Signal Analysis
External disturbance
External disturbance
Output Set-point Output
n
Set-point
Measurement noise
1 =-1 Measurement noise
(linear system)

Y (s ) =YR + YTd + YN = TR R(s ) + TT Td (s )+ TN N (s )

d

(Three inputs one output)

6
Error Signal Analysis
∑
External disturbance
Yi ( s ) PK Δ K
=Ti = K
, Y (s ) = ∑ Ti Ri
Ri ( s ) Δ Set-point Output
One loop: Δ =−
1 GcG ( − H ) =+
1 GcGH

Three inputs and one output: three paths

Path gain for R(s): GcG Measurement noise
Path gain for N(s): − HGcG
Path gain for Td(s): G(s)
Gc G G HGc G
∑ Ti Ri 1 + HG G R(s) + 1 + HG G Td (s) − 1 + HG G N (s)
Y (s ) ==
c c c

Gc G G Gc G
Y (s ) H ( s ) =1 = R(s ) + Td (s ) − N (s )
1 + Gc G 1 + Gc G 1 + Gc G
7
Error Signal Analysis (H(s)=1)
External disturbance
Consider special case H(s)=1
Set-point Output

GcG G GcG
Y ( s) = R( s) + Td ( s ) − N ( s)
1 + GcG 1 + GcG 1 + GcG -1

=-1 Measurement noise

Tracking error:

E ( s) = R( s) − Y ( s)
1 G GcG
E ( s ) = R( s) − Y ( s) = R( s ) − Td ( s ) + N ( s)
1 + GcG 1 + GcG 1 + GcG

8
Error Signal Analysis (H(s)=1)
1 G GcG
E ( s) = R( s ) − Td ( s ) + N ( s)
1 + GcG 1 + GcG 1 + GcG
Let L( s) = Gc ( s)G ( s)

Tracking error: -1

1 G( s) L( s )
E ( s) = R( s ) − Td ( s ) + N (s)
1 + L( s ) 1 + L( s ) 1 + L( s )
1 L( s )
Let S ( s ) = C ( s) =
1 + L( s ) 1 + L( s )

E (s) =
S (s)R(s) − S (s)G (s)Td (s) + C (s)N (s)
9
Error Signal Analysis (H(s)=1)
E (s) =
S (s)R(s) − S (s)G (s)Td (s) + C (s)N (s)
To minimizing E(s), for given G(s) Output

S(s) and C(s) must be small -1

But,
1 L( s )
S ( s) + C ( s) = + =1
1 + L( s ) 1 + L( s )

We cannot simultaneously make S(s) and C(s) small.

10
Trade-offs
E ( s ) = S ( s ) R( s ) − S ( s )G ( s )Td ( s ) + C ( s ) N ( s)
Consider that
1 Gc (σ + jω)G
S (=
s) + C (s) + = 1 for 0 ≤ ω ≤ ∞
1 + Gc (σ + jω)G 1 + Gc (σ + jω)G

S (σ + jω)) ω<ω + C (σ + jω) ω<ω =

1 (low frequencies)
0 0

In most case, the disturbance occurs at low frequencies and the measurement
noise happen to be at high frequencies. A low pass filter will be useful to filter
out the noise so that N(s) is smaller at lower frequency.

The conflict can be solved in the design phase by making the loop
gain, L(s) = Gc (s)G (s), large at low frequency. (generally associate
with the frequency of disturbance).
11
Low Pass filter
V2 (s ) I ( s ) / Cs
=
G (s ) =
V1 (s ) RI ( s ) + I ( s ) / Cs
1 1
= =
RCs + 1 RC (σ + jω) + 1
Low pass filter V (s ) = V1 (s ) (1st Order system)
RC (σ + jω) + 1
2

V2 ↓ as ω ↑
N effective = Glow− pass N ( s )
|| N effective ||≈|| N ( s ) ||ωlow ≈ ε

E ( s ) =S ( s )[ R ( s ) − G ( s )Td ( s )] + C ( s ) N ( s )
Glow− pass
≈ S ( s )[ R ( s ) − G ( s )Td ( s )] + C ( s )ε
≈ S ( s )[ R ( s ) − G ( s )Td ( s )] + ε 12
System Sensitivity
Sensitivity function is defined as the ratio of the change in the system
transfer function to the change of a process transfer function
ΔT
% change in T
SG =
T
= T
% change in G ΔG
G
(or parameter) for a small incremental change. ( ∆ → 0)
∂T
%change in T T ∂ ln T
=T
SG = =
%change in G ∂G ∂ ln G
G
G ∂T
SG =
T

