Lecture 7

Multivariable Control
Systems
Ali Karimpour
Associate Professor
Ferdowsi University of Mashhad

Lecture 7

References are appeared in the last slide.

Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability, Observability and Realization

Topics to be covered include:

• Controllability and Observability of Linear Dynamical Equations
• Output Controllability and Functional Controllability
• Realization of Proper Rational Transfer Function Matrices
• Model Order Reduction of Non-Minimal Representations
• Model Order Reduction of Minimal Representations
Truncation Method
Residualization Method
Hankel Norm Approximation
2
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability of Linear Dynamical Equations

Definition 7-1
The state equation x Ax  Bu or the pair (A, B) is said to be controllable
if for any initial state x0 and any final state x1, there exists an input that transfers
x0 to x1 in a finite time. Otherwise (A,B) is said to be uncontrollable

Definition 7-2
The state equation
x Ax  Bu
y  Cx  Du
or the pair (A,C) is said to be observable if for any unknown initial state x0 , there
exists a finite time t1 > 0 such that the knowledge of the input u and the output y
over [0,t1] suffices to determine uniquely the initial state x0.
Other wise, the equation is unobservable.
3
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability of Linear Dynamical Equations

Theorem 7-1: Controllability

4
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability of Linear Dynamical Equations

Theorem 7-2: Observability

5
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability in Rosenbrock’s system matrix

Rosenbrock’s system matrix is:

 P( s ) Q( s) 
P( s )    G ( s )  R( s ) P( s) 1 Q( s)  W ( s)
  R ( s ) W ( s ) 
Now suppose one find the greatest left common factor of P and Q as

P(s)  L(s) P (s) Q(s)  L(s)Q (s)

Now if L(s) is not unimodular then there is i.d.z and so reduced order
system is:
G(s)  R(s) P(s)1 Q(s)  W (s)  R(s) P (s)1 Q (s)  W (s)
How to derive i.d.z. ?
[ P ( s ) Q ( s )] Smith form [ S ( s) 0]
6
Input decoupling zeros are roots of |S(s)|=0
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability in Rosenbrock’s system matrix

Rosenbrock’s system matrix is:

 P( s ) Q( s) 
P( s )    G ( s )  R( s ) P( s) 1 Q( s)  W ( s)
  R ( s ) W ( s ) 
Now suppose one find the greatest right common factor of P and R as

P( s)  Pˆ ( s) L( s) R( s)  Rˆ ( s) L( s)
Now if L(s) is not unimodular then there is o.d.z and so reduced order
system is:
G(s)  R( s) P( s) 1 Q( s)  W (s)  Rˆ (s) Pˆ ( s) 1 Q( s)  W ( s)
How to derive o.d.z. ?
 P( s )  S ( s)
 R( s ) Smith  form  0 
   
7
Output decoupling zeros are roots of |S(s)|=0
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability of Linear Dynamical Equations

Example 7-1: Find the i.d.z. and o.d.z. of following system.

s( s  1) 0 s
P( s)   0 s( s  2) s 
  1 1 0

 s( s  1) 0 s  s 0 0
P(s) Q(s)  
Smith Form

  0 s 0 
 0 s ( s  2) s   
So intput decoupling zeros are: |S(s)|=0  0 and 0
s( s  1) 0  1 0
 P( s )   Smith Form
0 s 

 R( s )  0 s ( s  2)    
   1 1  0 0

So output decoupling zero is: |S(s)|=0  0 8
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability and Observability of Linear Dynamical Equations

Example 7-2: Reduce the following system if it is possible.

s( s  1) 0 s
P( s)   0 s( s  2) s  Clearly order of system is 4.
  1 1 0

 s 0  s  1 0   s 0 1
P( s)   Q( s)    1
  s  2  0 s  
0 s   0
So the reduced order system is:
s  1 0 1
P( s)   0 s  2 1 Clearly order of system is 2.
  1  1 0
Exercise 7-1: Find the i.d.z. and o.d.z. of following system and also check the
controllability and observability of system.
( s  1)
2 3
s 
P( s )   
 1 2  s 9
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability, Observability and Realization

• Controllability and Observability of Linear Dynamical Equations

• Output Controllability and Functional Controllability
• Realization of Proper Rational Transfer Function Matrices
• Model Order Reduction of Non-Minimal Representations
• Model Order Reduction of Minimal Representations
Truncation Method
Residualization Method
Hankel Norm Approximation

10
Dr. Ali Karimpour Jan 2017
 Lecture 7

Output Controllability

Definition 7-3: Output Controllability

Dynamical system
x (t )  Ax (t )  Bu (t )
y (t )  Cx (t )  Du (t )


rank CB CAB ... CAn1B D  p 
11
Dr. Ali Karimpour Jan 2017
 Lecture 7

Functional Controllability

Definition 7-4: Functional controllability.

