Robust and Optimal Control

A Two-port Framework Approach

CSD approach to
H-infinity Control

1
Robust and Optimal Control - A Two-port Framework Approach

Recall the summary of Method I

P1* = G1M −1
z u u'
u'
 P12
12 11 
P11
w 0 I  w M
 w'
w'

PP1*1*
u uu u u'
u'
 I 0
K  −1

KK y
P
 22
22
P2121 w M w'
w'
y y P2*
2*

G2 =  −1
K = CSDl (, )
 = G2
z = LFTl ( P, K ) w
  = CSDr (G1 , CSDl (G2 , K )) w
= CSDr (G1 , CSDl (, CSDl (,  ))) w
= CSDr (G1 , CSDl (, )) w

with specified properties 2

Robust and Optimal Control - A Two-port Framework Approach

<Recall the J-lossless>

A system G is called J-lossless if
(1) G is J-unitary. ( G ~
J n1 ,k G = J n2 ,k )
(2) G* (s) J n ,k G(s)  J n ,k ,  Re s  0
1 2

Example  s −1   −s − 1   s −1 
 s +1 0   0  0 
−s + 1 1 0   s + 1
G=  G JG = 
~
  
 0 s − 1  0 − s − 1  0 −1 0 s − 1
 s + 1   − s + 1  s + 1
1 0 
=  =J
0 −1

3
Robust and Optimal Control - A Two-port Framework Approach

<Recall the Dual J-lossless>

u z

K y G w

Similarly, a system G is called dual J-lossless

if it satisfy
(1) GJ G ~ = J
k , n1 k , n2

(2) G( s) J k ,n1 G* ( s)  J k ,n2 ,  Re s  0

4
Robust and Optimal Control - A Two-port Framework Approach

z w
y P If P is inner (I-lossless), then
u
LFT ( P, K )  BH for all K  BH
l   

K
z  w  w
From   = P    z * z + y* y =  w* u *  P* P   = w*w + u *u
 y u  u 
Since P* P = I ( P is inner)  z 2 + y 2 = w 2 + u
2 2 2 2
2
(1)

u 2 = Ky 2  K y 2  y 2  u 2 − y 2 0
2 2
For K  BH 
(2)


z ( j )
z 2 − w 2 0  1 for all 
2 2 2
From (1)&(2),
w( j ) 2

z ( j )
LFT ( P, K ) = max 2
 1  LFT ( P, K )  BH
l  w ( j )0 w( j ) 2
l 

5
Robust and Optimal Control - A Two-port Framework Approach

z u
If G is J-lossless, then
G K
w y CSD (G, K )  1,  K  BH
r  

* *
z u   z   z  u  * u  I 0 
From   = G      J   =   G JG   where J =  
 w  y   w   w  y   y 0 −I 
*
u  * u  u 
 z z − w w =   G JG   = u*
* *
y  J   = u*u − y* y
*
(1)
 y  y  y
For K  BH  u 2 = Ky 2  K 
y2 y
 2

z 2 − w 2 = u 2 − y 2 0 (2)
z ( j )
z 2 − w 2 0  1 for all 
2 2 2
From (1)&(2),
w( j ) 2

z ( j )
CSD (G, K ) = max 2
 1  CSD (G, K )  BH
r  w ( j )0 w( j ) 2
r 
6
Robust and Optimal Control - A Two-port Framework Approach

z w
z
y P u
u
G K
w y
K
G is J-lossless P is lossless

CSDr (G, K ) = LFTl ( P, K )

If G is J-lossless, then If P is inner (I-lossless), then
CSD (G, K )  BH ,  K  BH LFT ( P, K )  BH for all K  BH
r   
l   

7
Robust and Optimal Control - A Two-port Framework Approach

A CSD approach is brought up by Kimura, (1991~1996)

z u z u
G K
w y
   −1 
w y
 P12 − P11P21−1P22 P11P21−1 
G= −1
 − P21 P22 P21−1 

G =  K = CSDr ( −1 , )
If Θ is J-lossless CSDr (G, K ) 
= CSDr (, ) 
1    1

u z u z
K y G w   −1 y  
w

 P12 −1 − P12 −1P11 

G= 
 P22 P12
−1
P21 − P22 P12 −1P11  G =  K = CSDl ( −1 , )
If  is dual J-lossless CSDl (G, K )  = CSDl (, )   1    1
Augmentation is needed if P12 or P21 is not invertible 8
Robust and Optimal Control - A Two-port Framework Approach

H∞ Optimal Control
1 1  A B1 B2 
 P11 ( s ) P12 ( s )    P11 P12  s  AP BP  
P( s) =     =  = = C D12 
  C P DP   1
, 0 P D11
 P21 ( s ) P22 ( s ) 
 
 P21 P22   C2 D21 0 

(I) ( A, B2 ) stabilizable and (C2 , A) detectable

(II)  A − j I B2 
rank   = m2 + n , rank (D12 ) = m2
 C 1 D12 

(i.e., the transmission zeros of P12 (s) are not located at the j )

 A − j I B1 
(III) rank   = q2 + n , rank D21 = q2
 C 2 D21 

(i.e., the transmission zeros of P21 (s) are not located at the j )

The H∞ optimization problem requires to find a stabilizing controller K such

that the closed-loop transfer function from w to z satisfies LFTl (P , K ) 
 ,
or equivalently, LFTl ( P , K ) 1 .