T ∂G
13
Table of Derivatives

14
Sensitivity of the Open-loop System

T (s ) = Gc (s )G (s )
G ∂T
S T
=
T ∂G
G

∂T
= Gc ( s )
∂G
G GGc
SGT = Gc = =1
T T

15
The Sensitivity of the Feedback System to
Change in the G(s)
G( s)
T ( s) =
1 + GH ( s )

∂T G (1 + GH ) − GH G 1 G
=
S T
= =
∂G T (1 + GH ) T (1 + GH ) 2
G 2
G
(1 + GH )
1
SGT = || SGT ||≤ 1
1 + G( s) H ( s)

16
The Sensitivity of the Feedback System to
Change in the H(s)
G (s )
T (s ) =
1 + GH (s )

∂T H − G 2
H −G (s )H (s )
=
S HT = =
∂H T (1 + GH ) 2 G 1 + G (s )H (s )
(1 + GH )
>
If G ( s ) H ( s ) 
>> 1, ?
>
− G( s) H ( s)
S HT = ≈ −1 Good or bad?
1 + G ( s ) H ( s ) GH >>1
Bad!!
17
Sensitivity of the Closed-loop System

Consider the case of Td(s)=0 , N(s)=0, and H(s)=1

Gc ( s )G ( s )
T ( s) = .
1 + Gc ( s )G ( s )
∂T G Gc G
=
S =
T
1
∂G T (1 + Gc G ) 2
G
Gc G SGT =
(1 + Gc G ) 1 + Gc ( s )G ( s )

18
Sensitivity of the Closed-loop System

Gc (s )G (s ) 1
T (s ) = SGT =
1 + Gc (s )G (s ) 1 + Gc ( s )G ( s )

Gc (s )G (s ) 1
T (s ) + S (s ) =
T
+ =
C (s ) + S (s ) =
1
G
1 + Gc (s )G (s ) 1 + Gc (s )G (s )

T (s ) + SGT (s ) =
1
19
Sensitivity of the Closed-loop System
SGT = S ( s ) =
1
=
1 s = σ + jω
1 + Gc ( s )G ( s ) 1 + L( s )

Typically, the loop gain GcG is designed to be large at low

frequencies and small at high frequencies, hence
S (0) ≈ 0 while S ( ∞) = 1
Noted that T ( s) + T
SG ( s ) = T ( s) + S ( s) = 1
This implies T (0) = 1 and T (∞) → small
= ω 0 ω→∞

20
More on the Sensitivity Function

• When G ( s ) = G (α , s )
Gc G ( s, α )
T=
1 + Gc G ( s, α )

SαT = SGT SαG

21
Example
Find SaT

How to reduce SaT ?

Solution T ( s ) = K / s( s + a ) = K
1 + K / s( s + a ) s 2 + as + K
a δT a − Ks
S =S S
T T G
= = ( 2 )
a G a
T δa ( K
) ( s + as + K ) 2

s 2 + as + K
− as
= 2
s + as + K K SaT

22
Example – Feedback Amplifier
+ + vo = − K a vin G ( s ) = G (α , s )
Open-loop vin
Gain
T = −Ka
SαT = SGT SαG
v0
−Ka
− −
S KT = 1
a

vin v0
vin + −Ka
Gain R1 v0
Rp
Closed-loop −Ka R2 +

R2 −Ka K a ∂T
β = R= 1
p R1 + R2 T = s KT a = = K a ↑⇒ S K ↓
T
Rp 1 + Ka β T ∂K a 1 + K a β a
1
If=
Ka =
4
10 β 0.1
T
SK = ≈0.001
a 1 + 103
1927 H. S. Black conceives of the negative feedback amplifier and H. W. Bode
analyzes feedback amplifiers. 23
Disturbance Signals in a Feedback Control Systems

• A disturbance signal is an unwanted input signal that affect

the output signal.

•An important effect of feedback system is the control and

partial elimination of the effect of disturbance signals.

• Many control system are subject to extraneous disturbance

signal that cause the system to provide an inaccurate output.