An m-input l-output system G(s) is functionally controllable if the
normal rank of G(s), denoted r, is equal to the number of outputs; that is,
if G(s) has full row rank. A plant is functionally uncontrollable if r < l.
Remark 1: The minimal requirement for functional controllability is
that we have at least many inputs as outputs, i.e. m ≥ l
Remark 2: A plant is functionally uncontrollable if and only if
 l (G( j))  0 , 
Remark 3: For SISO plants just G(s)=0 is functionally uncontrollable.

Remark 4: A MIMO plant is functionally uncontrollable if the gain is

identically zero in some output direction at all frequencies. 12
Dr. Ali Karimpour Jan 2017
 Lecture 7

Functional Controllability
Example 7-3: a)A Functionally controllable system that is not state controllable.
0  3 0 0  3 2
1  4 0 0  1 2
 1 2  x (t )    x(t )   u (t )
 s 1 0 0 0  1  1 
s  3
1
G(s)      
1 1   0 0 1  2  1 1
 
 s 1 s 1 0 1 0 0 
y (t )    x(t )
0 0 0 1 
b) A state and output controllable system that is not Functionally controllable.

0 0 0  0 1 
1 1
x (t )  1 0 0 x(t )  1 0u (t ) s
G(s)   s2 
0 1 0 0 0 1 1
 2 
0 1 0  s s 
3

y (t )    x(t ) 13
0 0 1  Dr. Ali Karimpour Jan 2017
 Lecture 7

Functional Controllability

Example 7-4: dc-dc boost converter

 2498s  3.049 106 

 xˆ1   s 2  609s  3.207  105  x1
 xˆ     
ˆ1 , functionally uncontrollable x2
  2.5 
 2   2498s  3.049 10 
10 5
s  1.217 610 8

 2 609s  3.207  1055  1
 xˆ1   ss2 609 s  3.207 10  ˆ
or    1 , functionally uncontrollable
ˆ new system design :
x2   2.5 10 suncontrollable
 1.217  10 
 functionally
5 8

 s 2 2498
 609ss3.049
3.20710105 
6
12.5 s + 7440 
 xˆnew  2 5  ˆ1 
1  system      
5 2
or  s 609
designs 3.207
: 1 0 s 609 s 3.207 10 
 xˆ   vˆ  , functionally controllable
 2   2.5 10  1.217
10610 6.2s5+710   in 
5 8 5
2498 s  s3.049 12.5 440
 2 609s  3.207  1055 s22  609s  3.207  1055
 xˆ1   ss2 609 s  3.207 10 s  609s  3.207 10  ˆ1 
 xˆ    vˆ  , functionally controllable
 2   2.5 10 s  1.217  10 6.25  10   in 
5 8 5

 s 2  609s  3.207 105 s 2  609s  3.207  105 

14
functionally controllable Dr. Ali Karimpour Jan 2017
 Lecture 7

Functional Controllability

An m-input l-output system G(s)  C(sI  A) 1 B is functionally uncontrollable if

1- The system is input deficient or rank( B)  l

2- The system is output deficient or rank(C)  l

3- The system has fewer states than outputs rank(sI  A)  l

If the plant is not functionally controllable, i.e. r  l then there are
l-r output directions, denoted yi which cannot be affected.
yiH ( j )G( j )  0 i  1,..., l  r

From an SVD of G(jω) the uncontrollable output directions yi(jω) are

the last l-r columns of Y(jω). 15
Dr. Ali Karimpour Jan 2017
 Lecture 7

Functional Controllability

Example 7-5:
 1 2 
 s 1 s 1 
G( s)  
2 4 
 
s  2 s  2

This is easily seen since column 2 of G(s) is two times column 1.

