II.

FEEDBACK CONTROL SYSTEMS

II.A Advantage of Feedback

• Idealized performance of control system:

Output Y(s) = Desired performance R(s)
(Up to possibly a scaling constant)

Control
R(s) Y(s)
System
1 1

• Use Step input here for illustration

II.A.1. Performance Sensitivity to Parameter Variations
• Process parameters change due to environment, aging, etc
 Process transfer function G(s) changes with time
• Process G(s) inside block diagram of overall system
System transfer function T(s)

Process
R(s) G(s) Y(s)

• Hence, while R(s) staying the same, if we have

G(s) changes  T(s) changes  Y(s) changes
• System Sensitivity:
When G ( s )  G ( s )   G ( s )
T (s)  T (s)  T (s)
T
Def: S T  T T T G
G G Very small G
SG 
G G T

(% change in T over % change in G

for small incremental change)
• Lower/higher(?) sensitivity is better
 system output less affected by variations in parameter
variations
• Generally,
- Open loop system: G(s )
R(s) Y(s)

T ( s )  G ( s )  S GT  1

- Closed loop system: R(s) +

- G(s ) Y(s)

H (s )
G 1
T (s)   SG 
T

1  GH 1  GH
- Normally |1+GH(s)|>>1, for s=j at  values of interest (Later!)
 S GT  1 i.e., closed loop system has much lower
sensitivity (to variations in G)
• Example 4.1 Feedback Amplifier
(a) In open loop configuration

Gain
Vin  Ka Vo

V0   K aVin
T   Ka
S KT a  1  This means: for input Vin stays unchanged,
if Ka changes by 10% ,
then T changes by 10% ,
and Vo changes by 10%
 (b) In feedback configuration
R2 Vin Vo
 Gain
Rp
 Ka Rp
R2
V0   K a (Vin   V0 ) Vo
 Ka
T
(1  K a  )
1
S KT a 
(1  K a  ) Vin +
 Ka Vo
-
With   0.1 
K a  104
1
S KT a    for input Vin stays unchanged,
999
if Ka changes by 10% ,
10%
Vo insensitive to changes then T changes by   0.01%
999
in system parameter Ka and hence Vo changes by  10999%  0.01%
• Other kind of Sensitivity, e.g., Sensitivity due to variations
in H(s) for closed loop system:

R(s) +
- G( s ) Y(s)

H (s )
For 1+GH>>1
T
T 
SH T   T H   GH   1
H H T 1  GH
H
This means:
S GT  1  variation in G(s) has small effect on T(s)
S HT   1  variation in H(s) has strong effect on T(s)
Hence, important to have small variations in feedback loop for
closed loop system
• For T ( s ) depending on parameter i.e., T ( s )  T ( s ,  )
T
Sensitivity due S T  T  T 
to variation in    T


• For T ( s ) depending on G(s) depending on 

T   T G  G 
S T     ST SG
G 
 T  G   T G
• Example: Armature-controlled DC Motor
- Recall: Block diagram with  L   a (Reduced Model)

Speed Position
(s ) 1  (s )
Va ( s )
s

Nature induced
Feedback
- Let Output=angular speed (s) (not angle position )
- Va is input voltage to DC motor Step input of E (volts) with
Ek potentiometer factor k2
- In Textbook, V  2 (Eqn. 4.40 in textbook)
a
s
Here, we just let Va be input voltage to DC Motor
- For simplicity, we also set Kb=0 in deriving the equations
 - Open loop speed control: Disturbance
Td(s) Armature-controlled
Desired DC Motor
performance Va(s) Speed
R(s) Ka
(s )
Gain
controller Kb=0

- Ka is open loop controller gain to be designed

Controller
- Transfer function: Gain

KK R1
 ( s)  1 a R( s)  Td ( s )
( 1s  1) ( 1s  1)
where
Ra J Km Ra
1  , K1  , R1 
( Ra b  K m K b ) ( Ra b  K m K b ) ( Ra b  K m K b )
Kb=0 Kb=0 Kb=0
 - Closed loop speed control Disturbance
Td(s) Armature-controlled
Desired DC Motor
performance Va
R(s) Ka Speed
(s)
Gain
controller

Kb=0

- Ka is closed loop controller gain to be designed

- Transfer function:
K a K1 R1
( s)  R( s )  Td ( s ) (XX)
( 1s  1  K a K1 K t ) ( 1s  1  K a K1 K t )
- Typical values for speed control motor:
K1=2, Kt=1, Ka large (from 40100)
 Td(s)
- Derivation of (XX):
_
+ + q(s) G(s)
R(s) _
P(s) (s)