24
Disturbance Rejection
E (s ) =
S (s )R(s ) − S (s )G (s )Td (s ) + C (s )N (s )
Set-point Output
When R(s)=N(s)=0, E ( s) = R( s) − Y ( s)

G( s)
E ( s ) = − S ( s )G ( s )Td ( s ) = − Td ( s )
1 + Gc ( s )G ( s )
= -1
Gc ( s ) ↑ ⇔ E ( s ) ↓

Example: Let us consider the speed control system for a steel mill.

25
Disturbance Rejection

An armature-controlled DC motor
26
Disturbance Rejection–Open-loop
E ( s) = R( s) − Y ( s) = R( s) − ω ( s), Td

Let R(s)=0, Td(s)=D/s (step function) .

0 − ω (s) =
E (s) = −ω ( s )
1 1/ (Js + b)
T= ∑ Pk ∆k −ω(s ) =
E (s ) = TTd Td = Td (s )
∆ 1 + (K m K b / Ra ) / (Js + b)
1
= Td (s )
Js + b + K m K b / Ra
 1 D
Steady-state error: lim e(t ) = lim sE ( s ) = lim s 
t →∞ s →0 s→0  Js + b + K m K b / Ra  s

D
D = −ωo (∞).
=
eopen −loop (∞) = b + ( K m Kb / Ra )
b + (K m K b / Ra ) 27
Disturbance Rejection–Closed-loop

1
Ka

Ka Km 1 K
G1 = G2 = H = Kt + b
( s ) R( s ) − ω ( s ),
E= Ra Js + b Ka
 K  K K K
E ( s ) = −ω ( s ) |R ( s ) =0
Ka Km
G1 ( s ) H ( s ) =  K t + b  ≈ a m t ,
Ra  Ka  Ra
G2 ( s )
E ( s ) = −ω ( s ) = Td ( s ).
1 + G1 ( s )G2 ( s ) H ( s )
 1 D
Steady-state error: =
lime(t ) =
limsE (s ) lim s  
t →∞ s →0 s →0  Js + b + K m / R a (K t K a + K b  s
)
D
Ra D = ;
eclosed −loop (∞) ≈ b + ( K m / Ra )( Kt K a + Kb )
K m Kt K a 28
Disturbance Response –Open-loop vs. Closed-loop
D Ra D
eopen −loop (∞) = eclosed −loop (∞) ≈
b + (K m K b / Ra ) K m Kt K a

eclosed −loop (∞) Ra b + K m K b (the ratio is usually less than 0.02)

=
eopen −loop (∞) K m Kt K a
ec ( ∞) ≤ eo ( ∞) ≈ 0.02eo ( ∞)
The speed-torque curves for The speed-torque curves for
the open loop system the closed-loop system

29
Measurement Noise Attenuation
E (s )= ER & N ( s ) + EN ( s )= S (s )R(s ) − S (s )G (s )Td (s ) + C (s )N (s )
When R(s)=Td(s)=0,
Gc ( s )G ( s ) Output
E ( s) = C ( s) N ( s) = − N ( s) Set-point
1 + Gc ( s )G ( s )
In most case, the disturbance occurs at low
frequencies and the measurement noise happen H(s)= -1
to be at high frequencies.
Gc (s) ↑ at low frequency ⇔ ER& D ( s) ↓
Gc (s) ↓ at high frequency ⇔ E (s) ↓
A low pass filter will be useful to filter out the noise so that
N(s) will have smaller magnitude at low frequency region.
30
Low Pass filter V2 (s ) I ( s ) / Cs
=
G (s ) =
V1 (s ) RI ( s ) + I ( s ) / Cs
1
= 1
RCs + 1=
RC (σ + jω) + 1
Low pass filter V1 (s ) (1st Order system)
V2 (s ) =
RC (σ + jω) + 1
V2 ↓ as ω ↑
N effective = Glow− pass N ( s )

|| N effective ||≈|| N ( s ) ||ωlow ≈ ε

E ( s ) =S ( s )[ R ( s ) − G ( s )Td ( s )] + C ( s ) N ( s )
Glow− pass
≈ S ( s )[ R ( s ) − G ( s )Td ( s )] + C ( s )ε
≈ S ( s )[ R ( s ) − G ( s )Td ( s )] + ε
31
Noise in the Measurement Loop
For R(s)=0 and Td(s)=0 R( s) + G1 ( s ) G2 ( s ) Y (s)

−G1G2 H 2 ( s ) −

Y ( s) = N ( s ).
1 + G1G2 H 1 H 2 ( s )
Sensor

H 2 (s) H1 ( s )
+

G1G2 H1H 2 >> 1 Y ( s ) ≅ −1 +

H1 ( s)
N ( s ). H
1 ↑ Yn ↓ N (s)

• This is equivalent to maximizing the signal-to-noise

ratio of the measurement sensor.
• This necessity is equivalent to requiring that feedback
element, H(s), be well designed and operated with
minimum noise, drift and parameter variation.
∂T H G −H −GH T
SH ≈ −1
T
SH = • =( )2 • = .
∂H T 1 + GH G /(1 + GH ) 1 + GH
32
Control of the Transient Response
• One of the most important characteristics of control system is
their transient response.