9
Robust and Optimal Control - A Two-port Framework Approach

The proposed coupled CSD method: Method I

z u
 G11 G12 
G 
z 1 1  w w  21 G22  w
  P11 
P12 
 
y  P21 P22  u
u u

K  11 12 
  
y  21  22  w

z = LFTl ( P, K ) w
= CSDr (G1 , CSDl (G2 , K )) w
= CSDr (G1 , CSDl (G2 , CSDl (,  ))) w
= CSDr (G1 , CSDl (, )) w

When G1 is J-lossless and  dual J-lossless, one can ensure the stability of the
interconnection system for any   BH  by the small gain theory.
10
 Robust and Optimal Control - A Two-port Framework Approach

G1
Step 1: To find rcf with a special numerator P1*
z u u
x  I 0 0  x   P12
0
P11 
I  M*
   0   u 
w  w w'
Let  u  =  Fu Wuu
 w  Fw Wwu Www   w P2*
u u u
 I 0
 x   A B2 B1   A B2 B1  K y P P21  w M* w'
 22
 z  C D D  C D D 
   1 12 11   1 12 11  G2 =  −1
 w  0 0 I  x  0 0 I  I 0 0  x 
     
u  =  0 I 0   u  =  0 I 0   Fu Wuu 0   u  
 y  C2 0 D21   w C2 0 D21   Fw Wwu Www   w
     
  
u 0 I 0   0 I 0 
 w  0   I 
   0 I  0 0
 A + B1 Fw + B2 Fu B1Wwu + B2Wuu B1Www 
C + D F + D F D W + D W D W 
 1 11 w 12 u 11 wu 12 uu 11 ww 
 Fw Wwu Www   x   G1 
 
= Fu Wuu 0   u   =  G2 
 C2 + D21 Fw D21Wwu D21Www   w  M * 
 
 Fu W uu 0 
 Www 
 Fw Wwu
11
Robust and Optimal Control - A Two-port Framework Approach

  A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 


G1 ( s ) = C1 + D12 Fu + D11 Fw D12Wuu + D11Wwu D11Www 
s


  Fw Wwu Www 

  A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www   A BG2 
 s
  G2
G2 ( s ) =  Fu Wuu 0 = 
  G
C DG2 
 C2 + D21 Fw D21Wwu D21Www   2 

  A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 

M  ( s ) =  0 
s

 Fu Wuu
  Fw Wwu Www 


To find G1  RH  being J-lossless , i.e, G1~ JG1 = J , G1* ( s ) JG1 ( s )  J ,  Re s  0 one has
 −
1

 D12 ( I − D11D11 ) D12 
0    T T −1  2
0 
W
WI =  uu =  where I − D11D11  0
T

 Wwu Www   −
1
− 
1

( I − D11D11 ) D11D12  D12 ( I − D11D11 ) D12  ( 11 11 ) 2 

−1 −1
−
T T T T 2 T
I D D

FI = − RI−1 ( BT X + D1*T C1 ) =  Fu Fw 
T

X = Ric( H X )  0
 A − BRI−1D1*C1 − B RI−1BT 
T

HX =  
T  dom(Ric)
1* I 1* 1 (
 −C T ( I − D R −1DT )C − A − BR −1D T C 
 1 I 1* 1) 
 D12T D12 D12T D11 
RI =  T  , D1* =  D12 D11  , B =  B2 B1  12
 D11D12 −( I − D11D11 ) 
T
Robust and Optimal Control - A Two-port Framework Approach

Step 2: To find lcf with a special numerator

 I H u H y   A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 0 0

  
 0 Wuu 0  Fu Wuu 0 I 0   ( s )  ( s ) 
0 Wyu Wyy   C2 + D21 Fw D21Wwu D21Www 0 I 
 
 A + B2 Fu + B1Fw B2Wuu + BW 
+
1 wu

s + H F + H (C + D F )
BW H y 21 ww 
D W
+ H uWuu + H y D21Wwu 1 ww

( s ) =  
u u y 2 21 w
 Wuu Fu WuuWuu 0 
 
 Wyu Fu + Wyy ( C2 + D21Fw ) WyuWuu + Wyy D21Wwu Wyy D21Www 

is dual J-lossless, it can be verified that

 1

W 0    T T −1

 D12 ( I − D11 D11 ) D12 
2 0 
WI =  uu = 1   R( m2 + q2 )( m2 + q2 )
Wyu Wyy   1 −
 D21 ( I − D11T D11 ) D11T D12
T −2  D ( I − DT D )−1 DT  2 
−1
 −  D21 ( I − D11T D11 ) −1 D21
  21 11 11 21  
(
 H u H y  = ZCGT2 − BG2 J m2 ,m1 DGT2 R −1 )
Z = Ric( H Z )  0

( ) 
T
CG2 RI−1CG2
T
−1
−
T
A B J D R C
H Z =    dom(Ric)
G2 G2 m2 , m1 G2 I G2

 2
BG ( J m ,
2 1m − J m ,
2 1m D
T

G 2
RI−1 DG2 J m2 ,m1 ) BGT2 (
− AG2 − BG2 J m2 ,m1 DG2 RI−1CG2
T

) 

RI = DG2 J m2 ,m1 DG2 T 13
Robust and Optimal Control - A Two-port Framework Approach

Step 3: K  = CSDl (,  ) = LFTl ( P ,  ),   BH 

 A + B1Fw + B2 Fu + H u Fu + H y ( C2 + D21Fw ) H u H y 
s
 
 (s) =  Wuu Fu Wuu 0   GH 

 Wyu Fu + Wyy ( C2 + D21Fw ) Wyu Wyy 

 A + B1Fw + B2 Fu + H y ( C2 + D21Fw ) H y H uWuu−1 

s
 
 P (s) =  − Fu 0 Wuu−1 

 W yy ( C 2 + D21 Fw ) W yy W yuW −1 
uu 
G1
z u
 G11 G12 
w G 
 21 G22  w '

