Control System I

Chapter 4
A First Analysis of Feedback

“Our greatest weakness lies in giving up. The

most certain way to succeed is always to try just
one more time.” — Thomas A. Edison

Chap 4, No. 1
Control System I
Chapter 4 Overview
The goal of the analysis in this chapter is to revisit each of requirement of
stability, tracking, regulation, and sensitivity in a linear dynamic setting,
and to develop equations that will expose constrained place on the
controller and identity elementary objectives for the controller.

The following topics are included:

4.1 Basic Equations of Control for stability, tracking, regulation, and
sensitivity.
4.2 Control of steady-state error to polynomial inputs: System type

4.3 The three-term controllers: PID control.

4.4 Feedforward control by plant model inversion.

(very briefly if go through)

Chap 4, No. 2
Control System I
The Basic Equations of Control
Consider the open-loop system:
W(s)
+ Yol  s 
R(s) Controller Plant
∑
Dol  s  + G s

Yol  GDol R  GW ,
The open-loop transfer function
The difference between R and Y: between output Yol and input R :
Eol  R  Yol
Yol  s 
 R  GDol R  GW  Tol  s    G  s  Dol  s  ,
R s
  1  GDol  R  GW ,

The open-loop transfer function

between output Yol and disturbance W :
Yol  s 
 G s
W s
Chap 4, No. 3
Control System I The Basic Equations of Control
Consider the closed-loop system:
W (s)
+
R(s)
E U s Plant Ycl  s 
∑ cl Controller ∑
+ Dcl  s  + G s
-
+
Y *
s ∑ V(s)
+
GDcl G GDcl V: Sensor noise
Ycl  R W V,
1  GDcl 1  GDcl 1  GDcl
Also Ecl  R  Y * , Y *  V  Ycl , Ycl  GW  GU
 
U  Dcl Ecl  Dcl R  Y *  Dcl  R  V  Ycl   Dcl R  DclV  DclGW  DclGU
  1  Dcl G  U  Dcl R  DclV  DclGW ,
Dcl GDcl Dcl
U  R W V.
1  GDcl 1  GDcl 1  GDcl

Chap 4, No. 4
Control System I The Basic Equations of Control
Consider the closed-loop system:
W (s)
+
R(s)
E U s Plant Ycl  s 
∑ cl Controller ∑
+ Dcl  s  + G s
-
+
Y *
s ∑ V
+ V: Sensor noise
The error:
 GDcl G GDcl 
Ecl  R  Ycl  R   R W V
 1  GDcl 1  GDcl 1  GDcl 
1 G GDcl
 R W V,
1  GDcl 1  GDcl 1  GDcl
1 GDcl
Define  ,   Therefore, the closed-loop
1  GDcl 1  GDcl transfer function
Then Ycl   R  G W   V Ycl GDcl
Tcl   
U  Dcl R   W  Dcl V R 1  GDcl
Ecl  R  G W   V
Chap 4, No. 5
Control System I
Stability - 1
For a open-loop system, the transfer function is Tol  G  s  Dol  s  .
bs c s
Let G  s   , and Dol  s  
as d s
The stability requires that there is no roots of a(s) and d(s) in the RHP.
In open-loop system, the root in RHP, resulting in unstable, cannot be
fixed by pole-zero cancellation. Similarly, a zero may cause a poor
response, but cannot be improved by cancellation of a pole.
For a closed-loop system, the system poles are the roots of
bs c s
1  G  s  Dcl  s   0. Let G  s   , and Dcl  s  
as d s
bs c s
1  0 or a  s  d  s   b  s  c  s   0
as d s
The stability requires no roots of a(s)d(s) + b(s)c(s) = 0 in the RHP.
Compared to a open-loop system, the feedback case provides more
freedom to the controller design. But still avoid pole-zero cancellation.

Chap 4, No. 6
Control System I
Stability - 2
For example in Chapter 2, the transfer function for an inverted pendulum,
G s
bs 1 1 Yol  s 
G s   2  , R(s) 1
a  s  s  1  s  1 s  1
 s  1 s  1
The pole s = 1 in the RHP to cause the system unstable.
c s K s   
If a closed-loop controller is designed, Dcl  s   
d s s
Dcl  s  G s
R(s)
  Ycl  s 
E
∑ cl K  s    U s 1
+
- s  s  1 s  1

Then the characteristics equation:  s  1 s  1 s     K  s     0.