• The transient response is the response of a system as a function

of time.

• To achieve the desired transient response, the controller must

be adjusted.

33
Example: Speed Control---Open-loop
ω( s) Km
=
Va ( s ) ( Ra + La s )( Js + b) + K b K m
K1
= (τ 2 << τ 1 )
(τ 2 s + 1)(τ 1s + 1)

ω( s) K1
= G( s) ≈
Va ( s ) (τ 1s + 1)
Km Ra J
K1 = τ1 = .
( Ra b + K b K m ) ( Ra b + K b K m )

k2 E
Va ( s ) = , (Step function)
s
K 1k 2 E
ω ( s ) = G ( s )Va ( s ) = .
(τ 1 s + 1) s
−t /τ1 Armature current control dc motor
ωopen
= −loop (t ) K1 ( k2 E )(1 − e ).
34
Example: Speed Control -- Closed-loop

K1
G ( s) =
(τ1s + 1)

ω (s) K a G (s) K a K1 K a K1 / τ1
T(s)= = = =
R( s) 1 + K a Kt G ( s) τ1s + 1 + K a Kt K1 s + [(1 + K K K ) / τ ]
a t 1 1
k2 E
ω ( s ) = T ( s ) R( s ) = T ( s )
s
K a K1
ω (t ) = ( k 2 E )(1 − e − pt ), p = (1 + K a K t K1 ) / τ 1 K a Kt K1 >> 1
1 + K a K t K1

1   −( K a Kt K1 )t 
ωclosed −loop (t ) ≈ ( k2 E ) 1 − exp   .
Kt   τ1  
35
Example: Speed Control-Improving Transient Response
−t /τ1 1   −( K a Kt K1 )t  
ωopen
= −loop (t ) K1 ( k2 E )(1 − e ). ωclosed −loop (t ) ≈ (k2 E ) 1 − exp   .
Kt   τ1  
Ra J
τ1 = =constant
( Ra b + K b K m )

Transient response
is the response of a
system as a function
of time.

36
Example: Speed Control-Parameter Variation
Sensitivity of closed-loop system to a variation in the motor
constant K m is
1 1 [ s + (1/ τ1 )]
S KT = SGT S KG = = = .
m m 1 + GH ( s ) 1 + K a Kt G ( s ) [ s + ( K K K
a t 1 + 1) / τ 1 ]

For typical values

( s + 0.10)
T
SK = .
m ( s + 10)

T
SK ≅ 0.1.
m s = jω = j1

37
Disturbance & Parameter Variation
Td 1 1
T
SG = ≅ .
R( s) + G1 ( s ) G2 ( s ) Y (s) 2 1 + G1G2 H ( s ) G1G2 H ( s )
−

Y (s) −G2 ( s ) −1
= ≅
Sensor
.
Td ( s ) 1 + G1G2 H ( s ) G1 H ( s )
H 2 (s) H1 ( s )
+
+

N (s)

G1 ( s ) = K a T
SG ↓
2
Y (s) |Td ↓
Td ( s )
+
R( s) + G (s) + Y (s)

− Y (s) 1
= = SG
T
H (s) Td ( s ) 1 + GH ( s )

38
Error and Actuating Error
Ea (s )
R(s) Y(s)

Ea (s ) is the actuating error and exists only inside a

closed loop system:
Ea ( s ) ≡ R( s ) − H ( s )Y ( s )
E ( s ) = R( s ) − Y ( s ) ≠ Ea ( s ) if H ( s ) ≠ 1

Using E(s) not Ea (s) in the final value theorem for stead-state error.