K G2

u u u u
−1
 11 12   11 12   11 12 
 y   y   y  
 21  22   21  22   21  22  w'

14
Robust and Optimal Control - A Two-port Framework Approach

Equivalent Framework
z u
 G11 G12 
G 
z 1 1  w w  21 G22  w
  P11 
P12 
 
y  P21 P22  u
u u

K  11 12 
  
y  21  22  w

The controller K  = CSDl (,  ) = LFTl ( P ,  ),   BH 

u y
u u
y  P ( s)
u
(s)  (s) y
y
(s)

 A + B1Fw + B2 Fu + H y ( C2 + D21Fw ) H y H uWuu−1 

s  A + B F + B F + H (C + D F ) Hy
s
 
 P ( s) =  − Fu Wuu−1   K 0 ( s ) = 
1 w 2 u y 2 21 w
0 

 − Fu 0 

 Wyy ( C2 + D21Fw ) Wyy WyuWuu 
−1

15
Robust and Optimal Control - A Two-port Framework Approach

z u
Summary of Method I  G11 G12 
G 
w  21 G22  w

u u
 11 12 
  
y  21  22  w

 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 

G1 ( s ) = C1 + D12 Fu + D11 Fw D12Wuu + D11Wwu D11Www 
s

 Fw Wwu Www 
 A + B2 Fu + B1 Fw B2Wuu + B1Wwu 
 B1Www + H y D21Www 
s + H F + H (C + D F ) + H uWuu + H y D21Wwu
( s ) =  
u u y 2 21 w
 Wuu Fu WuuWuu 0 
 
 yu u
W F + W yy ( C 2 + D F
21 w ) W W
yu uu + W D W
yy 21 wu W D W
yy 21 ww 
 A + B1 Fw + B2 Fu + H u Fu + H y ( C2 + D21Fw ) H u H y 
s
 
 (s) =  Wuu Fu Wuu 0 

 Wyu Fu + Wyy ( C2 + D21 Fw ) Wyu Wyy 
K  = CSDl (,  ) = LFTl ( P ,  ),   BH 
16
 Robust and Optimal Control - A Two-port Framework Approach

Summary of Method I (continued)

 −
1

0   D ( I − D D ) D  2
T T −1
0 
W  12 11 11 12 
WI =  uu =   where I − D11 D11  0
T

 Wwu Www   −
1
− 
1

( I − D11 D11 ) D11 D12  D12 ( I − D11 D11 ) D12  ( I − D11D11 ) 2 
T −1 T −1 2 T
T T

F 
FI =  u  = − RI−1 ( BT X + D1*T C1 )
 Fw 
(I) ( A, B2 ) stabilizable and (C2 , A)detectable
X = Ric( H X )  0
 A − BRI−1 D1*C1
T
− B RI−1BT   A − j I B2 
HX =    dom(Ric) (II) rank  C  = m2 + n , rank (D12 ) = m2
 12 
( )
D
 −C T ( I − D R −1 DT )C − A − BR −1 D T C 
T 1

 1 1* I 1* 1 I 1* 1
  A − j I B1 
 D12 D12
T
D12 D11 
T (III) rank   = q2 + n , rank D21 = q2
 , D1* =  D12 D11  , B =  B2 B1   21 
RI =  T C 2 D
 D11 D12 −( I − D11 D11 ) 
T

 1

Wuu 0    D12 ( I − D11 D11 ) D12  2
T T −1
0 
WI =  = 1 1   R( m2 + q2 )( m2 + q2 )
−
Wyu Wyy   −  D ( I − DT D ) −1 DT  − 2 D ( I − DT D )−1 DT D  ( ) T  2
−1
  21 11 11 21  21 11 11 11 12 D
 21 I − D T
D
11 11
D 21  
(
 H u H y  = ZCGT2 − BG2 J m2 ,m1 DGT2 R −1 )
Z = Ric( H Z )  0

( ) 
T
CG2 RI−1CG2
T
−1
−
T
A B J D R C
H Z =    dom(Ric)
G2 G2 m2 , m1 G2 I G2

−1
 BG2 ( J m2 ,m1 − J m2 ,m1 DG2 RI DG2 J m2 ,m1 ) BG2
T
T
(
− AG2 − BG2 J m2 ,m1 DG2 RI−1CG2
T

) 

17
RI = DG2 J m2 ,m1 DG2 T
Robust and Optimal Control - A Two-port Framework Approach

Example I
Given a block diagram as shown below, please use CSD approach to find the 𝐻∞ controller
K(s) such that 𝐿𝐹𝑇𝑙 (𝑃, 𝐾) ∞ < 𝛾 where B=10 , 𝐾𝑡 = 2 , J=1 , 𝛾 ≥ 1

18
Robust and Optimal Control - A Two-port Framework Approach

Example I
−10 −1 0 2
s 𝐴 𝐵1 𝐵2
−0.6 0 0.6 0
Cut down the loop and P matrix 𝑃 𝑠 = 𝐶1 𝐷11 𝐷12 =
0 0 0 1
𝐶2 𝐷21 0
−1 0 1 0

𝐴 = −10
𝐵 = 𝐵2 𝐵1 = 2 −1 0

𝐶1 −0.6
𝐶= = 0
𝐶2
−1
x 𝑇𝑑 𝜔∗ u 𝐷11 𝐷12 0 0.6 0
𝐷= = 0 0 1
 B 1 Kt  𝐷21 0
𝑥ሶ  − J −
J
0
J 
0 1 0
 