A simple solution is take   1 and the common (stable) factor cancels.
Note that the cancellation is fine in this case, because (s + 1) is a stable
pole. The resulting 2nd-order equation can be easily solved to place the
remaining two poles at any points desired. Chap 4, No. 7
Control System I
Stability - 3
A closed-loop controller is designed, then the transfer function:
Ycl GDcl
 ,
R 1  GDcl
The the characteristics equation: 1  GDcl  0,
K s   
1
1
 0,  s  1 s  1 s     K  s     0.
 s  1 s  1 s  
For   1 the resulting 2nd-order equation:
s 2     1 s  K  0. Exercise
Routh array  s  1 s  1 s     K  s    
s2: 1 K s 3   s 2   K  1 s   K      0.
s 1:  1 0 s3: 1 K -1
s0: s2:  K - 
s 1:
s0:
Chap 4, No. 8
Control System I
Tracking
The tracking problem is to cause the output follow the reference input
as closely as possible.
In principle, the controller can be designed to reach the desired
performance. However, there are three points to be concerned:
1. The controller transfer function must be proper; that is, the numbers
of poles  the numbers of zeros.
2. The response of a physical system has its limitation. DO NOT have
an unrealistically fast design for the system, even it is stable.
3. In principle, any pole in the LHP can be cancelled for better system
response. However, to cancel a pole barely inside the LHP may
result in a response of unacceptable transients.

In this figure the output y

is tracking the reference 1
y, r r ess 
input r, though it appears Kv
a steady-state error ess . y

Chap 4, No. 9
Control System I Regulation
The regulation problem is to keep the error small when the reference is
at the most a constant setpoint and disturbances W are presented.
Yol  GDol R  GW ,
In a open-loop system, the controller has no influence at all on the
system response to either of the disturbance, W, or V, so this open-loop
structure is useless for regulation.
For a closed-loop system, a conflict between W and V happens when
one search a good controller. For example, from the error equation
1 G GDcl
Ecl  R W V,
1  GDcl 1  GDcl 1  GDcl
Select Dcl large to lower the effect of W, but will keep the term of V at
unity and the sensor noise is not reduced.
The resolution of this dilemma is observed y
that each of these terms is a function of Set
frequency so one of them can be large for point
some frequencies and small for others.
t
Dcl : large at low freq. to reduce the effect of W
Dcl : small at high freq. to reduce the effect of V
Chap 4, No. 10
Control System I
Sensitivity - 1
The block-diagram of open-loop system,
W(s) Yol  GDol R  GW ,
+
R(s) Controller Plant Yol  s 
∑
Dol  s  + G s

A plant with gain G at a frequency, it changes to G   G in operation.

Thus from
Tol  G  s  Dol  s  .
The change of  Tol will cause the system: (Notice that no change of Dol )
Tol   Tol  Dol G   G   DolG  Dol  G.  Tol  Dol  G.
T
The sensitivity of transfer function SG is defined:

SGT   Tol Tol  G  Tol

 ,
  G G  Tol  G
Also  Tol Dol  G  G
  SGT  1 for a open-loop system
Tol DolG G
In a open-loop system, the % change of G will cause
the same % change of Tol .
Chap 4, No. 11
Control System I
Sensitivity - 2
For a closed-loop system,
Tcl 
GDcl
, Tcl   Tcl 
G   G  Dcl
,
1  GDcl 1  G   G  Dcl
The first-order variation is proportional to the derivative,
dTcl
 Tcl   G,
dG
Tcl G dTcl
Tcl
SG Sensitivity of Tcl with respect to G, or SG
Tcl dG

SGTcl
G  1  GDcl  Dcl  Dcl GDcl   1 .
GDcl  1  GDcl   1  GDcl 
2
1  GDcl
In feedback control, the error in overall transfer function gain is less
sensitive to variations in the plant gain by a factor of
1
 , : sensitivity function
1  GDcl
Compared to errors for in open-loop control gain, if  1  GDcl  
100, a 10% change in plant gain G will cause only 0.1% change in
the steady-state gain.
Chap 4, No. 12
Control System I
Sensitivity - 3
Redefine the sensitivity function for a feedback system,
1