H(s)

- Td(s)=0, PG ( s )
( s)  R( s)
(1  PGH ( s ))
- R(s)=0,
 ( s )  G ( s ) q( s ) G( s)
 ( s)   Td ( s )
q( s )   PH ( s ) ( s )  Td ( s ) (1  PGH ( s ))

PG ( s ) G( s)
- Hence,  ( s )  R( s)  Td ( s )
(1  PGH ( s )) (1  PGH ( s ))
- (XX) obtained by P( s )  K a K m , G ( s )  1 , H ( s )  1
Ra Js  b
• Sensitivity analysis for Open Loop speed control

- Let disturbance torque Td  0, then

T(s)
K a K1
 ol ( s)  R( s)
( 1s  1)
Sensitivity with
* Ka -- control gain that we set respect to motor gain
* K1 -- motor gain in G(s) varies (slightly) with
each unit and time T
* T(s) changes with change in K1: S KT  T 1
1 K1
K
* Hence, 10% change in K1 1
 10% change in T(s)
 10% change in  ol ( s )  T ( s ) R( s )
Open Loop control performance sensitive to changes in
system parameter K1.
- Example: Let R(s)= step input and desired performance is
 ()  1
- With step input R(s)=1/s, output given by
K1K a K1K a
 ( s) 
ol
R( s) 
( 1s  1) s( 1s  1)
- Output steady state value using FVT

 ol ()  lim s ol ( s)  K a K1
s 0
 ol (t )
OL speed
1 control system K a K1
t t
1
- Based on our knowledge of K1, one can hence set K a 
K1
to achieve desirable steady state value  ()  1
ol
- Hence, if motor gain K1=2 according to our knowledge,
we can set gain Ka=0.5 so that
 ol ()  K a K1  0.5(2)  1
to satisfy the desired performance

- However, if K1 were to change by 10%, i.e., K1  2.2 ,

without we knowing it so we still use Ka=0.5 as before,
then
 ol ()  K a K1  (0.5)( 2.2)  1.1 10% change
in  ()
ol

- Note: change in K1 yields same percentage of change in

steady state value of  ol () (because SKT  1 in this case)
1

- Open loop control can attain desired performance only if

K1 is known and unchanged!
 • Sensitivity analysis for Closed Loop speed control

- Let disturbance torque Td  0 , then T(s)

K1 K a
 ( s) 
cl
R( s)
( 1s  1  K1 K a K t )
* T(s) change with change in K1:
 1s  1
Sensitivity with S T

( 1s  1  K1 K a K t )
K1
respect to motor gain

* Typical values: K1=2, Kt=1, Ka large (from 40100)

 K 1 K a K t from 80 200  S K 1  1
T

* Change in K1 yield vey small change in  ( s )  T ( s ) R( s )

cl

- Results consistent with previous sensitivity analysis on

OL and CL system
- Example: Let R(s)= step input and desired performance is
 ()  1
1 K1K a
- With R( s )  ,  ( s ) 
cl

s s( 1s  1  K1K a K t )
- Steady state value using FVT
K1 K a
 cl ()  lim s cl ( s) 
s 0 1  K t K1 K a
CL speed  cl (t )
K1K a
1 control 1  Kt K1K a
t system t
- One can set Ka large so that K 1 K a K t  1 to yield
close-to-desirable  cl ()  1
- E.g., when K1=2, Kt=1, Ka =50, K 1 K a K t  100
K1 K a 100
 ( ) 
cl
  0.99009  1
1  K t K1 K a 101
- Now, if K1 were to change by 10%, i.e., K1  2.2
(2.2)50 110
 ( ) 
cl
  0.99099  1
1  (2.2)50 111
 cl () quite insensitive to variations in K1!
- Hence, if K1 stays constant and known forever, OL speed
control is “better” in the sense that  ol ()  1while
 cl ()  1 only
- On the other hand, CL speed control can withstand change
in motor gain K1 and maintain  cl ()  1 always while
 ol () cannot
- Life is always changing  CL speed control is better overall
Summary: 1st advantage of CL system over OL system:
Lower sensitivity in system parameter variations (in forward
loop)
II.A.2. Improving Transient Response

Process

Can we adjust for “good” transient response?