A B1 B2  𝑧  −We 0
We
0 
P( s ) =  C1 D12  =   
s 𝑒
D11  Let 𝑊𝑒 = 0.6, 𝑊𝑢 = 1, 𝛾 = 1
 
C2 D21 0  𝑧  0 0 0
Wu 
𝑢
 
y  −1 
 0 1 0  19
Robust and Optimal Control - A Two-port Framework Approach

Example I

Check Controllability and Observability

Calculate by MATLAB:

(1)rank(crtb(A,B2 ))=2 →is stabilizable

(2)rank(obsv(C2 ,A))=2 → is detectable

𝐴 − 𝑠𝐼 𝐵2
(3)rank( )=1+1=2, rank 𝐷12 =1
𝐶1 𝐷12

𝐴 − 𝑠𝐼 𝐵1
(3) rank( )=1+1=2, rank 𝐷21 =1
𝐶2 𝐷21

The system is controllable and observable!

20
Robust and Optimal Control - A Two-port Framework Approach

Example I

Solve solution

 −
1
 1 0 0 
 D12 ( I − D11 D11 ) D12 
  T T −1  2
0  
WI =   Wuu 0 
 = 0 1 0  =   where I − D11T D11  0

( I − D11T D11 ) 2  0 0 1.25 
−
1 1 Wwu Www 
( I − D11 D11 ) D11 D12  D12 ( I − D11 D11 ) D12 
−1 −
T T
 T T −1
 2

1 0 0 
 D12T D12 D12T D11  
RI =  T  = 0 −1 0 
 D11 D12 −( I − D11 D11 )  0 0 −0.64 
T

 
D1* =  D12 D11  , B =  B2 B1 
 A − BRI −1 D1*T C1 − BRI −1 BT   −10 −3
Hx =  T  =  
 −C1 ( I − D1* RI D1* )C1 −( A − BRI D1*C1 )   −0.5625 10 
−1 T −1 T T

 −10 −3
X = Ric  H x  = Ric   = 0.028 ( eigenvalue of H x :  10.084 ; trace( H x ) = 0 )
 − 0. 5625 10 
 −0.056 
F 
FI = − RI−1 ( BT X + D1*T C1 ) =  −0.028  =  u 
 −0.5625  w 
F

21
Robust and Optimal Control - A Two-port Framework Approach

Example I

Solve solution
 −10.084 2 −1 0 
 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www   −0.9375 0 0 0.75

G1 ( s ) = C1 + D12 Fu + D11 Fw 
s
D12Wuu + D11Wwu D11Www  =  −0.056 1 0 0 
 
 Fw Wwu Www   −0.028 0 1 0 
 −0.5625 0 0 1.25 
 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 
 AG2 BG2   −10.084 2 −1 0 
G2 ( s ) =    =  −0.056 1 0 0 
s
Fu Wuu 0 =
 C2 + D21 Fw  CG2 DG2 
  −1.5625 0 0 1.25
D21Wwu D21Www 

22
Robust and Optimal Control - A Two-port Framework Approach

Example I

Solve solution

 1

Wuu 0   
 12
D T
( I − D D
11 11
T −1
) D12 
 2 0  1 0 
WI =  = 1 1  = 0 0.8
−
 yu
W W yy 
  −  D ( I − DT D ) −1 DT  2 D21 ( I − DT D ) D11T D12  D ( I − DT D )−1 DT  2  
− −1 
  21 11 11 21  11 11  21 11 11 21  
 ( AG2 − BG2 J1,2 DG2 T RI −1CG2 )T CG2 T RI −1CG2 
HZ =  −1 −1

B ( J − J D T
 G2 1,2 1,2 G2 I G2 1,2 G2
R D J ) B T
− ( A G2 − B J
G2 1,2 G2D T
R I CG2 
) 
 −9.972 −1.5594 
Z = Ric  H z  = Ric   =0.0499 ( eigenvalue:  10.0499 ; trace(H z )=0 )
 − 1 9.972 
1 0 
RI = DG2 J1,2 DG2 T =  
0 −1.5625
H I = ( ZCGT2 − BG2 J1,2 DGT2 ) RI −1 =  H u H y  =  −2.0028 0.0499

23
Robust and Optimal Control - A Two-port Framework Approach

Example I Solve solution

 A + B2 Fu + B1Fw B2Wuu + BW 
1 ww + H y D21Www 
1 wu
 BW  −10.0499 −0.028 −1 0.0624 
s + H F + H (C + D F ) + H uWuu + H y D21Wwu
( s ) = 
u u y 2 21 w  =  −0.056 1 0 0 
 Wuu Fu WuuWuu 0  
   −1.25 0 0 1 
 Wyu Fu + Wyy ( C2 + D21Fw ) WyuWuu + Wyy D21Wwu Wyy D21Www 
 A + B2 Fu + B1 Fw 
 H u H y   −10.0499 −2.0028 0.0499 
s + H F + H (C + D F )
 ( s) = 
u u y 2 21 w  =  −0.056 1 0 
 Wuu Fu Wuu 0  
   −1.25 0 0.8 
 yu u
W F + W yy ( C 2 + D F
21 w ) W yu W yy 

 A + B2 Fu + B1 Fw 
 Hu Hy 
 11 ( s ) 12 ( s)  =  + H u Fu + H y ( C2 + D21 Fw )
s


 Wuu Fu Wuu 0 

 A + B2 Fu + B1 Fw   A + B2 Fu + B1Fw 
 I − H uWuu   Hu Hy 0  0 Hy − H uWuu 
 −1  
+ H u Fu + H y ( C2 + D21 Fw )  =  + H y ( C2 + D21Fw ) 
0 Wuu  
 Wuu Fu Wuu 0 I   Fu I 0 −1 
Wuu 

