www.covenantuniversity.

ed

Raising a new Generation of Leaders

Module 4:
State-variable design
Contents
• Introduction
• State feedback controller design using the pole-
placement approach
• State observer design using the pole-placement
approach
• Optimal control design
• Computer simulation
• Tutorial

2
2
4.1 Introduction
• In the classical control system design, the output of a
dynamic system is measured and then compared with
the reference input for the purpose of computing a
control signal based on this comparison.

• Generally, the control signal is computed using a

known controller configuration having the error signal
as its input.

3
3
• This conventional design involves output feedback as
opposed to the modern design in which the state
variables of the system are fed back.

• However, the eventual feedback or closed-loop poles

determine the performance of the system. See Fig. 4.0
for the general control system design steps.

4
Fig. 4.0:
Flowchart
showing
control
system
design
steps

5
5
• Generally, the poles of a system determine the modes or
natural behavior of the system.

• A mode may be stable or unstable, slow or fast,

oscillatory or non-oscillatory depending on the location
of the respective poles in the s-plane.

6
• Therefore, deliberate pole placement or assignment
allows us to modify the behavior of the system.

• This pole placement involves re-locating system poles

to selected locations in the s-plane so that a desired set
of performance criteria can be met.

7
• The pole-placement technique is employed
in the following slides to carry out the
design of:
 a state-feedback controller; and
a state observer.

8
4.2 State-feedback controller design using the pole-
placement approach
Consider the dynamic system defined by
dx t 
 Ax t   Bu t  (4.1)
dt
y  Cx t 
This system, which is open-loop, is shown in Fig. 4.1.
u B + x y
+ ∫ C
x
A
Fig. 4.1: An open-loop control system

9
9
• A controller with gain matrix K can be included in Fig.
4.1 to produce a closed-loop system in Fig.4.2.

r + u B + x y
- + ∫ C
x
A
K
Fig. 4.2: A closed-loop control system with u = -Kx + r

10
•The controller is described by u  Kx  r.
•We are interested in a case where r = 0, that is, a
situation in which only the initial conditions of the
system are reckoned with. Systems of this type are
called regulators. Hence, the control law becomes
u  Kx (4.2)
And the resulting closed-loop system is shown in Fig.
4.3.

11
11
 u B + x y
+ ∫ C
x
A
-K
Fig. 4.3: A closed-loop control system with u = -Kx

Substituting equation (4.2) into the state equations of

equation (4.1) gives the closed-loop dynamic equation
x  A  BKx (4.3)
The characteristic equation of the closed-loop system
given in equation (4.3) is thus:
sI  A  BK  0 (4.4)
12
12
• The state-feedback controller design problem involves
finding a feedback vector K = [k1 k2 . . . kn] such that
the required dynamic properties of the system are met.

• There are three methods of solving this problem, and

they are treated separately in the following.

13
13
4.2.1 Direct substitution method for state-feedback
controller design
• The design steps are as follows:
Confirm that the plant is CONTROLLABLE. This is a
necessary and sufficient condition for state-feedback
controller design.
Decide on pole locations for suitable closed-loop
dynamics. Let the desired closed-loop poles be τ1, τ2, . .
., τn. Form the desired closed-loop characteristic
equation from the desired poles as
ϕd(s) = (s- τ1)(s- τ2) (s- τ3). . . (s- τn) = 0 (4.5)
14
14
Use the closed loop system matrix to evaluate the
actual closed-loop characteristic equation as

c s  sI  A  BK  0 (4.6)
The characteristic polynomial ϕc(s) will involve
elements k1, k2 , . . ., kn of k. These elements are
determined by equating ϕd(s) from equation (4.5) to
ϕc(s) from equation (4.6) and comparing coefficients
of powers of s.