39
Error and Actuating Error
• Regardless of the open-loop or the closed-loop, the
error of the system is always defined as the difference
between the desired input and the output of the system:
E ( s) ≡ R( s) − Y ( s)
For open loop: Y ( s) = G ( s) R( s) Eopen −loop (s )= (1 − G (s ))R(s )

For closed-loop: Y ( s ) = T ( s ) R( s ) Eclosed-loop (s )= (1 − T (s ))R(s )

Steady-State Error
The tracking error is Eopen −loop= (1 − G (s ))R(s )
1
Eclosed −loop (s ) = R(s )
T=
d & N =
0, H (s ) 1 1 + Gc G
The steady-state error is the error after the transient
response has decayed, leaving only the continuous
response.
1
Eopen −loop (s )= (1 − G (s ))
For a step input ( R(s)=1/s): s
1 1
Eclosed −loop (s ) =
1 + Gc G s
Steady-state error can be estimated using the final value theorem
lim e(t ) = lim sE ( s )
t →∞ s →0
41
Steady-State Error
lim e(t ) = lim sE ( s )
t →∞ s →0
1
For R(s)=1/s: eopen −loop (∞) = lim
s →0
s(1 − G (s)) = 1 − G (0)
s
1 1 1
eclosed −loop (∞) =lim s = lim
s → 0 1 + G (s )G (s ) s s → 0 1 + G (0)G (0)
c c

If we want the steady state error to go to zero for a finite G (0),

we need to design a controller Gc (0)to be infinite, i.e. Gc (0)
must contain one integrator 1/ s.

1 = lim= 1 s
eclosed −loop (∞) =lim =
lim 0
1 s→0 s + G (0)
s→0 1 + G (0)G (0)
+
s→0
c Gc (0) =
1 1 G (0)
s s
42
Example: Steady-State Error

G(s)

1 K 1 Gc ( s )G ( s )
R( s) = G( s) = Gc ( s ) = T ( s) =
s τs + 1 τs + 1 1 + Gc ( s )G ( s )
1
eo (∞) = lim s(1 − Gc ( s )G ( s )) = 1 − Gc (0)G (0) = 1 − K
s →0 s
=
Gc (0) 1=
G (0) K
1 K 1
ec (∞) = lim s (1 − T (s )) = 1 − T (0) = 1 − =
s →0 s 1+ K 1+ K

43
Example: Steady-State Error
eopen −loop (∞) = 1 − K
G(s)
1
eclosed −loop (∞) =
1+ K
For open loop, K=1, the steady state error is zero.
For closed-loop, it is prefer to have large gain K.
In practice, K can vary. Consider the case of ΔK/K (a
10% change), when K is select to be 100 then, we have
1
eclosed −loop (∞) = Δeopen −loop (∞) =0.1
101 K =100
1 1
Δeclosed −=
loop (∞ ) =- 0.0011
101 91
The relative change in the closed-loop is two order of magnitude
44
lower than that of the open-loop.
E4.11: Closed-loop System with Non-unity Feedback
K
G( s) =
( s + 10)
14
H ( s) = 2
s + 5s + 6

(a) Compute the transfer function T(s)=Y(s)/R(s).

(b) Computer E(s) and determine the steady-state error for R(s)=1/s.
(c) Computer the transfer function T(s)=Y(s)/Td(s). E(s) when Td(s)=1/s
(d) Computer the sensitivity S KT .
(a) The closed-loop transfer function is:
G (s) K ( s 2
+ 5s + 6)
T (s ) = = 3
1 + H ( s )G ( s ) s + 15s + 56 s + 60 + 14 K
2

45
E4.11: Closed-loop System with Non-unity Feedback
K
G (s ) =
(s + 10)
14
H (s ) = 2
s + 5s + 6

(b) With E(s)=R(s)-Y(s) we obtain

 G (s)  1 − G ( s )(1 − H ( s ))
E (s ) =
1 − 1+G (s )H (s )  R ( s ) = R( s)
  1+G (s )H (s )
s 3 + (15 − K ) s 2 + (56 − 5 K ) s + (60 + 8 K ) 1
⋅ .
s + 15s + 56 s + 60 + 14 K
3 2
s
Then, using the final value theorem we find
60 + 8 K
lim sE (s ) = .
s →0 60 + 14 K 46
E4.11: Closed-loop system with non-unity feedback
K
G( s) =
( s + 10)
14
H ( s) = 2
s + 5s + 6

(c) The transfer function from the disturbance Td(s) to the input is
1 s 3 + 15s 2 + 56 s + 60
Y (s) = Td ( s ) = 3 Td ( s ).
1+G (s )H (s ) s + 15s + 56 s + 60 + 14 K
2