K CSDl (Π, Φ) CSDl (Π, 0) H u Fu 在求 Π11 1Π12過程被消掉

With Central Controller (Φ 0)
A B2 Fu B1 Fw H y C2 D21 Fw Hy
K0 CSDl (Π, 0) Π11 1Π12
Fu 0
0.0028
K0 24
s 10.1621
Robust and Optimal Control - A Two-port Framework Approach

Example I

Solve solution  −1.875 0.9375 3

 s + 10.084 s + 10.084 4
 −10.084 2 −1 0   
 −0.9375   s + 9.9722 0
0 0 0.75 0.056
s   s + 10.084 
s + 10.084
G1 ( s ) =  −0.056 1 0 0   G1 ( s ) =  
   −0.056 s + 10.1145
0
 −0.028 0 1 0 
 s + 10.084 s + 10.084 
 −0.5625 0 1.25   1.125
0 0.5625 5
 
 s + 10.084 s + 10.084 4
The eigenvalues of H x with negative real - part are the pole of G1 s
(eig ( H x ) 10.084, pole of G1 ( s ) 10.084)

 −10.0499 −0.028 −1 0.0624   s + 10.05 0.056 −0.0035 

 s + 10.0499 s + 10.0499 s + 10.0499 
( s ) =  −0.056 0   ( s) = 
s
1 0 
 0.0035 1.25 s + 9.9722 
 −1.25 0 0 1   s + 10.0499 s + 10.0499 s + 10.0499 

The eigenvalues of H z with negative real - part are the pole of Θ( s )

(eig ( H z ) 10.0499, pole of Θ( s ) 10.0499)
25
Robust and Optimal Control - A Two-port Framework Approach

Example I

Solve solution

K CSDl (Π, Φ) LFTl (Π P , Φ)

1
A B1 Fw B2 Fu H y (C2 D21Fw ) H y H uWuu
s
ΠP Fu 0 Wuu
1
Wyy (C2 D21 Fw ) Wyy WyuWuu
Central Controller (Φ 0)
s A B1 Fw B2 Fu H y (C2 D21Fw ) H y 10.1621 0.0499
K 0 (s)
Fu 0 0.056 0
0.0028
K 0 (s)
s 10.1621
1 1 0.0028
K (s) Kt 2
Js B s 10 s 10.1621 0.0056
因此閉迴路轉移函數為： 1 1 0.0028
1 K ( s) Kt 1 2 ( s 10.1122)( s 10.0499)
Js B s 10 s 10.1621 26
Robust and Optimal Control - A Two-port Framework Approach

Example I LFT diagram

P( s)

K (s)

eigenvalues 10.1122, 10.0499

(by Matlab eig function)
 A B2 Fu B1 
s  
LFTl ( P( s ), K ( s )) =  − H y C2 A + B1Fw + B2 Fu + H y (C2 + D21Fw ) − H y D21 
 C D12 Fu D11 
 1
eig ( H z ) 10.0499
  A B1 B2 
 s
 
 P( s ) =  C1 D11 D12 
 C2 D21 0 
where 
 s  A + B F + B F + H (C + D F ) −H y 
 K (s) =  0.0028

1 w 2 u

Fu
y 2 21 w
 =
0  s + 10.1621
與前述結果相呼應 27

Robust and Optimal Control - A Two-port Framework Approach

Example I

MATLAB CODE

28
Robust and Optimal Control - A Two-port Framework Approach

A unified approach to control synthesis

Objective: Requirements on coupled

Find a such that CSD matrices (Method I)
G G12   I 12 
G1 =  11   =    RH 
I 
LFT ( P, K )is stable RH  ,
Stabilization problem 0  0  22 

G G   I 12 
min LFT ( P, K ) (i) G1 =  11 12   RH  ,  =    RH 
0 22 
2
H2 control problem 0 I 

(ii) G11 inner &  22 co-inner

LFT ( P , K ) 
1 (i) G1 J-lossless
H∞ control problem
(ii)  dual J-lossless

29
Robust and Optimal Control - A Two-port Framework Approach

H-infinity Solution of
Special SCC
Formulations

30
Robust and Optimal Control - A Two-port Framework Approach

Formulation of Formulation of
Corresponding Plant Corresponding Plant
System System

A B1 B2   A B1 B2 
PDF ( s) =  C1 D12 
Disturbance
POE ( s) =  C1 D11 I 
s s
D11 Output Estimation
Feedforward C2 I 0  C2 D21 0 
(OE)
(DF) with (A − B1C2 ) is Hurwitz with (A − B2C1 ) is Hurwitz

 A B1 B2 
 C1 D11 D12 
A B1 I 0 
Full Information
PFI ( s) = 
Full Control  
s s

  I  0 0  PFC ( s) =  C1 D11 0 I 

(FI)       
 0  I  0 
(FC) C2 D21 0 0

A B1 B2  A B1 I 
POI ( s) =  C1 D11 0
State Feedback Output Injection
PSF ( s ) = C1 D12 
s s
D11
(SF)  I 0 0  (OI) C2 D21 0

31
Robust and Optimal Control - A Two-port Framework Approach

 A − j I B2 
Disturbance Feedforward (DF) problem rank  = m2 + n , rank (D12 ) = m2
 C1 D12 