15
15
Example 4.1
 0 1   0
x   u; y  1 0x
Consider the plant described by x    1
  2  3   
Using the direct substitution method, design the state
feedback control u = -Kx such that the closed-loop poles
are [-1, -4].
Solution
Step 1: Check for the controllability of the system.
0 1 
M c  B AB   
1  3
Since the rank of Mc is 2, then the system is completely
controllable, and a state-feedback controller is possible.
16
16
Step 2: The desired closed-loop poles are -1 and -4.
Therefore, the desired characteristic equation is
 d s   s  1s  4   s  5s  4  0
2
Step 3: The actual characteristic equation is
c s   sI  A  BK  0
 s 0  0 1   0
     k k   0
 0 s    2  3  1  1 2
     

c s   s  k 2  3s  k1  2  0
2

17
Step 4: Equating ϕd(s) and ϕc(s), and comparing
coefficients of powers of s, we have
s  5s  4  0  s  k 2  3s  k1  2  0
2 2
Hence,
5  k2  3  k2  5  3  2

4  k1  2  k1  4  2  2
The gain matrix K = [k1 k2]. That is,
And the control law u = -Kx. That is, K  2 2

 x1 
u  2 2  
 x2 
18
18
4.2.2 Transformation matrix method for state-feedback
controller design
• The design steps are as follows:
Confirm that the plant is CONTROLLABLE.
Form the characteristic polynomial of the open-loop
system: sI  A  s n  a n 1    a1s  a 0 (4.7)
n 1s
Decide on pole locations for suitable closed-loop
dynamics. Let the desired closed-loop poles be τ1, τ2, . .
., τn. Form the desired closed-loop characteristic
polynomial: d s   s  1 s   2  s   n 
n
s  n 1    1s   0  0
n 1s (4.8)
19
19
Obtain the transformation matrix P that converts the
system to the controllable canonical form. Note that if
the system is already in this form, then P = I (an identity
matrix).
Hence, the feedback gain matrix K can be found from
K   0  a 0 1  a1  2  a 2   n 1  a n 1  P 1 (4.9)

20
20
Example 4.2: Example 4.1 Revisited
 0 1   0
Consider the plant described by x   x   u; y  1 0x
  2  3  1 
   
Using the transformation matrix method, design the state
feedback control u = -Kx such that the closed-loop poles
are [-1, -4].
Solution
Step 1: Check for the controllability of the system.
0 1 
M c  B AB   
1  3
Since the rank of Mc is 2, then the system is completely
controllable, and a state-feedback controller is possible.

21
21
Step 2: The characteristic polynomial of the open-loop
system is  s 0  0 1   s 1 
sI  A      
 0 s    2  3  2 s  3
     

sI  A  s 2  3s  2
So a1 = 3;a0 = 2.
Step 3: The desired closed-loop poles are -1 and -4.
Therefore, the desired characteristic polynomial is
 d s   s  1s  4   s  5s  4
2

So α1 = 5;α0 = 4.

22
22
Step 4: Since the system is in the controllable
canonical form, P = I.
Step 5: The feedback gain matrix K is
K   0  a 0 1  a1  2  a 2   n 1  a n 1  P 1

  0  a 0 1  a1 I 1  4  2 5  3  2 2
Therefore,
K = [k1 k2] = [2 2].
And the control law u = -Kx. That is,

 x1 
u  2 2  
 x2 
23
23
4.2.3 Ackermann’s formula for state-feedback controller
design
• For a completely state controllable system, the following
expression, called Ackermann’s formula, can be used to
compute the state feedback gain matrix K:
K  0 0  0 1 M c 1 d A  (4.10)
where Mc is the controllability matrix, and ϕd(A) is the
desired characteristic polynomial in A. That is, if
 d s   s n   n 1s n 1   n  2s n  2    1s   0
then
 d A   A   n 1A
n n 1   n 2A n  2    1A   0 I
24
24
Example 4.3: Example 4.1 Revisited
 0 1   0
x   x   u; y  1 0x
Consider the plant described by   2  3  1 
   
Using Ackermann’s formula, design the state feedback
control u = -Kx such that the closed-loop poles are [-1, -
4].
Solution
Step 1: Check for the controllability of the system.
0 1 
M c  B AB   
1  3
Since the rank of Mc is 2, then the system is completely
controllable, and a state-feedback controller is possible.