The uncontrollable output directions at low and high frequencies are,

respectively,

1 1 1 2
y0 (0)   1 , y0 ( )   1
2   5 
16
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability, Observability and Realization

• Controllability and Observability of Linear Dynamical Equations

• Output Controllability and Functional Controllability
• Realization of Proper Rational Transfer Function Matrices
• Model Order Reduction of Non-Minimal Representations
• Model Order Reduction of Minimal Representations
Truncation Method
Residualization Method
Hankel Norm Approximation

17
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Transfer Function Matrices

Dynamical equation The input-output description

(state-space) description This transformation (transfer function matrix)
x  Ax  Bu is unique.
y  Cx  Eu G(s)  C(sI  A) 1 B  E
The input-output description Dynamical equation
(transfer function matrix) Realization (state-space) description
is not unique
1 x  Ax  Bu
G(s)  C(sI  A) B  E
y  Cx  Eu
Theorem 7-3
A transfer function matrix G(s) is realizable by a finite dimensional
linear time invariant dynamical equation if and only if G(s) is a proper
rational matrix.
18
Proof: See “Linear system theory and design” Chi-Tsong Chen
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realizations
Definition 7-6
A linear time-invariant dynamical equation is said to be reducible if and only if there
exist a linear time-invariant dynamical equation of lesser dimension that has the same
transfer function matrix. Otherwise, the equation is irreducible.
Theorem 7-5
A linear time invariant dynamical equation is irreducible if and only if it is controllable
and observable.
Theorem 7-6

19
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realizations
Definition 7-5: Characteristic polynomial and degree of G(s)

Theorem 7-4:

20
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of proper rational transfer functions

ˆ0 s n  ˆ1s n1  ......  ˆn 1 s n1   2 s n2  ......   n ˆ0

g ( s)  , ˆ 0  0 g ( s)  
 0 s  1s  ......   n
ˆ n
ˆ n 1
ˆ s  1 s  ......   n
n n 1
ˆ 0
There are different forms of realization
 Observable canonical form  Controllable canonical form
 x  0
1
0    x   
0 ... n 1 n   x   0
1
1 0 ... 0   x  0 
1
 x  1 0   x  
0 ...   x   0 0 1 ... 0   x  0 
  
2
   n 1 2 n 1
   
2
   
2

 x   0 0     x   
1 ... n2 n2
u  x    . . . ... .   x   0  u
  
3
  
3
   
3
   
3

 .  . .
. ... .  .   .  .  0 0 0 ... 1  .   . 
 x  0
n
1     x   
0 ... 1 n 1
  x   
n n
 n 1
 n2
...   1
  x  1
n

x  1 x  1

x  x 
  ˆ 2
  ˆ
2

y  0 0 ... 0 1 x   u 0
y     ...   x   u 0

  ˆ   ˆ
3 n n 1 n2 1 3
0

.
0

.
 x  n
 x 
n

21
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of proper rational transfer functions

Realization from the Hankel matrix (Minimal)

 0 s n  1s n 1  ......   n g (s)  h(0)  h(1)s 1  h(2)s 2  h(3)s 3  ......
g ( s)  n
s  1s n 1  ......   n
The coefficients h(i) will be called Markov parameters.

 h(1) h(2) h(3) ... h(  ) 

 h(2) h(3) h(4) ... h(   1) 

 h(3) h(4) h(5) ... h(   2) 
H ( ,  )   
 . . . ... . 
 . . . ... . 
 
h( ) h(  1) h(  2) ... h(    1)

22
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of proper rational transfer functions

Realization from the Hankel matrix

Theorem 7-7: Consider the proper transfer function g(s) as

 0 s n  1 s n 1   2 s n 2  ......   n
g ( s) 
s n   1 s n 1  ......   n

then g(s) has degree m if and only if

 H (m, m)    H (m  k , m  l )  for every k , l  1, 2 , 3 , ....