A B1 B2  z u
 P11 ( s ) P12 ( s )  
D12  with (A − B1C2 ) is Hurwitz  P12 P11 
s
PDF ( s ) =   =  C1 D11 0
 21
P ( s ) P ( s )  C w  I  w
22
 2 I 0 
P1*
One adopts CSDr CSDl u u
 I 0
K y P P21  w
A B1 B2   22
C D11 D12  P2*
 P1*, DF ( s )  s  1
P = 0 0 I 
 2*, DF   0
( s )
I 0 

 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 

G1, DF ( s ) = C1 + D12 Fu + D11 Fw D12Wuu + D11Wwu D11Www   RH 
s
C2 0 I 
 Fw Wwu Www 
 P1*, DF   G1, DF  −1
P*, DF =  = G  M DF  A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 
 P   2, DF 
G2, DF ( s ) =  0   RH 
2*, DF s
Fu Wuu
 C2 + Fw Wwu Www 

So far the procedure is identical to the general approach !

32
Robust and Optimal Control - A Two-port Framework Approach

lcf:
I − B2 − B1   A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 0 0 
0 Wuu −1 0    Fu Wuu 0 I 0 

0 −Www −1WwuWuu −1 Www −1   C2 + Fw Wwu Www 0 I 
 A − B1C2 0 0 − B2 − B1 
  DF ( s )  DF ( s )  =  0 
s
Wuu −1 Fu I 0 Wuu −1
Www −1 ((C2 + Fw ) − WwuWuu −1 Fu ) 0 I −Www −1WwuWuu −1 Www −1 

 A − B1C2 − B2 − B1 
 DF ( s ) =  0   GH 
s
Wuu −1 Fu1 Wuu −1
Www −1 ((C2 + Fw ) − WwuWuu −1 Fu ) −Www −1WwuWuu −1 Www −1 

 I 0 −1
where  DF = 0 I  and  DF = G 2, DF . Since (A-B1C2) is Hurwitz,  DF derived
 
for this DF problem is actually an identity matrix. Then the H-infinity controllers can

be generated by K = CSDl (G −12, DF ,  ) = CSDr (G2, DF ,  ),   RH 

z = LFT ( PDF , K DF ) w = CSDr (G1 , CSDl (,  DF )) w
= CSDr (G1 , CSDl ( I , 0)) w = CSDr (G1, DF , 0) w = G12G22 −1w as  DF = 0
33
Robust and Optimal Control - A Two-port Framework Approach

 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 

 G11 ( s ) G12 ( s )  s 
G1, DF ( s ) =   = C1 + D12 Fu + D11 Fw D12Wuu + D11Wwu D11Www 
 21
G ( s ) G22 ( s )   Www 
 Fw Wwu
 A + B2 Fu + B1 Fw B1Www 
 G12 ( s )  s 
  = C1 + D12 Fu + D11 Fw D11Www 
G22 ( s )   Www 
 Fw

 A + B2 Fu + B1 Fw B1Www   A + B2 Fu B1 
C + D F + D F  D11 
D11Www   I 0  C1 + D12 Fu
 1 12 u 11 w
 =
 Fw Www   −Www −1 Fw Www −1   0 I 
   −1 −1 
 0 I   −Www Fw Www 

−1
 A + B2 Fu
s B1 
 G12G22 ( s ) =  
 C1 + D12 Fu D 11 
 z = CSDr (G1 , CSDl ( I , 0)) w = CSDr (G1, DF , 0) w = G12G22 −1w as  DF = 0

Coprime factorization

34
Robust and Optimal Control - A Two-port Framework Approach

 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 

 G11 ( s ) G12 ( s )  s 
G2, DF ( s ) =  = Fu Wuu 0 
 21
G ( s ) G 22 ( s )   C2 + Fw Www 
 Wwu
 A + B2 Fu + B1 Fw B1Www 
 G12 ( s )  s 
  = Fu 0 
G22 ( s )   C2 + Fw Www 


 A + B2 Fu + B1 Fw B1Www   A + B2 Fu − B1C2 B1 
  0 
Fu 0   I 0   Fu
  =
 C2 + Fw Www   −Www −1 (C2 + Fw ) Www −1   0 I 
   −1 −1 
 0 I   −Www (C2 + Fw ) Www 

s  A+ B F − BC B1 
 G22 −1 ( s ) = 
2 u 1 2
−1 −1 
 −Www (C 2 + Fw ) Www 
 A + B2 Fu − B1C2
−1
s B1 
 K ( s ) = CSDr (G2, DF , 0) = G12 ( s )G22 ( s ) = 
 Fu 0 

K = CSDl (G −12, DF ,  ) Coprime factorization

= CSDr (G2, DF , ),
  RH  35
Robust and Optimal Control - A Two-port Framework Approach

 A − B1C2 − B2 − B1 
G2, DF ( s ) −1 =  DF ( s ) =  0 
s
Wuu −1 Fu Wuu −1
Www −1 ((C2 + Fw ) − WwuWuu −1 Fu ) −WwuWuu −1 Www −1 

 I B2Wuu   A − B1C2 − B2 − B1  s A+ B F − B C 0 − B1 

 =
2 1 2
      
0 W  W F W 0   11  0 
−1 −1 ( s ) 12 ( s )
 uu   uu u uu  Fu I

−1 −1
s A + B2 Fu − B1C2 B1 
K ( s ) = CSDl (G , 0) = CSDl ( DF ( s), 0) = − 12 = 
2 , DF 11
 Fu 0 

−1
s  A + B2 Fu − B1C2 B1 
K ( s ) = CSDr (G2, DF , 0) = G12 ( s)G22 ( s) = 
 Fu 0 