25
25
Step 2: The desired closed-loop poles are -1 and -4.
Therefore, the desired characteristic polynomial is

 d s   s  1s  4   s 2  5s  4
So
 d A   A 2  5A  4I

 0 1   0 1   0 1  1 0
  .   5   4 
  2  3   2  3   2  3  0 1 
       

 2 2 
 
  4  4
 
26
26
Step 3: Using Ackermann’s formula, we have
K  0 0  0 1 M c 1 d A 
1
0 1   2 2
 0 1    
1  3 - 4 - 4

 3  1  2 2
 0 1     2 2
  1 0  - 4 - 4
K = [k1 k2] = [2 2], and the control law u = -Kx.
That is,

 x1 
u  2 2  
 x2 
27
27
4.3 State observer design using the pole-placement
approach
• Most modern control design techniques, such as pole-
placement, optimal control, etc., require the plant’s state
vector for feedback control purposes.

• There are applications where some of the state variables

may not be available, or it is too costly to measure them
or send them to the controller.

28
28
• If the state variables are not available because of
system configuration or cost, it is possible to estimate
the states from the available outputs.

• Estimated states, rather than actual states, are then fed

to the controller.

29
• An observer, sometimes called an estimator, uses
the model of the plant to calculate the state
variables that are not accessible from the plant.

• When the observer estimates n state variables,

where n is the dimension of the state vector, then
the observer is termed a full-order state observer.

30
30
• But when the number of state variables estimated is
less than n, then the observer is termed a reduced-order
state observer. We will only consider a full-order state
observer here.

• A typical state feedback control design scheme showing

a combination of a controller and an observer is shown
in Fig. 4.4 below.

31
 + x y
B
+ ∫ C
A Plant
0 + u
-
+
L -
+ +
B ∫ C
+
Controller A

K
Observer
Fig.4.4: State-feedback controller-observer scheme

32
32
Consider the plant
  Ax  Bu
x (4.11)
The model of the observer, as can be seen in Fig. 4.4, is
ˆx  Aˆx  Bu  Ly  ˆy  (4.12)
where L is called the observer gain matrix.
The state equations for the observer can be rewritten as
ˆx  Aˆx  Bu  LCx  Cˆx 

 Aˆx  LC ˆx  Bu  LCx
 A  LCˆx  Bu  LCx (4.13)

33
33
Subtracting equation (4.13) from equation (4.11) yields
  ˆx  Ax  Bu  A  LCˆx  Bu  LCx
x

  A  LCx  A  LCˆx
x  ˆx  A  LC x  ˆx  (4.14)

34
e  A  LC e (4.15)

35
35
• The full-order state observer design problem
consists in solving for the values of L to yield a
desired characteristic equation or response for
equation (4.15).
• In other words, the problem reduces to that of pole-
placement where L is found such that the modes of
the estimation error dynamics die away as quickly
as possible.

36
36
• The observer characteristic equation is found from
equation (4.15) as
sI  A  LC   0 (4.16)

37
• Now we select the eigenvalues of the observer to
yield stability and a desired transient response that
is two to five times faster than the controlled
closed-loop response.
• There are also three state observer design methods
which we will consider in the following, but it is
essential to note that observer design is done
independently of controller design.

38
38
4.3.1 Direct substitution method for state observer design
• The design steps are as follows:
Confirm that the plant is OBSERVABLE. This is a
necessary and sufficient condition for state observer
design.
Decide on desired pole locations for the observer
dynamics. Let the desired observer poles be μ1, μ2, . . .,
μn, and form the desired observer characteristic equation
as
ϕd(s) = (s- μ1)(s- μ2) (s- μ3). . . (s- μn) = 0 (4.17)

39
39
Obtain the actual observer characteristic equation as

c  
 s  sI  A  LC  0  (4.18)
The characteristic polynomial ϕc(s) will involve
elements l1, l2 , . . ., ln of L. These elements are
determined by equating ϕd(s) from equation (4.17) to
ϕc(s) from equation (4.18) and comparing coefficients
of powers of s.

