?

PART IV
Major Types of Linear Controllers
University of Waterloo - SD 352 99 of 186
?
Linear controllers are the module components of a linear system that would
transform error signals into actuating signals for the plant.
R(s)
sensors
plant
U(s)
C(s)
E(s)
controllers
University of Waterloo - SD 352 100 of 186
?
Proportional (P) Controllers
E(s) U(s)
C(s)
R(s)
p K
H(s)
G(s)
U(s) = K
p
E(s)
u(t) = K
p
e(t)
C(s)
R(s)
=
K
p
G(s)
1 + K
p
GH(s)

E(s)
R(s)
=
1
1 + K
p
GH(s)
e

= lim
s0
sE(s) = lim
s0
s
1 + K
p
GH(s)
R(s)
Therefore,
Larger K
p
smaller error, faster response, larger overshoot (OS).
Higher type smaller error.
University of Waterloo - SD 352 101 of 186
?
Proportional-Derivative (PD) Controllers
K
U(s) E(s) R(s) C(s)
p G(s)
d
G(s)
K
s
H(s)
^
U(s) = (K
p
+ sK
d
)E(s)
u(t) = K
p
e(t) + K
d
d e(t)
d t

G(s) = (K
p
+ sK
d
)G(s)
E(s)
R(s)
=
1
1 +

GH(s)
=
1
1 + (K
p
+ sK
d
)GH(s)
e

= lim
s0
sE(s)
=
lim
s0
sR(s)
1 + K
p
lim
s0
GH(s)
University of Waterloo - SD 352 102 of 186
?
TF =
C
R
=

G
1 +

GH
=

G
1 + (K
p
+ K
d
s)GH
=

G
1 + (K
p
+ K
d
s)

2
n
s(s+2
n
)
=
A
s
2
+ s(2
n
+ K
d

2
n
) + K
p

2
n
, for some constant A
The denominator shows that a PD controller increases the damping of
the system.
0 5 10 15 20 25 30 35 40
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Time
R
e
s
p
o
n
s
e
Without PD
With PD
Advantage: More damping.
Disadvantage: Slower response and larger T
p
.
University of Waterloo - SD 352 103 of 186
?
Proportional-Integral (PI) Controllers
s
U(s) E(s) R(s)
K
C(s)
p G(s)
^
G(s)
i
K
H(s)
U(s) = (K
p
+
K
i
s
)E(s)
u(t) = K
p
e(t) + K
i
_
e(t)dt

G(s) = (K
p
+
K
i
s
)G(s)
E(s)
R(s)
=
1
1 + (K
p
+
K
i
s
)GH
e

= lim
s0
sE(s)
= lim
s0
sR(s)
1 + (K
p
+
K
i
s
)GH
0 1 2 3 4 5 6 7
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Time
R
e
s
p
o
n
s
e
With P Controller
With PI Controller
Increase in the system type
Better (smaller) steady-state error.
Higher overshoot.
University of Waterloo - SD 352 104 of 186
?
PID Controllers
K
d
U(s) E(s) R(s) C(s)
s
G(s) Kp
G(s)
^
K
i
s
H(s)
U(s) = (K
p
+
K
i
s
+ K
d
s)E(s)
u(t) = K
p
e(t) + K
i
_
e(t)dt +
d e(t)
d t

G(s) =
_
K
d
s
2
+ K
p
s + K
i
s
_
G(s)
Advantages:
Smaller error (due to the proportional gain).
Smaller overshoot (due to the derivative gain).
Smaller steady-state error due to the integral gain.
A compromise eect between the two gains.
Disadvantages:
No formal way for determining optimal PID values.
Use optimization techniques.
University of Waterloo - SD 352 105 of 186
?
Proportional and Tacho Feedback Controller
p
E(s) R(s)
d
K
C(s)
s
G(s)
H(s)
K
U(s)
In this case, the sensor H(s) acts as a lter to lter the usually noisy
signal c(t).
The block K
d
s measures the derivative (velocity) of the signal being
fed to it by the sensor.
A tacho feedback is used to compute the velocity of a signal if it is
not available.
University of Waterloo - SD 352 106 of 186
?
Velocity Feedback Controller
s C(s) C(s) E(s) R(s)
G(s) Kp

K
f
K s
1
s

H(s)
For DC servo systems,
(s)
V (s)
=
K
1
s(s+K
2
)
and

(s)
V (s)
=
K
1
s+K
2
, for certain gains
K
1
and K
2
.
Remark: In practice, it is better to lter the output signal before applying
it to the feedback loop.
University of Waterloo - SD 352 107 of 186
?
A Typical Representation of a Tacho Feed-
back
C(s) R(s)
K
t
s
G(s)