• Open loop speed control (Td=0 still)
(s) K1 K a

R( s ) ( 1s  1)
1
- With R(s) a step function R( s ) 
s
- Open loop speed response:
K a K1
 (s) 
ol

s( 1 s  1)
- Open loop speed response: K a K1
 (s) 
ol

s( 1 s  1)
t

- Inverse Laplace Transform:  ol (t )  K K (1  e 1
)
a 1

- Time constant 1 , hence transient response, fixed by

process parameter and cannot be changed

- Typically,  1  10 sec for speed control motor

 Time to reach steady state long
(roughly 40-50 sec) for  ol (t )
• Closed Loop Control (Td=0 still)
(s) K1 K a

R( s ) ( 1s  1  K1 K a K t )

- With R(s) a step function

- Closed loop speed response:
K1 K a
( s) 
s( 1s  1  K1 K a K t ) Time constant for
closed loop system
- Inverse Laplace transform:
K1 K a  t
 cl (t )  (1  e cl
)
(1  K1 K a K t )
1
where  cl  is adjustable by K a and K t
1  K1K a K t
- With 1  K 1 K a K t  1
1
 cl  much smaller than 1
1  K1 K a K
 response to final value much faster
- Specifically, with
 1  10 sec and
K 1 K a K t  100

Summary: 2nd advantage of CL system over OL system:

Transient adjustable and faster with closed loop system
II.A.3. Ability to Reject Disturbance
• Open loop speed control
K1 K a R1
( s)  R( s )  Td ( s )
( 1s  1) ( 1s  1)
Governing transient and Undesirable Error
steady state performance E(s) if Td  0
- Question: how “large” is this undesirable error?
D
- Let step disturbance torque be a constant D: Td ( s ) 
s
R1 D
E ( s)  
ol

s( 1s  1)
- Steady state undesirable error due to disturbance
e ol ( )  lim sE ol ( s )   R1 D
s 0 Undesirable error added
to  () caused by
ol
(By FVT)
disturbance
• Closed loop speed control
K a K1 R1
( s)  R( s )  Td ( s )
( 1s  1  K a K1 K t ) ( 1s  1  K a K1 K t )

Governing transient and Undesirable Error

steady state performance
E(s) if Td  0
- For step disturbance Td ( s )  D
s
R1 D
E (s)  
cl

s( 1s  1  K a K1 K t ) K a K1 K t  1
Hence,
R1 D
e ( )  lim sE ( s )  
cl cl
 e ol ( )
(By FVT)
s 0 (1  K a K1 K t )

- Much smaller undesirable error added to  cl () due to Td

Summary: 3rd advantage of CL system over OL system:
Effects of disturbance much smaller in CL system
II.A.4. Disadvantages of Feedback
• Require feedback signal:
 more instrumentations (sensors, comparators, etc.)
 system more complex, cost, maintenance
• May have small steady state error: not reaching desired value
DC motor example
OL Final value=1

CL Final value=0.99
• Require more powerful actuators – for present example, motor
input Va for CL control is 100 times than that of OL

Va(t) for OL and CL speed control

Ka=0.5

Time (sec)

• System may be unstable due to improper controller assignment

and system uncertainties, etc.  require proper design
II.A.5. Design Example: Disk Drive Read System (CL Position Control)
Motor and arm G(s)
Armature-current controlled
Va(s)

• Output y(t) = head position = (t) from DC Motor

• Zero back emf: Kb=0
• Parameters:
 • Close Loop Performance (using Full Model)
Disturbance
D(s )

R(s) Y (s )

K a G1 ( s )G2 ( s ) G2 ( s )
Y ( s)  R(s)  D( s )
1  K a G1 ( s )G2 ( s ) 1  K a G1 ( s )G2 ( s )
5000 K a s  1000
 R(s)  D(s )
s 3  1020s 2  20000s  5000 K a s 3  1020s 2  20000s  5000 K a

Response due to R(s) governing Undesirable Error

transient and steady state performance Due to D(s)
• Performance due to R(s) only
- R(s)=1/s (Step), D(s)=0
(Textbook Fig. 4.36)

- Transient changeable by Ka
* Ka=10 – too slow?
* Ka=80 – too oscillatory?
• Error due to disturbance D(s) only
- R(s)=0 (No input), Disturbance D(s)=1/s (Step)
(Textbook Fig. 4.37)

- Undesirable error due to constant disturbance torque D(s)

for Ka=10 is 8 times than that for Ka=80
• Performance with BOTH input and disturbance
- Overall motor response with R(s)=1/s and D(s)=0.1/s

Design issue  “Best” value of Ka?