1  GDcl
The complementary sensitivity function is defined,
GDcl
  1
1  GDcl
Clearly   1
The results in this section assume that the steady-state error occurs
due to constant inputs, either reference or disturbance.
Example: a feedback amplifier and loudspeaker system for Hi-Fi sound.
The error of the feedback system
1  1 
Ecl  j0   R  j0  E 
,  cl 1  GD R, if W  V  0 . 
1  G  j0  Dcl  j0   cl 
To reduce errors to 1% of the input, it needs 1  GDcl  100, or
G  j0  Dcl  j0   100,
where 60  0 2  15 , 000 Hz (frequency range of human hearing)
Chap 4, No. 13
Control System I
4.2 Control of Steady-State Error to Polynomial Inputs:
System Type

Overview of this section:

In a number of important cases the reference input will not be
constant but can be approximated as a linear function of time for a
time span long enough for the system to reach steady-state.
Therefore, it needs to know the steady-state error.
Stable systems can be classified into system types according to the
polynomial degree for which the error is constant.
System types can be defined with regard to reference inputs or
disturbance input.
Determining the system type involves calculating a certain transfer
function and then applying the Final Value Theorem.
The definition of system type help us to identify quickly the ability of a
system to track polynomial. System type is a robust property with
respect to parameter change in the unity feedback structure.

Chap 4, No. 14
Control System I
4.2 Control of Steady-State Error to Polynomial Inputs:
System Type
Consider a closed-loop control structure as shown
W (s)
R(s) +
Ecl Controller Plant Y (s)
∑ ∑
+ Dcl  s  + G s
-
+
∑ V
1 tk 1
Ecl  R, Let r  t   1 t  ,  R  s   k  1
1  GDcl k! s
1 1 1
Then lim e  t   ess  lim sEcl  s   lim s R  s   lim s ,
t  s 0 s  0 1  GD s  0 1  GD s k  1
cl cl
For a unit-step input, R(s) = 1/s, and r(t) is a polynomial of degree 0.
1 1 e e 1
ess  lim s ,  ss  ss  ,
s  0 1  GD s
cl rss 1 1  GDcl  0 
The system is defined to be a Type 0 and GDcl  0  K p as the “position
error constant.” A polynomial of degree 0 is the highest degree a
system of Type 0 can track at all.
Chap 4, No. 15
Control System I 4.2 Control of Steady-State Error to Polynomial Inputs:
System Type
GDclo  s  All the terms except the pole(s) at the origin, and it is finite
if s = 0. Then
GDclo  s  n = 0 if GDcl has no integrator;
GDcl  s  
Define GDclo  0  K n sn n = 1 if GDcl has one integrator
1 1 1 sn 1
ess  lim s R  s   lim s  lim ,
s  0 1  GD s 0 GDclo  s  s k 1 s  0 s  Kn s
n k
cl
1
if n  k , ess  0, sn
if n  k , ess   ,
if n  k  0, ess  1  1  K 0  ; n  k  0, ess  1 K n
Error Type:
If n = k = 0, it is a unit-step input, and r(t) is a polynomial of degree 0.
The constant K 0 is called the “position constant” written as K p and it
is classified as Type 0.
If n = k = 1, the input is a 1st degree polynomial, a ramp or a velocity
input, and the constant K1 is called the “velocity constant” written as
Kv and it is classified as Type 1.
A similar way can be used to define Type 2 and higher.
Chap 4, No. 16
Control System I 4.2 Control of Steady-State Error to Polynomial Inputs:
System Type
The results can be
summarized by the
following equations:
K p  limGDcl  s  , n  0,
s 0

Kv  lim sGDcl  s  , n  1,
s 0

K a  lim s 2GDcl  s  , n  2.
s 0

In steady-state operation, the

output velocity is exactly the
same as the input velocity,
but there is a proportional
error, 1 Kv

Chap 4, No. 17
Control System I
System Type using Proportional and Integral Control
Example 4.1: A speed control with P controller.
A
G s  , Dcl  s   kP ,
s 1
Solution: kP A kP A
G  s  Dcl  s   , K p  limGDcl  s   lim  kP A
s 1 s 0 s 0  s  1

The system is Type 0 since n = 0, no pole at s = 0. ess 

Example 4.2: A speed control with PI controller.