Note that K = CSDl (G −12, DF ,  ) = CSDr (G2, DF , ),   BH 

36
Robust and Optimal Control - A Two-port Framework Approach

P( s)

K (s)

I  A I B2 Fu B1   A + B2 Fu B2 Fu B1 
s  s   s  A + B2 Fu B1  
LFTl ( P ( s ), K ( s )) =  B1C2 A + B2 Fu − B1C2 B1  =  A − B1C2 = 
D11  
0 0 
 C1 + D12 Fu
 C D12 Fu D11  C1 + D12 Fu D12 Fu D11    
 1
  A B1 B2 
 s
  s A B2 Fu B1
 P( s ) =  C1 D11 D12  LFTl ( P( s ), K ( s ))
 C2 I 0  C1 D12 Fu D11
where 
 s A+ B F − B C B1 
 K (s) =  2 u 1 2

  Fu 0  A矩陣的Eigenvalue為A B2 Fu , A B1C2
37
Robust and Optimal Control - A Two-port Framework Approach

Full Information (FI) problem

 A B1 B2 
 
 P11 ( s) P12 ( s)  s  C1 D11 D12 
PFI ( s) =  =
 P21 ( s) P22 ( s )   I   0   0  
      
 0   I   0  

One adopts CSDr CSDl

 A B2 B1 
C   A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 
D D
  G1, FI ( s ) = C1 + D12 Fu + D11 Fw D12Wuu + D11Wwu D11Www   RH 
1 12 11
s

 P1*, FI ( s )  s  0 0 I 
 P ( s)   =   Fw Wwu Www 
 2*, FI   0 I 0 
 I   0   0    A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www 
        
 0   0   I   F W 0
G2, FI ( s ) =    RH
s
u uu

  I   0   0 


 P1*, FI   G1, FI  −1       
P*, FI =  = G  M FI   w
F  wu 
W  ww  
W
P
 2*, FI   2, FI 

The procedure is also identical to the general approach !

38
Robust and Optimal Control - A Two-port Framework Approach

lcf:
 I − B2 ( B1Fw + B2 Fu − B1 )   A + B2 Fu + B1Fw B2Wuu + BW BW 0 (0 0) 
   1 wu 1 ww

 0 Wuu−1 ( uu u )   
− W −1
F 0 Fu Wuu 0 I (0 0) 
    I   0   0   0   I 0 
 0   0  I 0             
  
 0   −WwwWwuWuu 
− −  −1     Fw   Wwu   Www   0   0 I  
 
1 1
  0 W ww 

 A + B2 Fu + B1Fw 0 0 − B2 B1Fw + B2 Fu − B1 
s
 0 I 0 Wuu−1 −Wuu−1Fu 0 
  FI ( s )  FI ( s )  = 
 I 0 0 0 I 0 
 −1 −1 −1 −1 −1 
 −WwwWwuWuu Fu1 + Www Fw 0 I −Www WwuWuu−1 0 Www 

 A + B2 Fu + B1Fw 0 0 
  
0 
I 0
 s 0 I
 FI ( s ) =  0   0   =
   0   0 
 0   I   
I
−1 −1 −1     
−
 ww wu uu u
W W W F + Www w 
F  0   I  
and the central controller is given by
x 
u = K FI yFI = CSDl (  FI , 0 ) yFI = − 11
−1
12 yFI =  Fu 0   = Fu x
 w

39
Robust and Optimal Control - A Two-port Framework Approach
z u
 P12 P11 
State Feedback (SF) problem w 0
 I  w

A B1 B2  P1*
 P11 ( s ) P12 ( s )  s  One adopts CSDr CSDl and u
D12 
u
PSF =   = C1 D11  I 0
 P21 ( s ) P22 ( s )   I 0  the state space of lcf is given by K y P
 22 P21  w
 0
P2*

 A + B2 Fu + B1 Fw B2Wuu + B1Wwu B1Www   A + B2 Fu + B1Fw B2Wuu + B1Wwu B1Www 

G1, SF ( s ) = C1 + D12 Fu + D11 Fw D11Www  , G2, SF (s ) =  0 
s s
D12Wuu + D11Wwu Fu Wuu
 Fw Wwu Www    I 0 0 

 I −( B2 + B1WwuWuu−1 ) ( B2 + B1WwuWuu−1 ) Fu   A + B2 Fu + B1Fw B2Wuu + B1Wwu B1Www 0 0

  
0 Wuu−1 −Wuu−1 Fu  Fu Wuu 0 I 0 
0 0 I   I 0 0 0 I 
 
 A + B2 Fu + B1 Fw 0 B1Www −( B2 + B1WwuWuu−1 ) ( B2 + B1WwuWuu−1 ) Fu 
s
 
  SF ( s )  SF ( s )  =  0 I 0 Wu−u1 −Wu−u1 Fu 
 I 0 0 0 I 
 

I 0 Wu−u1 −Wu−u1Fu 
 SF ( s )  SF ( s )  =   and the central controller is given by
0  22 ( s)  21 ( s )  22 ( s ) 
CSDl ( SF ( s ), 0) = Fu , i.e., u = K y = Fu x(t ), since C2 = I
therefore z = CSDr (G1 , CSDl (, 0)) w = CSDr (G1 , 0) w = G12G22 −1w 40
Robust and Optimal Control - A Two-port Framework Approach
z u
 G11 G12 
G 
w  21 G22  w