23
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of proper rational transfer functions

Now consider the dynamical equation

x  Ax  bu g (s)  c(sI  A) 1 b  e  s 1c( I  s 1 A) 1 b  e
y  cx  eu
 e  cbs 1  cAbs 2  cA2bs3  .....
h(i)  cAi 1b i  1, 2 , 3 , ......
 h(1) h(2) ..... h( n) 
 h(2) ..... h(n  1) 

 . . ..... . 
H (n  1, n)   
 . . ..... . 
 h( n) h(n  1) ..... h(2n  1)
 
h(n  1) h(n  2) ..... h(2n) 
Let the first m rows be linearly independent and the (m +1) th row of H(n+1,n) be
linearly dependent on its previous rows. So

[a1 a2 ..... am 1 0 ..... 0] H (n  1, n)  0 24

Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of proper rational transfer functions

[a1 a2 ..... am 1 0 ..... 0] H (n  1, n)  0

We claim that the m -dimensional dynamical equation
 0 1 0 ..... 0 0   h(1) 
 0 0 1 ..... 0 0   h(2) 
  
 0 0 0 ..... 0 0   h(3) 
   
x  . . . . ..... . .  x   . u (I)
 . . . ..... . .   . 
   
 0 0 0 ..... 0 1  h(m  1)
 a a a .....  am 1  am   h(m) 
 1 2 3

y  1 0 0 ..... 0 0x  h(0)u

is a controllable and observable (irreducible realization).

Exercise 7-2: Show that (I) is a controllable and observable (irreducible realization) of
x  Ax  bu
25
y  cx  eu Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of proper rational transfer functions

Example 7-6: Derive three different realization for following system.

2s 3  18s 2  48s  32 6s 2  26s  20
g ( s)   3 2
s  6s  11s  6
3 2
s  6s  11s  6
2

Observable canonical form realization is: Controllable canonical form realization is:
 x1  0 0  6   x1  20  x1   0 1 0   x1  0
 x   1 0  11  x   26u  x    0 0 1   x   0  u
   2    
2   2  2   
 x 3  0 1  6   x3  6   x3   6  11  6  x3  1 
 x1   x1 
y  0 0 1 x 2   2u y  20 26 6 x2   2u
 x3   x3 
 6  10 14  We can show that the rank of H(4,3) is 2. So
 10 14  10
H (4,3)    0 1  6 
 14  10  34 6 5 1 0H (4,3)  0 
x   6  5 x   10u
     
 10  34 230  y  1 0x  2u 26
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

There are many approaches to find irreducible realizations for proper

rational matrices.

1. One approach is to first find a reducible realization and then apply

the reduction procedure to reduce it to an irreducible one.

2. In the second approach irreducible realization will yield directly.

27
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Method I: Gilbert diagonal representation.

Each element of G(s) has real distinct poles.

Reminders

Gk  Ck Bk Ck is q  k Bk is k  p


A  diag 1I 1 , ... , r I r 
 B1 
C  C1 ... Cr  B    
 Br  28
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-7: Derive Gilbert diagonal representation for following system.
 1 2s  4 
 s 1 s  1s  4
G ( s)  
 1 1 
 s  1s  2 s2 
 1 2 1 0 0  0 
  1 0  0 1  11 1 01 1
      

1  1 2 1 0 0  1 0 0 
G(s)   
s  1  1 0 s  2 1 1 s  4 0 0
   

 1 0 0 0  1 0
 0 1 0 0  0 1  1 0 0  1 0
x   x   u x   0  1 0  x  0 1u
 0 0  2 0  1 1
    minimal realization  0 0  2 1 1
0 0 0  4 1 1 

 1 2 0 0  1 2 0
y y x
 x   1 0 1 
 1 0 1 0  29
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Method I: Gilbert diagonal representation.

Repetitive real poles.
 2 1 0 0  0 0 0  1 1 1 1 
0 0  0 0 0 s  2
x  
2 1
x 
s  22 s  23 s  24 
0 0  2 1  0 0 0
u
a k  0 1 1 1  0 0 0
d g  0

0 0 0  2 1
  
2 3 G ( s )   b e h

l   s2 s  22 s  23  0 0
 1 1  0 0 0
a k  c f i m  0 0 2 
 
d g
 s2 s  2  1 2 3
y  b e h l  x  1 
 c m  0 0 0 
f i  s2 

a  d  g  k 
1    e 1 2 3  1  h 1 2 3  1  l 1 2 3
G( s)  b 1 2 3 
1
s  24   s  23  
 f 
s  2
2  
 i 
s2 
m
c 

30
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Method I: Gilbert diagonal representation.

Repetitive real eigenvalues.

r1 Jordan block of order 3

r2 - r1 Jordan block of order 2

r3 – r2 Jordan block of order 1

31
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-8: Derive Gilbert diagonal representation for following system.