A
G s  , Dcl  s   kP  k I s , ess 
s 1
Solution:
A kP s  k I
G  s  Dcl  s   ,
s 1 s
The system is Type 1 since n = 1, one pole at s = 0.
A kP s  k I
Kv  lim sGDcl  s   lim s  Ak I .
s 0 s 0 s 1 s
Chap 4, No. 18
Control System I
System Type using Error Transfer Function
Consider a closed-loop control structure as shown
W (s)
R(s) +
E
∑ cl Controller
G s Plant Y (s)
∑
+ Dcl  s  + G s
-
+
Y s
*
∑ V
Y s GDcl
  ,  Ecl  s   R  s   Y  s   R  s    R  s  ,
R s 1  GDcl

 Ecl  s   R  s  1    s   ,
ess  lim e  t   lim sE  s   lim s 1    s   R  s  ,
t  s 0 s 0
1
If the reference input is a polynomial of degree k, then R  s   k  1
s
1  s  1  s 
 ess  lim s k 1
 lim k
,
s 0 s s  0 s
The system is referred to Type k if ess is a nonzero constant.
Notice a system of Type 1 or higher has a closed-loop DC gain of
1.0, meaning   0   1 in these cases. Chap 4, No. 19
 Control System I Closed-loop System with Sensor Dynamics
Example 4.3: An electric motor position (#29 of Ch. 2) with tachometer feedback
W (s)
R(s) + H(s) is used to
∑ E Controller
G s Plant Y (s) improve dynamic
∑
+ Dc  s  + G s response, but
-
+ will increase
Sensor V(s) steady-state
∑
H s error.
For
1
G s  , Dc  s   kP , H  s   1  kt s, E  s   1    s   R  s  ,
s  s  1
Solution: kP
Dc  s  G  s  s  s  1 kP
 s    ,
 t  s  sP 1 s  s  1   1  kt s  kP
1  H  s  Dc  s  G  s  1  1  k s k
 
1  s  1  s  s  1  kP kt  
 ess  lim  lim k  ,
s 0 s k s  0 s  s  s  1 
 1  k t  P 
s k 
  0, if k  0 , kP
 The system is Type 1 with Kv 
ess   1  kt kP 1  kt k P
  , if k  1 Chap 4, No. 20
 kP
Control System I 4.2.2 System Type for Regulation and Disturbance Rejection
The transfer function from the disturbance input W(s) to the error E(s):
In general, E  s  Y s
  Tw  s  ,
W s W s
because, if the reference is zero, the output is the error.
Example 4.4: System type for a DC motor position control
(a) Dc  s   kP
B A
 1
E s A s  s  1 B
  ,
W s 1
Ak P s  s  1  AkP
s  s  1
Type 0 since if W(s) = 1/s, ess  B AkP .
B A
(b) Dc  s   kP  k I  1
s E s A s  s  1  Bs
  2 ,
W s A  kP  k I s  s  s  1  A  kP s  k I 
1
s  s  1
Type 1 since if W  s   1 s 2 , ess  B Ak I .
Chap 4, No. 21
Control System I
Summery of Steady-State Error and System Type
 Position error, velocity error, and acceleration error mean steady-state
deviation in the output position. Position here means the physical
output such as position, temperature, pressure, voltage, water level,
etc.

 The error constants K p , Kv , and K a describe the ability of a system to

reduce or eliminate steady-state error.

 To improve the steady-state performance we can increase the type of

the system by adding an integrator (like PI control in Example 4.4 ) or
integrators to the feedforward path. This, however, introduces an
additional stability problem. (The design of a satisfactory system with
more than two integrators in series in the feedforward path is generally
difficult.)

We cannot tell if a Type 1 or higher if one only review the steady-state

error.