2
n
s
2
+2ns+
2
n
C(s)
R(s)
=
G(s)
1 + G(s)K
t
s
=

2
n
s
2
+2
n
s+
2
n
1 +

2
n
s
2
+2
n
s+
2
n
(K
t
s)
=

2
n
s
2
+ (2
n
+
2
n
K
t
. .
2
t

n
)s +
2
n
=

2
n
s
2
+ 2
t

n
s +
2
n
University of Waterloo - SD 352 108 of 186
?
2
t

n
= 2
n
+
2
n
K
t

t
= +
1
2

n
K
t
Hence, a tacho feedback loop increases the damping of the system
without aecting the speed of the response.
0 5 10 15 20 25
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Time
R
e
s
p
o
n
s
e
Without Tacho Feedback
With Tacho Feedback
j

t n

n

n

|
n
| < |
t

n
|
University of Waterloo - SD 352 109 of 186
?
A Comparison of Dierent Controllers
c
C(s)
G(s)
R(s)
G (s)
H(s)
G(s) =
1
(s+1)(s+2)
H(s) = 0.8
R(s) =
1
s
P Controller
G
c
(s) = 20 Char. Eq.: s
2
+ 3s + 18 = 0.
This leads to,

n
=

18 =
3
2
n
= 0.35
K
p
= 8 e
ss
= 0.111
University of Waterloo - SD 352 110 of 186
?
PD Controller
G
c
(s) = (20 + 4s) Char. Eq.: s
2
+ 6.2s + 18 = 0.
This leads to,

n
=

18 =
6.2
2
n
= 0.73 e
ss
= 0.111
Higher damping PD controller improves overshoot.
University of Waterloo - SD 352 111 of 186
?
PI Controller
G
c
(s) = (20 +
2
s
)
Char. Eq.: s
3
+3s
2
+18s+16 = 0 (s+0.09)(s
2
+2.91s+17.3) = 0.
This leads to,

n
=

17.37 = 4.21 =
2.91
2
n
= 0.345 e
ss
= 0
Advantages:
Zero steady-state error.
Disadvantages:
Higher overshoot (due to the eect of the 3rd pole, which is close
to the j-axis).
University of Waterloo - SD 352 112 of 186
?
PID Controller
G
c
(s) = (20 + 4s +
2
s
)
Char. Eq.: (s + 0.0917)(s
2
+ 6.11s + 17.44) = 0.
This leads to,

n
=

17.44 =
6.11
2
n
= 0.73 e
ss
= 0
PID controller leads to a compromise between PD and PI controllers.
University of Waterloo - SD 352 113 of 186
?
PART V
Stability of Linear Feedback Systems
University of Waterloo - SD 352 114 of 186
?
The Concept of Stability
- -
Q(s) U(s) Y (s)
y(t) = L
1
[Q(s)U(s)] = y
ss
(t) + y
t
(t) , where
y
ss
(t) is the steady-state response. It is due to the input U(s) and
it has the same wave form as U(s).
y
t
(t) is the transient response. Its wave form is determined by the
poles of Q(s).
If s
1
, s
2
, . . . , s
n
are the poles of Q(s), then y(t) has the following form
y(t) = A
1
e
s
1
t
+ A
2
e
s
2
t
+ + A
n
e
s
n
t
If Q(s) is the transfer function of a CLS and it is of the form Q(s) =
G(s)
1+GH(s)
, then the poles of the CLS, s
1
, s
2
, . . . , s
n
, are determined by
the CLSs characteristic equation: 1 + GH(s) = 0.
University of Waterloo - SD 352 115 of 186
?
If all s
i
s, i = 1, . . . , n, are in the LHS of the s-plane, i.e., Re(s
i
) =

i
< 0

s
s
i
j
i
lim
t
A
i
e
s
i
t
= lim
t
A
i
e

i
t
e
j
i
t
= lim
t
A
i
e
j
i
t
since lim
t
e

i
t
= 0.
In this case, the system is called asymptotically stable.
University of Waterloo - SD 352 116 of 186
?
If one or more poles are on the j-axis while the rest of the poles are
in the LHS of the s-plane, i.e., Re(s
i
) =
i
= 0

j
j
i

i
A
i
e
s
i
t
= A
i
e

i
t
e
j
i
t
= A
i
e
j
i
t
= A
i
(cos
i
t + j sin
i
t)
The term (cos
i
t + j sin
i
t) never dies out.
In this case, the system is called marginally or critically stable.
University of Waterloo - SD 352 117 of 186
?
If one or more poles are in the RHS of the s-plane, i.e., Re(s
i
) =
i
> 0
lim
t
A
i
e
s
i
t
= lim
t
A
i
e