2 Jordan block of order 2

0 Jordan block of order 1

32
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-9: Derive Gilbert diagonal representation for following system.

33
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-10: Derive Gilbert diagonal representation for following system.

C (:,1)
B(4, :)

C (:,2) C(:,5)
B(4, :)

34
B(7, :) Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-10: Derive Gilbert diagonal representation for following system.
C(:,3) C(:,5) B(4, :)
 0 0 1   1 1  2
M 3   0 0 0  1 0  1 B(7, :)
 1 5 3  1 0 1 
C(:,8) B(9,:)

C (:,4) C (:,7) B(4, :)

0 0  1  1 1  2
M 4  0 0 0   1 0  1 B(7, :)
0  1 0   1 0 1 
C(:,9) B(9,:)

35
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Example 7-10: Derive Gilbert diagonal representation for following system.

36
Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of Proper Rational Matrices

Method II: Hankel form realization of a proper G(s). Let

G(s)  H (0)  H (1)s 1  H (2)s 2  .......... .

Consider the monic least common denominator of G(s) as

 (s)  s m  1 s m1   2 s m2  ...   m
Then after deriving H(i) one can simply show
H (m  i)   H (m  i  1)   H (m  i  2)  ...   H (i)
1 2 m
i  1 (I )

Let {A, B, C and E} be a realization of G(s) then we have

G(s)  E  C(sI  A) 1 B  E  CBs 1  CABs 2  CA 2 Bs 3  .......... .
Then {A, B, C and E } be a realization of G(s) if and only if
E  H (0) H (i  1)  CA B i  0 , 1, 2 ,....
i

37
Exercise 7-3: Proof equation (I)(just PhD students) Dr. Ali Karimpour Jan 2017
 Lecture 7

Realization of proper rational transfer functions

Then {A, B, C and E } be a realization of G(s) if and only if

E  H (0) H (i  1)  CA B i  0 , 1, 2 ,....
i

There are different forms of realization

 Observable canonical form  Controllable canonical form
   H (1)    I  p

   H ( 2)    0 
        p

x   M x   .. u x   N  x   .. u
       
   H ( m  1)    0  p

   H (m)    0 
p

y  I q
0 0 ... 0x  H (0)u y  H (1) H (2) h(3) ... H (m)x  H (0)u
 0q Iq 0q ... 0q  0 p 0p ... 0 p mI p 
 0  I 0p ... 0 p   m 1 I p 
 q 0q Iq ... 0q   p
M  . . . ... .  N  0 p Ip ... ..0 p .   m  2 I p 
   
 0q 0q 0q ... Iq   . . ... ... . 
  m I q   m 1 I q   m2 I q ...   1 I q  0 p   1 I p 
  0p ... I p
38
Proof ? Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of Proper Rational Matrices

Now we shall discuss in the following a method which will yield directly
irreducible realizations. This method is based on the Hankel matrices.
We also define the two following Hankel matrices

 H (1) H (2) H ( m)   H (2) H (3) H (m  1) 

 H (2) H (3) H (m  1)   H (m  2)
~  H (3) H (4)
T  T 
 . . .   . . . 
   
 H (m) H (m  1) H (2m  1)  H (m  1) H (m  2) H ( 2m) 

Derive SVD of T

39
Dr. Ali Karimpour Jan 2017
 Lecture 7
Irreducible realization of Proper Rational Matrices
Derive SVD of T

40
Exercise 7-4: Proof theorem 7-8. Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible realization of Proper Rational Matrices

Example 7-11: Derive an irreducible realization for the following proper rational function.
  2s 2  3s  2 1 
 
G( s)   ( s  1) 2
s 
 4s  5  3s  5 
 s 1 s  1 
Least common denominator of G(s), is
 (s)  s(s  1) 2
 2 0  1 1  1  2 0 2 3 0  3  4 0  4 5 0  5  6 0 6
G(s)     1  2 s    1 2 s  1  2 s    1 2 s  .1  2 s    1 2 s .....
 4  3            

1 1 2 0 3 0
Non-zero singular values of T
 1  2 1 2 1  2  are 10.23, 5.79, 0.90 and 0.23.
 H (1) H (2) H (3)   
 2 0 0 4 0 
T   H (2) H (3) H (4)  
3
 So, r = 4.
 1 2 1  2  1 2
 H (3) H (4) H (5)   
3 0 4 0 5 0
 