Chap 4, No. 22
Electric Circuits
The Classical Three-term PID Controllers
The classic three-term PID controllers are often been used:
 Proportional feedback (P controller)
-- Reduce error due to disturbance but it still allow a nonzero steady-
state error.
-- Increase the speed of response but has a much larger transient
overshoot.
 If integral feedback (I controller) is added, then 𝒆𝒔𝒔 = 𝟎 , but deteriorate
the dynamic response (worse in transient response)
 If derivative feedback (D controller) is added, then it damps the dynamic
response (worse in transient response)
 PI and PD controllers are also employed in industries for the
combination advantages of P, I and, D.
 PID controller is widely used in the process industries and other
applications. The transfer function of PID controller:
Dc  s   kP  k I s  kD s,

Chap 4, No. 23
Electric Circuits
Proportional Control (P-controller)
In P controller, the feedback control signal is linearly proportional to
the system error. The control law
W (s)
u  t   kP e  t  , Dcl  s   k P +
R(s)
kP : the proportional gain ∑ E Controller
U Plant Y (s)
∑
+ Dcl  s  + G s
The control transfer function is: -
U s
 Dcl  s   kP
E s
Proportional feedback (P control) my  cy  ky  f  t  ,
A mechanical
system
just like a “knob” to adjust the gain. The characteristic eqn.
Consider a 2nd-order system ms 2  cs  k  0 k c
A
G s  2 ,
s  a1s  a2 m
The characteristic equation is: y
f
AkP
1  kP G  s   0,  1  2 0 d 2v dv 1 1  
1 c
s  a1s  a2 L 2 R  v  VS , 2 mk
dt dt C C
 s 2  a1s  a2  kP A  0
k
The characteristic eqn: n 
kP : affect the natural frequency, m
but cannot affect the damping.  Ls 2  Rs  1 C   0
Chap 4, No. 24
Electric Circuits
Proportional controller (P-controller)
This system (last page) is Type 0, one can make kP large enough to get
the small steady-state error. However, the damping may be too small
and cause the worse transient response (large overshoot).
Let a1  1.4 , a2  1, and A  1.
Given the unit-step input r(t) = 1, with kP  1.5 and 6.
Y s kP
 2 ,
R  s  s  1 .4 s  1  k P
kP :  ,  M p : , t r : , e ss :
Y and U due to the disturbance W:
Y s G s U s kPG  s 
 , 
W  s  1  kPG  s  W  s  1  k PG  s 
Like a reference input, a unit-step disturbance
w will yield a steady-state tracking error.

For the system with higher order (more than 2nd-order), a higher gain
will increase the speed of response but typically at cost of a lager
transient overshoot and less over all damping.
Chap 4, No. 25
Control System I Integral Control (I control)
I Controller : to minimize the steady-state tracking error and the steady-
state output response to disturbance.
The control law:
u  t   k I tt0 e  d , k I : the integral gain
In I Controller, at each instant of time the control signal is the summation
of past values of error; the control action is based on the “history” of the
system error.
The control signal at any instant of time is
proportional to the area under the system
error curve for time t1
The control transfer function is:
U s k
 Dcl  s   I ,
E s s
I controller has infinite gain at DC (for s = 0),
hence it will have a superior performance in
steady-state error.
It can provide a finite value of control with zero system error because u(t)
is a function of all past values of e(t) rather than just the current value.
Chap 4, No. 26
Control System I
Integral Control (I control)
Use an example to discuss the function of a pure I-Controller
a2 k
G s  2 , Dcl  s   I ,
s  a1s  a2 s
kI
E s 1 s U s s kI
  ,  
R  s  1  kI G s s  k IG  s  R  s  1  kI G s
    s  k IG  s 
s s
Y s k IG  s 
  s 
R s s  k IG  s 
For a unit-step input R(s) = 1/s, and using the Final Value Theorem (FVT):
kIG  s  1 k IG  0 
y     lim sY  s  R  s   lim s   1.
s 0 s 0 s  k IG  s  s 0  k IG  0 
Similarly
0 kI
e    0, u      G  0   1.
1