40
40
Example 4.4
  1 3  1
Consider the plant described by x   x   u; y  1 0x
 0  2  1
   
Using the direct substitution method, design a state
observer with eigenvalues [-9, -10].
Solution
Step 1: Check for the observability of the system.
 C   1 0
Mo      
CA   1 3

Since the rank of Mo is 2, then the system is completely

observable, and the design of a state observer is possible.

41
41
Step 2: The desired observer poles are -9 and -10.
Therefore, the desired observer characteristic equation
is
d s   s  9 s  10   s  19s  90  0
2

Step 3: The actual observer characteristic equation is

c s   sI  A  LC  0
 s 0    1 3   l1 
     1 0  0
 0 s   0  2 l 
     2

 c s   s  3  l1 s  2 l1  3 l2  2  0
2

42
42
s  19s  90  0  s  3  l1 s  2 l1 3 l2  2  0
2 2

19  3  l1  l1  19  3  16

90  2 l1  3 l2  2  l2  90  34  / 3  18 .67

43
43
4.3.2 Transformation matrix method for state observer
design
• The design steps are as follows:
Confirm that the plant is OBSERVABLE.
Form the characteristic polynomial of the open-loop
n n 1
system: sI  A  s  a n 1s    a1s  a 0 (4.19)
Decide on desired pole locations for the observer
dynamics. Let the desired observer poles be μ1, μ2, . . .,
μn, and form the desired observer characteristic equation
as d s   s  1 s   2  s   n 
n
s  n 1    1s   0
n 1s (4.20)
44
44
Obtain the transformation matrix Q that converts the
system to the observable canonical form. Note that if the
system is already in this form, then Q = I (an identity
matrix).
Hence, the observer gain matrix L can be found from
 0  a 0 
 
 1  a1 
 
L  Q.   2  a 2  (4.21)
 
  
 
  
 n 1 a n 1 
45
45
Example 4.5   1 3  1
x   x   u; y  1 0x
Consider the plant described by  0  2  1
   
Using the transformation matrix method, design a state
observer with eigenvalues [-9, -10].
Solution
Step 1: Check for the observability of the system.
 C   1 0
Mo      
CA   1 3
Since the rank of Mo is 2, then the system is completely
observable, and the design of a state observer is possible.

46
46
Step 2: The characteristic polynomial of the open-loop
system is
 s 0  1 3 
sI  A    
 0 s   0  2
   
So a1 = 3;a0 = 2.  s 2  3s  2
Step 3: The desired observer poles are -9 and -10.
Therefore, the desired observer characteristic polynomial
is

So α1 = 19;α0 = 90.
 d s   s  9 s  10   s 2  19s  90

47
47
Step 4: Since the system is not in the observability
canonical form, the transformation matrix Q is found as
follows. Q  MV 1
where a1 1 3 1  C   1 0
M   ;V    
 1 0 1 0 CA   1 3
1
 3 1  1 0  0 1 
Q     .    
 1 0  1 3  1 3  2 3
  
Step 5: Therefore, the observer gain matrix L is
 0 1   0  a 0   0 1  90  2
L .   .  

1 3  
 2 3  1  a1 
 
1 3  2 3
 19  3 

 16   16 
L  
56 3 18.67
48
48
4.3.3 Ackermann’s formula for state observer design
• For a completely state observable system, Ackermann’s
formula for computing the observer gain matrix L is
0 
 
0 
 
L  d A  M o 1    (4.22)
 
0 
 
1
 
where Mo is the observability matrix, and ϕd(A) is the
desired observer characteristic polynomial in A. That is,
d A   d s 
sA
49
49
Example 4.6
  1 3  1
Consider the plant described by x   x   u; y  1 0x
 0  2  1
   
Using Ackermann’s formula, design a state observer with
eigenvalues [-9, -10].
Solution
Step 1: Check for the observability of the system.
 C   1 0
Mo      
CA   1 3

Since the rank of Mo is 2, then the system is completely

observable, and the design of a state observer is possible.