 1  2  1 2 1  2 41
Dr. Ali Karimpour Jan 2017
 Lecture 7

Irreducible Realization of Proper Rational Matrices

- 0.3413 0.2545 - 0.8902 - 0.1621  - 0.4003 - 0.0196 0.4905 - 0.6574

- 0.2357 - 0.5238 - 0.0581 - 0.0071  0.1049 0.5872 - 0.6022 - 0.5306
  
 0.5127 - 0.2078 - 0.1054 - 0.8264   0.5496 - 0.1057 - 0.0978 0.1026 
Yr    Ur   
 0.2357 05238 0.0581 0.0071   - 0.1382 - 0.5432  0.3875 - 0.1888
- 0.6738 0.2627 0.4316  0.5392  - 0.6989 0.2311 - 0.2949 0.4522 
   
- 0.2357 - 0.5238  0.0581  0.0071  0.1382 0.5432 0.3875 0.1888 

 - 1.0915 0.6121 - 0.8443 - 0.0770 

 - 0.7539 - 1.2598 - 0.0551 - 0.0034

 1.6398 - 0.4999 - 0.1000 - 0.3923 
Yˆ  Yr S 1 / 2  
 0.7539 1.2598 0.0551 0.0034 
 - 2.1553 0.6317 0.4093 - 0.2560 
 
 - 0.7539 - 1.2598 - 0.0551 - 0.0034
 - 1.2803 0.3355 1.7579 - 0.4421 - 2.2356 0.4421 
 - 0.0471 1.4124 - 0.2543 - 1.3066 0.5557 1.3066 
Uˆ  S 1 / 2U r  
H

 0.4652 - 0.5711 - 0.0927 - 0.3675 - 0.2797 0.3675

 
 - 0.3121 - 0.2519 0.0487 - 0.0896 0.2147 0.0896  42
Dr. Ali Karimpour Jan 2017
 Lecture 7
Irreducible Realization of Proper Rational Matrices
 - 0.1067 - 0.0737 0.1603 0.0737 - 0.2107 - 0.0737
 0.1058 - 0.2178 - 0.0864 0.2178 0.1092 - 0.2178 
Y  S Yr  
ˆ † 1 / 2 
H

 - 0.9386 - 0.0613 - 0.1112 0.0613 0.4551 - 0.0613 

 
 - 0.3415 - 0.0149 - 1.7406 0.0149 - 1.1356 - 0.0149
 - 0.1251 - 0.0081 0.5171 - 1.3847 
 0.0328 0.2441 - 0.6349 - 1.1176 
 
 0.1718 - 0.0440 - 0.1031 0.2161 
Uˆ †  U r S 1 / 2   
 - 0.0432 - 0.2259 - 0.4086 - 0.3976
 - 0.2185 0.0961 - 0.3109 0.9525 
 
 0.0432 0.2259 0.4086 0.3976 
 - 1.2497 0.0369 0.2155 - 0.1904   - 1.2803 0.3355 
 0.1588 - 1.0139 - 0.1604 0.0772   - 0.0471 1.4124 
~
A  Yˆ T Uˆ  
† †  B  UˆI p , pm  
T 
 - 0.2227 - 0.1800 - 0.2888 0.8076  0.4652 - 0.5711 
   
 0.1246 - 0.1181 0.1354 - 0.4476   - 0.3121 - 0.2519
 - 1.0915 0.6121 - 0.8443 - 0.0770   2 0 
C  I q ,qmYˆ    E  H (0)   
 - 0.7539 - 1.2598 - 0.0551 - 0.0034  4  3
Exercise 7-5: Derive state space model of g(s) by theorem 7-8.
2s 3  18s 2  48s  32
g ( s)  43
s 3  6s 2  11s  6 Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability, Observability and Realization

• Controllability and Observability of Linear Dynamical Equations

• Output Controllability and Functional Controllability
• Realization of Proper Rational Transfer Function Matrices
• Model Order Reduction of Non-Minimal Representations
• Model Order Reduction of Minimal Representations
Truncation Method
Residualization Method
Hankel Norm Approximation

44
Dr. Ali Karimpour Jan 2017
 Lecture 7

Model Order Reduction of Non-Minimal Representations

Theorem 7-9 The controllability and observability of a linear time-invariant dynamical

equation are invariant under any similarity transformation.
Theorem 7-10 x  Ax  Bu
Consider the n-dimensional linear time –invariant dynamical equation y  Cx  Eu