0  kIG  0  0  k IG  0 
The steady-state error will be zero no matter what the value of k I .
I Controller : The gain k I can be selected to provide an acceptable dynamic
response; however it will cause instability if raised suddenly high. It may reduce
the stability and may not have acceptable transient behavior.
Chap 4, No. 27
Control System I
Integral Control (I control)
Y and U due to the disturbance W:
Y s sG  s  U s k IG  s 
 , 
W  s  s  k IG  s  W  s  s  k IG  s 
For a unit-step input W(s) = 1/s, and using the Final Value Theorem (FVT):
sG  s  1 0  G 0
y     lim sY  s  W  s   lim s   0.
s 0 s 0 s  k IG  s  s 0  k IG  0 
Similarly kIG  0 
u      1.
0  kIG  0 
No steady-state error for y and the control signal u will cancel out the
disturbance w; that is,
Y  s   G  s  U  s   W  s   , u     w     1  1  0,  y     0.

W (s)
R(s) +
∑ E Controller G  s  ∑ Plant Y (s)
+ Dcl  s  + G s
-

Chap 4, No. 28
Control System I
Integral Control (I control)
It concludes that integral feedback (I-controller) results in zero steady-
state output error in both tracking and disturbance rejection.
The results are independent of the plant parameters, of which the
change of plant parameters can be tolerated.
Regardless the value of k I , the asymptotic tracking and disturbance
rejection properties are preserved, if the closed-loop system is stable.
These properties of integral control are referred to as robust.
The additional I-Controller cause the closed-loop system to become a
Type 1 and will have a constant tracking error to a ramp reference input.

The responses of output y and control signal u due to disturbance w.

k I  0 .5 k I  0 .5

Chap 4, No. 29
Control System I Example for the effect of I Controller
W (s) B
A k
G s  , Dcl  s   I R(s)
E + Y (s)
s 1 s
+
∑ Dcl  s  ∑ G s
+
-

Ak I
AB
Y s s  s  1
, Y s 
Ak I
  2 s 1  2
ABs
R  s  1 Ak I  s  s  Ak I W  s  Ak I  s  s  Ak I
,
s  s  1 1
s  s  1
E s s  s  1 The characteristic equation:
 2 ,
R  s   s  s  Ak I  s 2  s  k I A  0,  
1
,  kI ,  
2  Ak I
(1) if R = 1/s , then yss = 1 = r.
(2) yss = 0, due to W (step function)
Integral action tends to produce a control signal after the error has
vanished, while suggests that the I controller should be aware that the error
is approaching zero.
Chap 4, No. 30
Control System I 4.3.3 Derivative Feedback (D-Control)
Derivative controller (D-control), also called rate feedback, is used to
improve closed-loop system stability as well as speeding up the transient
response and reducing overshoot (damp out oscillation).
The control law:
u  t   kDe  t  , kD : the derivative gain
The control transfer function is:
E s
 Dcl  s   kD s,
U s
Unless the system has a natural proportional term such as a spring or
equivalent feature, derivative control will never be used alone since it will
not drive the error to zero if e(t) is constant. Usually it is used with PD or
PID controller.
In D control, the correction depends on the rate of change of the error.
de e  t 0  Δt   e  t 0 
 lim , Δt  t  t 0 ,
dt t  t0 Δt  0 Δt
Its value depends on the future e  t 0  Δt  , and is called anticipatory
response.
Chap 4, No. 31
Control System I 4.3.3 Derivative Feedback (D-Control)
An important effect of D-control is to give a sharp response to suddenly
changing signals
Therefore, the derivative term is sometimes introduced into the feedback
path, as shown, in a PID controller, with the transfer function:
k
Dc  s   kP  I  kD s
s
Notice that the zeros from the reference
to the output are different in two cases.
With derivative in the feedback path, the
reference is not differentiated, which is
how the undesirable response to sudden
changes is avoided. (A)
G  s   kP  k I s 
(A) Y  s  
R  s  1  G  s   kP  k I s  kD s 
(B) Y  s  G  s   kP  k I s  kD s 

R  s  1  G  s   kP  k I s  kD s  (B)
Disadvantage of D Controller : it tends to amplify noise, discussed in Chapter 6.
Chap 4, No. 32