The steady-state error to a unit step disturbance is ,

lim sE ( s ) =
lim s ( R ( s ) − Y ( s )) =
− lim sY ( s )
s →0 s →0 s →0

s 3 + 15s 2 + 56 s + 60 1 60
=− lim s 3 ⋅ =− .
s →0 s + 15s + 56 s + 60 + 14 K s
2
60 + 14 K 47
E4.11: Closed-loop System with Non-unity Feedback
K 14
=G (s ) = , H (s )
(s + 10) s 2 + 5s + 6
K
T=
( s + 10)( s 2 + 5s + 6) + 14k

(d) The sensitivity is

∂T K ∂T ∂G K
=s =T

∂K T ∂G ∂K T
K

1 1 K (1 + GH ) 1
= =
(1 + GH ) 2 s + 10 G 1 + GH

(s + 10)( s 2
+ 5s + 6)
SK =
T

(s + 10)(s 2 + 5s + 6) + 14 K
48
Advantages of the Closed-loop Feedback Control
•Decreased sensitivity of the system to variations in the
parameters of the process.

•Improved rejection of the disturbances.

•Improved reduction of the steady-state error of the system.

•Easy control and adjustment of the transient response.

49
The Cost of Feedback
•Increased Number of Components and Complexity

•Loss of Gain
Open loop gain Gc ( s )G ( s )

Gc ( s )G ( s ) 1
Closed-loop gain = Gc ( s )G ( s ) < Gc ( s )G ( s )
1 + Gc ( s )G ( s ) 1 + Gc ( s )G ( s )
•Instability

•Trade off in Design

50
Eurotunnel (design example)

51
Building the Eurotunnel

Proportional and
derivative controller
(PD)
t ) Ke + 11e
uc (=

Design objective: K=? eTd

52
Building the Eurotunnel

Using Mason’s formula, we have

Y ( s ) = TR ( s ) R( s ) + TTd ( s )Td ( s )
K + 11s 1
= 2 R( s) + 2 Td ( s ).
s + 12 s + K s + 12 s + K
• Note that the linear system satisfies the superposition rule.
Therefore, one can find the transfer functions separately for
input R(s) and disturbance Td(s). The total output is the sum
of both. 53
Block Diagram Approach
• Td(s)=0

Gc ( s )G ( s ) ( K + 11s )/s( s + 1) = K + 11s

T ( s) = =
1 + Gc ( s )G ( s ) 1 + ( K + 11s )/s( s + 1) s + 12 s + K
2

• R(s)=0
TTd ( s ) =
G( s) 1/s( s + 1)
=
Td(s) 1 + Gc ( s )G ( s ) 1 + ( K + 11s )/s( s + 1)
1
= 2
s + 12 s + K
Y ( s ) = TR ( s ) R ( s ) + TTd ( s )Td ( s )
K + 11s 1
= 2 R( s ) + 2 Td ( s )
s + 12 s + K s + 12 s + K 54
Step Responses
K=180

yr (t) K=20

Time (seconds)

K + 11s 1
Y ( s) = 2 R( s) + 2 Td ( s )
s + 12 s + K s + 12 s + K
55
Steady-State Error
The steady-state error of the system to a unit step input R(s)=1/s is
1 K + 11s
lim e(t ) = lim s(1 − T ( s )) = lim(1 − 2 )=0
t →∞ s →0 s s →0 s + 12 s + K
The steady-state value of y(t), when the disturbance is a unit step input
Td(s)=1/s and R(s)=0, is
1 1 1
lim yTd (t ) = lim s(TTd ( s )) = lim( 2 )=
t →∞ s →0 s s→0 s + 12 s + K K
The sensitivity of the system to a change in the process G(s) is
1 s( s + 1) s ( s + 1)
SGT = = = 2
1 + GcG s( s + 1) + 11s + K s + 12 s + K

When |s|<1, SG s
T
≈ .
K
56
Error Responses

e(t)

K=20

K=180

Time (seconds) 57
Disturbance Responses
K=20

yd(t)

K=180

Time (seconds) 58
Common Step Response Parameters

Typically, we increase the gain in the loop to obtain a faster response and
increased bandwidth, risking instability and saturating actuators. 59
Well Tuned Response
Yr(t)

Yd(t)

Figure 4.19 The response y(t) for a unit step input (solid line)
and for a unit step disturbance (dashed line) for K = 20.
60
Response of the Boring System
The over design results are summarized in Table 4.1. The
preferred loop gain is K=20.