If the controllability matrix of the dynamical equation has rank n1 (where n1<n ), then
there exists an equivalence transformation
x  Px
which transform the dynamical equation to
 xc   Ac A12   xc   Bc   xc 
         u 
y  Cc 
Cc    Eu
 xc   0 Ac   xc   0   xc 
and the n1-dimensional sub-equation
x c  Ac xc  Bc u
y  C c xc  Eu 45
is controllable and has the same transfer function matrix as the first system.
Dr. Ali Karimpour Jan 2017
 Lecture 7

Model Order Reduction of Non-Minimal Representations

Theorem 7-11 x  Ax  Bu
Consider the n-dimensional linear time –invariant dynamical equation y  Cx  Eu

If the observability matrix of the dynamical equation has rank n2 (where n2<n ), then
there exists an equivalence transformation

x  Px
which transform the dynamical equation to
 xo   Ao 0   xo   Bo   xo 
         u 
y  Co 
0    Eu
 xo   A21 Ao   xo   Bo   xo 
and the n2-dimensional sub-equation
xo  Ao xo  Bou
y  Co xo  Eu
46
is observable and has the same transfer function matrix as the first system.
Dr. Ali Karimpour Jan 2017
 Lecture 7

Model Order Reduction of Non-Minimal Representations

Theorem 7-12 (Canonical decomposition theorem) x  Ax  Bu

Consider the n-dimensional linear time –invariant dynamical equation
y  Cx  Eu
There exists an equivalence transformation
x  Px
which transform the dynamical equation to
 xco   Aco A12 A13   xco   Bco   xco 
 x    0
 co   Aco

A23   xco    Bco u
   
y 0  Cco
 

Cc  xco   Eu
 xc   0 0 Ac   xc   0   xc 

and the reduced dimensional sub-equation

x co  Aco xco  Bco u
y  Cco xco  Eu
is observable and controllable and has the same transfer function matrix as the first
system. 47
Dr. Ali Karimpour Jan 2017
 Lecture 7

Controllability, Observability and Realization

• Controllability and Observability of Linear Dynamical Equations

• Output Controllability and Functional Controllability
• Realization of Proper Rational Transfer Function Matrices
• Model Order Reduction of Non-Minimal Representations
• Model Order Reduction of Minimal Representations
Truncation Method
Residualization Method
Hankel Norm Approximation

48
Dr. Ali Karimpour Jan 2017
 Lecture 7

Model Order Reduction of Minimal Representations

Consider following system

 x  A A  x  B 
  A 
1 11 12 1 1

x (t )  Ax (t )  Bu (t )  x    u
 2   21
A  x  B
22 2 2 
y (t )  Cx (t )  Du (t ) x 
y  C C    Eu
1
1 2
x  2

There are several model order reduction procedure:

• Truncation Method.
• Residualization Method (Singular Perturbation).

• Hankel norm truncation Method.

• Hankel norm residualization Method (Singular Perturbation).

• ……………….. 49
Dr. Ali Karimpour Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Truncation Method
Consider following system  x  A A  x  B 
  A 
1 11 12 1 1

 x    u
 2   21
A  x  B
22 2 2 
x 
y  C C    Eu
1
1 2
x  2

• Truncation Method. Let x2=0 x1  A11 x1  B1u

y  C1 x1  Eu
High frequency response is not changed by truncation method.
G()  G ()  E r

• Residualization Method (Singular Perturbation). Let x2  0

1 1
x1  ( A11  A12 A22 A21 ) x1  ( B1  A12 A22 B2 )u
1 1
y  (C1  C2 A22 A21 ) x1  ( E  C2 A22 B2 )u
Exercise 7-6: Show that steady state behavior is not changed by
residualization method 50
G (0)  Gr (0) Dr. Ali Karimpour Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Truncation Method

Truncation procedure
1 0 ... 0  b1T  1 0 ... 0  b1T 
0   
... 0  b2T  0  ... 0 
 T
b2 
x   2
x u Truncation Method xr   2
xr   u
 ... ... ... ...   ...    ... ... ... ...   ... 
       T
0 0 ... n  bnT  0 0 ... k  bk 
y  c1 c2 ... cn x  Eu y  c1 c2 ... ck x  Eu