i
t
e
j
i
t
=
since lim
t
e

i
t
= .
In this case, the system is called unstable.
University of Waterloo - SD 352 118 of 186
?
Methods to Determine Stability
Finding poles directly
Routh-Hurwitz criterion
Nyquist criterion
Bode-plot method
Lyapunovs criterion
University of Waterloo - SD 352 119 of 186
?
Routh-Hurwitz Criterion
Introduction
If the characteristic equation 1 +GH(s) = 0 is written in the form of
a
n
s
n
+ a
n1
s
n1
+ + a
1
s + a
0
= 0 ,
then for the CLS to have no poles in the RHS of the s-plane, it is
necessary but NOT sucient that
all the coecients a
n
, a
n1
, . . . , a
0
, must have the same sign, and
None of these coecients is zero, i.e., i = 0, . . . , n, a
i
= 0.
A necessary and sucient condition for the CLS not to have poles
in the RHS of the s-plane is given by the Routh-Hurwitz criterion.
University of Waterloo - SD 352 120 of 186
?
Routh-Hurwitz Table
s
n
a
n
a
n2
a
n4
. . . . . .
s
n1
a
n1
a
n3
a
n5
. . . . . .
s
n2
b
n2
b
n4
b
n6
. . . . . .
s
n3
c
n3
c
n5
c
n7
. . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
s
0
h
0
where
b
n2
=
1
a
n1

a
n
a
n2
a
n1
a
n3

=
a
n1
a
n2
a
n
a
n3
a
n1
b
n4
=
1
a
n1

a
n
a
n4
a
n1
a
n5

=
a
n1
a
n4
a
n
a
n5
a
n1
b
n6
=
1
a
n1

a
n
a
n6
a
n1
a
n7

=
a
n1
a
n6
a
n
a
n7
a
n1
etc...
University of Waterloo - SD 352 121 of 186
?
c
n3
=
1
b
n2

a
n1
a
n3
b
n2
b
n4

=
b
n2
a
n3
a
n1
b
n4
b
n2
c
n5
=
1
b
n2

a
n1
a
n5
b
n2
b
n6

=
b
n2
a
n5
a
n1
b
n6
b
n2
etc...
Routh-Hurwitz Criterion
The number of roots of the characteristic equation in the RHS of the
s-plane is equal to the number of sign changes in the elements of the
rst column of the Routh-Hurwitz table.
The system is stable if and only if there are no sign changes in the
rst column of the Routh-Hurwitz table.
The Routh-Hurwitz criterion provides an idea about the absolute sta-
bility of the system (stable or unstable). It does not say anything
about how stable (or unstable) the system is.
University of Waterloo - SD 352 122 of 186
?
Example
Ch. Eq.: s
3
4s
2
5s + 6 = 0
Since not all coecients have the same sign and according to the
necessary condition, the system is unstable (one or more roots are in
the RHS of the s-plane).
Verifying using the Routh-Hurwitz criterion
s
3
1 5 0
s
2
4 6 0
s
1 206
4
= 3.5 0
s
0 3.560
3.5
= 6
First column: 2 sign changes
In fact, this can be clearly seen by factorizing the Ch. Eq. as
s
3
4s
2
5s + 6 = (s + 1)(s 2)(s 3) = 0.
University of Waterloo - SD 352 123 of 186
?
Example (Zero in the rst element of a row)
Ch. Eq.: s
3
+ 3s + 2 = 0
s
3
1 3 0
s
2
0 2 0
s
1 02
0
= what to do?
s
0
Replace 0 by a small positive number .
s
3
1 3 0
s
2
0 2 0
s
1 32

=
2

0
s
0
2
There are 2 sign changes in the rst column which means that there
are 2 roots of the Ch. Eq. in the RHS of the s-plane and hence the
system is unstable.
University of Waterloo - SD 352 124 of 186
?
Example (Row of zeros below the auxiliary equation)
Ch. Eq.: s
5
+ s
4
+ 4s
3
+ 24s
2
+ 3s + 63 = 0
s
5
1 4 3 0
s
4
1 24 63 0
s
3
20 60 0
s
2
21 63 0 Aux. Eq. Coes
s
1
0 0 0 Row of zeros
s
0
University of Waterloo - SD 352 125 of 186
?
Auxiliary equation: A(s) = 21s
2
+ 63 = 0.
Dierentiating A(s):
d A(s)
d s
= 42s
Now, use the coecients of
d A(s)
d s
in the row of zeros.
. . . . . . . . . . . . . . .
s
2
21 63 0
s
1
42 0 0 Coes of
d A(s)
d s
s
0
63
There are 2 sign changes in the rst column which means that there
are 2 roots of the Ch. Eq. in the RHS of the s-plane and hence the
system is unstable.
University of Waterloo - SD 352 126 of 186
?
Solving auxiliary equations:
Example
A(s) = 21s
2
+ 63 = 0 =s
1,2
= j
_
63
21
= j