K=20 is a reasonable design compromise.

61
Sequential Design Example: Disk Drive Read System
Goal: position accuracy 1 µm. Moving the head from track a to track
b within 50 ms with subject to external shocks and vibration.

Td(s)=0 K a G1 ( s )G2 ( s ) = 5000 K a

T ( s) =
1 + K a G1 ( s )G2 ( s ) s + 1020s + 2000s + 5000 K a
3 2

L( s ) = K a G1 ( s )G2 ( s )
The forward path transfer function has three more poles than
zeros and has no poles in the right half plane (is stable).

62
Sequential Design Example: Disk Drive Read System
Goal: position accuracy 1 µm. Moving the head from track a to track
b within 50 ms with subject to external shocks and vibration.

Td(s)≠0 R(s)=0 G2 (s )
TTd (s ) =
1 + K a G1 (s )G2 (s )
1
= 3
s + 1020 s 2 + 2000 s + 5000 K a

63
Disk Drive Read
• Steady-state error when Td(s)=0 and R(s)=1/s:
E ( s ) = R( s ) − Y ( s ) = (1 − T ( s )) R( s )
1
E ( s ) = (1 − T ( s )) R( s ) = R( s )
1 + K a G1 ( s )G2 ( s )
 1 1
lim e(t ) = lim s 
t →∞ s →0 1 + K a G1 ( s )G2 ( s ) 
s
1
= lim =0
s →0 500
1 + Ka
( s + 1000) s( s + 20)
the integrator
(s + 1000)s (s + 20)
= lim
= 0
s → 0 (s + 1000)s (s + 20) + 500 K
a 64
Disk Drive Read
Note that S_S error is
zero for all values of
K as a result of the
integrator in the loop.

Goal: position accuracy 1 µm.

Moving the head from track a
to track b within 50 ms.

Still have a long way to ago!

500ms 65
Summary
The fundamental reasons for using feedback, despite its cost and
Additional complexity, are the follows
•Decrease in the sensitivity of the system to variations in the
parameters of the process.
•Improvement in the rejection of the disturbances.

•Improvement in the reduction of the steady-state error of the system.

•Ease of control and adjustment of the transient response of the system.

66
MECH3610 HOMEWORK 5
P 5.1 It is important to ensure passenger comfort on ships by stabilizing the ship’s
oscillations due to waves. Typical ship stabilization systems employ fins to stabilize
external torque from the sea’s dynamics. A simple diagram of the system is shown in
Fig. 5.1. The rolling motion of a ship can be modeled as an oscillating pendulum with
a deviation from the vertical of 9 degrees and a typical period of 3 seconds.
The transfer function is G(s) = s + 2ξωω s + ω
2
n
2 2
n n

Where natural frequency ωn =3 rad/s and damping ratio ξ=0.2. The expected amplitude of
rolling rotation caused by sea’s dynamics is 18 degrees and the oscillations would decay
due to damping effect.
Determine and compare the open-loop and closed-loop system for
•Sensitivity to changes in the actuator constant K a and the roll sensor K1 .
•The ability to reduce the effects of step disturbances of the waves. Note that the desired roll θd
is zero degrees.
Hints: a) Open-loop

Closed-loop
For Td = 0,

? ?

b) θdesired ( s ) − θ ( s ) =
E (s) = −θ ( s ) θdesried
= ( s ) 0,=
θ ( s ) T ( s )Td ( s )
ess = lim sE ( s ) What is the T(s) ?
s →0

Open-loop ess = lim sEo ( s )

s →0

Closed-loop ess = lim sEclosed −loop ( s )