1 , 2 ,..., k are dominant poles and others are insignificant.

n
ci biT
G ( s)  Gr ( s)  
i  k 1 s  i
Error is Error value related to:
n
 (c b )
T
G( s)  Gr ( s)    i i

i  k 1 Re(i ) 51
Dr. Ali Karimpour Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Hankel Norm Approximation

Consider following system x  Ax  Bu

y  Cx  Eu
Controllability gramians and observability gramians are:

P   e BB e dt
At T AT t
0

Q   e C T Ce At dt
AT t
0

Minimum energy required to steer the state of system from 0 to xr is:

u  xrT P 1 xr
2

Maximum energy produced by observing the output of the system with initial state
x0 is:
y  x0T Qx0
2
52
Dr. Ali Karimpour Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Hankel Norm Approximation

Consider following system x  Ax  Bu

y  Cx  Eu
Controllability gramians and observability gramians are changed by similarity
transformation.
A balanced realization is a realization with following property.

P  Q    diag1  2 ...  n   i   i 1 Hankel singular values

If  k   k 1 k is suitable value for reduced order realization.

 x  A A  x  B 
  A   u
1 11 12 1 1

x (t )  Ax (t )  Bu (t ) A balanced realization  x
2   21
  
A  x  B 
22 2 2

y (t )  Cx (t )  Du (t ) x 
y  C C    Eu
1

Hankel norm truncation method. 1 2

x  2 53
Hankel norm residualization Method Dr. Ali Karimpour Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Hankel Norm Approximation

Example 7-12: Consider following system.

a) Derive a reduced 1st order system by Hankel truncation method.

b) Derive a reduced 1st order system by Hankel residualization method.
c) Draw Bode plot of real system and all reduced orders in the same plot.
d) Draw step response of real system and all reduced orders in the same plot.
0 1 0 0   0.8741  1.1929 0.3438   1.1176 
x   0

0 1  x  0  u
  
 x   1.1929  0.8161 1.5679  x   0.5585u
   
 4  5  2 1  0.3438  1.5679  0.3098  0.2510
y  5 2 1x y  1.1176 0.5585  0.2510x
Matlab: system=pck(A,B,C,D); sysbal(system)
x  0.8741x  1.1176u 1.249
Hankel truncation method  g ( s) 
y  1.1176x s  0.8741
ht

x  1.4754x  1.452u  0.1789s  1.844

 g ( s)  54
Hankel residualization method y  1.452 x  0.1789u hr
 1Karimpour
s Ali
Dr. .4754Jan 2017
 Lecture 7
Model Order Reduction of Minimal Representations
Hankel Norm Approximation

Example 7-13: Consider following system.

1.249  0.1789s  1.844
g ( s)  g ( s) 
s  0.8741 s  1.4754
ht hr

55
Dr. Ali Karimpour Jan 2017
 Lecture 7

Exercises

Exercise 7-1: Mentioned in the lecture. Exercise 7-2: Mentioned in the lecture.
Exercise 7-3: Mentioned in the lecture(just for PhD student).
Exercise 7-4: Mentioned in the lecture. Exercise 7-5: Mentioned in the lecture.
Exercise 7-6: Mentioned in the lecture.

Exercise 7-7: Check the contollability and observability of following systems.

Exercise 7-8: Find irreducible

realization for following systems.

56
Dr. Ali Karimpour Jan 2017
 Lecture 7

Exercises

Exercise 7-9: Find a reduced order(2nd order) for following

System:
a) By Hankel truncation method.
b) By Hankel residualization method.
c) Draw Bode plot of real system and all reduced orders in the same plot.
d) Draw step response of real system and all reduced orders in the same plot.

57
Dr. Ali Karimpour Jan 2017
 Lecture 7

References

References

• Multivariable Feedback Design, J M Maciejowski, Wesley,1989.

• Multivariable Feedback Control, S.Skogestad, I. Postlethwaite,

Wiley,2005.
• Control Configuration Selection in Multivariable Plants, A.
Khaki-Sedigh, B. Moaveni, Springer Verlag, 2009.

• ‫ دکتر علی خاکی صديق‬،‫تحليل و طراحی سيستم های چند متغيره‬

Web References

• http://www.um.ac.ir/~karimpor
• http://saba.kntu.ac.ir/eecd/khakisedigh/Courses/mv/

58
Dr. Ali Karimpour Jan 2017