50
50
Step 2: The desired observer poles are -9 and -10.
Therefore, the desired observer characteristic
polynomial is
 d s   s  9 s  10   s 2  19s  90
So  d A   A 2  19 A  90 I

 1 3   1 3   1 3  1 0
  .   19   90 
 0  2  0  2  0  2 0 1
       

 72 48 
 
 0 56 
 
51
51
Step 3: Using Ackermann’s formula, we
have 0 
 
L  d A M o 1    
 
1
1
72 48  1 0 0
    
 0 56   1 3 1

72 48  1 0  0  16 

    
 0 56  1 3 1 3 1 18.67 

52
52
4.4 Optimal Control
• Determination of the state-feedback gain matrix based
on pole placement has been the subject of discussion
so far. Several sets of pole locations can be tried, and
the system performance is observed for these various
sets.
• The set of poles that best satisfies the performance
specifications is then chosen. Extensive system
simulation is very useful in this regard.

53
53
• The most effective, widely used, and systematic way of
finding the gain matrix K is the so-called linear
quadratic regulator (LQR).
• The design problem is that of finding a control input, u,
that minimizes the quadratic cost function


J   x Qx  u Ru dt
T T
(4.23)
0

where Q and R are weighting factors; Q is a

symmetric matrix, and R relates to the control energy
expenditure. Note that Q ≥ 0 (i.e., the eigenvalues of
Q are nonnegative) and R > 0 (i.e., the eigenvalues
of R are positive).
54
Design steps for state-feedback optimal controller design
• Given the plant dx t   Ax t   Bu t  and y  Cxt ,
dt
the design steps are as follows:
Confirm that the plant is CONTROLLABLE. This is a
necessary and sufficient condition for state-feedback
optimal controller design.
Obtain suitable values for Q and R. Note that Q is
positive-definite (or positive-semidefinite) and R is
positive-definite.

55
Determine the value of k (for the optimal control u = -
kx) as k  R B P
1 T (4.24)
where P is a symmetric matrix whose eigenvalues are
positive.
Determine P, which is the solution to the algebraic
Riccati equation (ARE)
A P  PA  PBR B P  Q  0
T 1 T
(4.25)
The parameters k1, k2 , . . ., kn are extracted from k, as
obtained in equation (4.24). The values of these
parameters make the closed-loop poles optimal.

56
Example 4.7
Determine the optimal gain matrix k for the plant
described by
  2 0 1
x    x   u; y  1 0x.
 1 0 0
1 0
Q    and R  1.
0 4
Solution
Step 1: The rank of the controllability matrix Mc is 2, so
the system is completely controllable.

57
Step 2: The values of Q and R are given, and they are
both positive-definite.
Steps 3 and 4: To determine optimal gain k, matrix P is
first obtained from
A T P  PA  PBR 1BT P  Q  0
 - 2 0   p11 p12   p11 p12  - 2 0   p11 p12  1   p11 p12 
T
1 0
           1 0    
 1 0   p21 p22   p21 p22  1 0   p21 p22  0   p21 p22   0 4
 - 2p11  p21 - 2p12  p22   - 2p11  p12 0   p112 p11 p12  1 0
          
 0 0   - 2p 21  p22 0   p11 p21 p21 p12  0 4
 - 4p11  p12  p21  p11 - 2p12  p22  p11 p12  1 0
2

    
 - 2p 21  p22  p11 p21  p21 p12   0 4

58
Four equations result as follows:
- 4p11  p12  p21  p112  1;
- 2p12  p22  p11 p12  0;
- 2p 21  p22  p11 p21  0; and
 p21 p12  4;
1 2
Thus, P   
 2 6
and
1 2
k  1.1 0.   1 0
 2 6
59