s →0
P 5.2 One important objective of the paper-making process is to maintain uniform consistency
of the stock output as it progresses to drying and rolling. A diagram of the thick stock
consistency dilution control system and its block diagram are shown in Fig. 5.2. The amount of
𝐾𝐾 1 𝐴𝐴
water added determines the consistency. Let 𝐻𝐻 𝑠𝑠 = 1, 𝐺𝐺𝑐𝑐 𝑠𝑠 = , 𝐺𝐺 𝑠𝑠 = .R s = .
10𝑠𝑠+1 2𝑠𝑠+1 𝑠𝑠
Determine:
𝑌𝑌(𝑠𝑠)
(a)The closed-loop transfer function T s = .
𝑅𝑅(𝑠𝑠)
(b)The sensitivity 𝑆𝑆𝑇𝑇𝑇𝑇 .
𝐴𝐴
(c)The steady-state error for a step change in the desired consistency R s = .
𝑠𝑠
(d)Calculate the value of K required for an allowable steady-state error of 2%.
𝐾𝐾 𝐾𝐾 𝑠𝑠+𝑧𝑧
1 𝑠𝑠+𝑧𝑧
2 1.74(𝑠𝑠+2.65)(𝑠𝑠+4)
(e)If 𝐺𝐺𝑐𝑐 𝑠𝑠 = 𝐾𝐾𝑝𝑝 + 𝑖𝑖 + 𝐾𝐾𝑑𝑑 𝑠𝑠 = = what will be the effect on
𝑠𝑠 𝑠𝑠 𝑠𝑠
steady state error for a step change in the desired consistency
Hints: a) The closed-loop transfer function
Gc ( s )G ( s ) K 1
=
T (s) = ?= Gc ( s ) = , G ( s) b)
1 + Gc ( s )G ( s ) 10 s + 1 2s + 1
c)
Eclosed −loop ( s ) = R( s ) − Y ( s ) = (1-T(s))R(s)
ess = lim sEclosed −loop ( s )
s →0

d) | eSS |< 0.02 A =

as t → ∞ =
ess lim sEclosed −loop ( s ) f ( K )
s →0

What is the K value? 0.02 A >| ess |=

f (K )

e) = Gc ( s )G ( s ) 1 1.74( s + 2.65)( s + 4)
T (s) = ?= G (s) ,=
Gc ( s )
1 + Gc ( s )G ( s ) 2s + 1 s
Eclosed −loop ( s ) =
R( s ) − Y ( s ) =−
(1 T ( s )) R( s )
ess = lim sEclosed −loop ( s )
s →0
Ki
PID controller Gc ( s ) = K p + + K d s
s
Ki K d s + K p s + Ki
2
K d (s 2 + K p / K d s + Ki / K d )
Gc ( s ) = K p + + Kd s = =
s s s

K d ( s + z1 )( s + z2 ) K d ( s 2 + ( z1 + z2 ) s + z1 z2 ) 1.74( s + 2.65)( s + 4)
= =
s s s
where
K d =1.74, K p / K d = z1 + z2 = 6.65 and K i / K d = z1 z2 =10.6

=
Kd ? =
K p ?=
Ki ?
Appendix: Waterbed Effect
L(s)

L( s )
T (s) =
1 + L( s ) Closed-loop transfer Function
1
S ( s) = Sensitivity Function
1 + L( s )

L( s ) 1
T ( s ) + S ( s=
) + = 1
1 + L( s ) 1 + L( s )
72
Waterbed Effect

Bode's sensitivity integral, discovered by Hendrik Wade Bode,

is a formula that quantifies some of the limitations in feedback
control of linear parameter invariant systems. Let L be the loop
transfer function and S be the sensitivity function. Then the
following holds:
∞ ∞ 1 Np
π
∫0 ln S ( jω ) dω = ∫0 ln dω = π ∑ Re( pk ) − lim sL( s )
1 + L( jω ) k =1 2 s →∞
where are the poles of L in the right half plane (unstable poles).
73
Waterbed Effect
N L (s)
IfL( s ) = has no poles in the right half plane (is stable) and
DL ( s )
has at least two more poles than zeros, the equation simplifies
to: ∞ ∞ 1 Np
π
∫0 ln S ( jω ) dω = ∫0 ln dω = π ∑ Re( pk ) − lim sL( s )
1 + L( jω ) k =1
0
2 s →∞
0
N (s) 1 π π sN ( s ) π s π 1
(if ≈ 2 ⇒ limsL(s )= lim ≈ lim 2 =lim 0)
D( s) s 2 s →∞ 2 s →∞ D( s ) 2 s →∞ s 2 s →∞ s

∞
∫0 ln S ( jω ) dω = 0
This equality shows that if sensitivity to disturbance is suppressed
at some frequency range, it is necessarily increased at some other
range. 74
Waterbed Effect

∞
∫0 ln S ( jω ) dω = 0

In other words, the waterbed effect will exist for a system

if L has no poles in the right half plane (is stable) and has
at least two more poles than zeros.

75
Disk Drive Read & Waterbed Effect

Sensitivity function S

1 1
S= =
1 + GK 1 + L( s )

5000 K a
L(s ) =
s (s + 1000)(s + 20)

76