3
This means that the Ch. Eq. has two roots on the j-axis.
Thus, should there be no roots of the Ch. Eq. in the RHS of the
s-plane, the system is marginally stable.
University of Waterloo - SD 352 127 of 186
?
Example
Ch. Eq.: s
6
+ 3s
5
3s
4
9s
3
4s
2
12s = 0
Routh-Hurwitz table:
s
6
1 3 4 0
s
5
3 9 12 0
s
4 9+9
3
= 0
12+12
3
= 0 0 Row of zeros!
Auxiliary equation: A(s) = 3s
5
9s
3
12s = 0
s
1
= 0
3s
4
9s
2
12 = 0
Let z = s
2
=3z
2
9z 12 = 0
=z
1,2
= +1.5

2.25 + 4 =
_
+4
1
=s
2,3
=

z
1
= 2 and s
4,5
=

z
2
= j
This can be easily veried as the Ch. Eq. can be factorized as
s(s
2
+ 1)(s
2
4)(s + 3) = 0.
University of Waterloo - SD 352 128 of 186
?
Dierentiating A(s):
d A(s)
d s
= 15s
4
27s
2
12
Routh-Hurwitz table becomes
s
6
1 3 4 0
s
5
3 9 12 0
s
4
15 27 12 0
s
3
3.6 9.6 0
s
2
67 12 0
s
1
8.96 0
s
0
12
One sign change = one root in the RHS of the s-plane.

j
3 2 +2
j
+j
splane
Note: It is important to remember that the j-axis is considered
neither in the RHS nor in the LHS of the s-plane.
University of Waterloo - SD 352 129 of 186
?
Determining the Gain Range for Stability
Example
C(s) R(s)
0.1
G(s)
K
s(s
2
+2s+4)
Ch. Eq: 1 + GH(s) = s
3
+ 2s
2
+ 4s + 0.1K = 0
Routh-Hurwitz table:
s
3
1 4 0
s
2
2 0.1K 0
s
1
4 0.05K 0
s
0
0.1K
For the characteristic roots (poles of the CLS) not to be in the RHS of
the s-plane, there should be no sign change in the entries of the rst
column of the Routh-Hurwitz table.
University of Waterloo - SD 352 130 of 186
?
In other words, the following two conditions have to be satised:
4 0.05K > 0 = K < 80
0.1K > 0 = K > 0
= The CLS is stable when 0 < K < 80.
The gain K
c
= 80 is called the critical gain.
When K = K
c
= 80, all the entries in the third row of the Routh-
Hurwitz table become zeros making the CLS critically stable.
In that case, the corresponding auxiliary equation is
A(s) = 2s
2
+ 0.1K
c
= 0
whose solution is s
1,2
= j2 (obviously, the roots are on the j-axis)

c
= 2 rad/s is called the critical angular frequency.
University of Waterloo - SD 352 131 of 186
?
Relative Stability Using the Routh-Hurwitz
Criterion
Example
Problem:
Given the following Ch. Eq.
s
3
+ 9s
2
+ 36s + 80 = 0.
Are the roots of the Ch. Eq. to the left of the line
0
= 1, for
instance?
University of Waterloo - SD 352 132 of 186
?
Solution:
Let us dene a new -axis as
=
0
= + 1
splane
j

+2 1 2
2 1 +1 +2
j
3
splane

+1
Thus, the new s variable is
s = s
0
= s + 1
Substituting s = s 1 into the Ch. Eq. we get
( s 1)
3
+ 9( s 1)
2
+ 36( s 1) + 80 = 0
= s
3
+ 6 s
2
+ 21 s + 52 = 0
University of Waterloo - SD 352 133 of 186
?
Routh-Hurwitz table:
s
3
1 21 0
s
2
6 52 0
s
1
21
52
6
= 12.33 0
s
0
52
No sign change in the rst column
=No roots in the RHS of the s-plane
=No roots to the right of line
0
= 1 of the s-plane
University of Waterloo - SD 352 134 of 186
?
Example
Problem:
Consider the following CLS
c
C(s)
G(s)
R(s)
G (s)
H(s)
G
c
(s) =
K
s
G(s) =
10
s
2
+ s + 10
H(s) =
1
s + 2
Find the range of K for the CLS to be stable.
University of Waterloo - SD 352 135 of 186
?
Solution:
The CLS transfer function is given by
CLTF =
10K
s
3
+s
2
+10s
1 +
10K
s
3
+s
2
+10s

1
s+2
=
10K(s + 2)
s
4
+ 3s
3
+ 12s
2
+ 20s + 10K
The Routh-Hurwitz table:
s
4
1 12 10K
s
3
3 20 0
s
2
16/3 10K
s
1
a
s
0
b
where
_
a =
(1620)/330K
16/3
, and
b = 10K
For the CLS to be stable,
320 90K > 0 K < 32/9, and
10K > 0 K > 0.
Hence, 0 < K < 32/9 is the required gain range for the CLSs stability.
University of Waterloo - SD 352 136 of 186