State Feedback (SF) problem

u
 A + B2 Fu + B1Fw
u
1 ww 
BW  11 12 
 G12 ( s )  s 
D11Www 
  
G ( s )  = C1 + D12 Fu + D11Fw  21  22 
y w

 22   Www 
 Fw
 x   A + B2 Fu + B1Fw 1 ww 
BW  A + B2 Fu B1 
 z  C + D F + D F  D11 
D11Www   I 0   x  C1 + D12 Fu
 = 1 12 u 11
 =
w  Fw Www   −Www −1Fw Www −1   w  0 I 
     −1 −1 
 w '  0 I   −Www Fw Www 

−1
s A + B2 Fu B1 
 G22 ( s ) =  −1 −1  z CSDr (G1 , 0) w T ( s ) w, u ' 0* w'
− W
 ww w F W ww 

s  A+ B F B1 
 T ( s ) = G12 ( s )G22 −1 ( s ) = 
2 u

C + D F D 
 1 12 u 11 
Coprime factorization

因此可求得閉迴路系統T的poles會等於eig ( A B2 Fu )

41
Robust and Optimal Control - A Two-port Framework Approach

P( s)

C2 I

K
 A B1 B2 
  D12 
s
s  A + B2 Fu B1   P( s ) = C1 D11
LFTl ( P( s ), K ) =  , where 
C1 + D12 Fu D11    I 0 0 
 K = Fu


可得出閉迴路系統poles為A B2 Fu (意即A B2 K )

42
Robust and Optimal Control - A Two-port Framework Approach

State-Space Formulae of H∞ Controllers (Method II)

z 1  w
A B1 B2   x   A B1 B2   x    P11 P12 
 P11 ( s ) P12 ( s )  s   
D12    z  =  C1 D12   w
y 1  u
P( s) =   =  C1 D11 D11
      P21 P22 
 21
P ( s ) P22 ( s )  C 0   y  C2 0   u   
 2 D21 D21

( A, B2 ) stabilizable and (C2 , A) detectable. K

First the stacked matrix is arranged by

A 0 − B1 B2 0
P* ( s ) =  P*1 ( s ) P*2 ( s )  =  C1 0  .
s
I − D11 D12
P*1
z z C2 0 − D21 0 − I 
 I − P11 
y 0 − P  w
 21 

P*2
z u
 P12 0
y P − I  y K
 22

43
 Robust and Optimal Control - A Two-port Framework Approach

Step 1: To find lcf with an upper block triangular CSD

Let
 x  I H z H y   x 
    
z
  = 0 Wzz W yz   ze 
 y  0 0 Wyy   ye 
   

 P*1 P*2  = M *−1 G1 G2 

I H z H y   A 0 − B1 B2 0 0 0
 
G1 ( s ) G2 ( s ) M * ( s )  = 0 Wzz Wyz   C1 I 0 
s

 I − D11 D12 0
0 0 Wyy  C2 0 − D21 0 −I 0 I 
 

where H =  H z H y  is chosen such that A + H z C1 + H y C2 is Hurwitz and Wzz and Wyy are nonsingular,

44
Robust and Optimal Control - A Two-port Framework Approach

 I H z H y   A 0 − B1 B2 0 0 0
 
0   C1 I − D11 I 0 
s
G1 ( s ) G2 ( s ) M * ( s )  = 0 Wzz D12 0
0 Wyz Wyy  C2 0 − D21 0 −I 0 I 
 
 A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 
s
 
G1 ( s ) =  Wzz C1 Wzz −Wzz D11 
 Wyz C1 + Wyy C2 Wyz −Wyz D11 − Wyy D21 

 A + H z C1 + H y C2 B2 + H z D12 −H y 
s
   AG2 BG2 
G2 ( s ) =  Wzz C1 Wzz D12 0  :=  
CG DG2 
 Wyz C1 + Wyy C2
 Wyz D12 −Wyy   2 

In order to make G1  RH is dual J-lossless, one has

 −
1

W 0   ( I − D11 D11 ) 2
T
0 
WII =  zz =
Wyz Wyy   −1 T − 2
1
T −2
1
  D21 ( I − D11 D11 ) D21  D21 D11T ( I − D11 D11T ) −1
T
 D21 ( I − D11T D11 ) −1 D21  
H II =  H z H y  = −(YC T + B1 D*1T ) RII−1 ,

( ) 
T
−1
− −C T RII−1C
T

 A B D R
1 *1 II C 
HY =   = 解得Y
( )
T
T −1 −1
 − B1 ( I − D*1 RII D*1 ) B1 − A − B1 D*1 RII C 
T
T
 
 −( I − D11 D11T ) D11 D21T  D  C 
RII =  T T 
, D*1 =  11  , C =  1 
 D21 D11 D21 D21   D21   C2  45
Robust and Optimal Control - A Two-port Framework Approach

Step 2: To find rcf with an upper block triangular CSD

G2 (s)  A + H z C1 + H y C2 B2 + H z D12 −H y 
 Wzz C1 Wzz D12 0   I 0 0 
    ( s ) 
 Wyz C1 + Wyy C2 Wyz D12 −Wyy    Fu Wuu 0  
    ( s )
 0 I 0   Fy Wyu Wyy   
 0 0 I 
To make  J-lossless, one has
 A + H z C1 + H y C2 ( B2 + H z D12 )Wuu 
 − H yWyy 
 11 ( s ) 12 ( s )  s  +( B2 + H z D12 ) Fu − H y Fy − H yWyu 
 ( s)  ( s)   =
 21   Wzz C1 + Wzz D12 Fu Wzz D12Wuu 0 
22

Wyz C1 + Wyy C2 + Wyz D12 Fu − Wyy Fy Wyz D12Wuu − WyyWyu −WyyWyy 

 −
1

Wuu 0    D12 ( I − D11 D11 ) D12  2
T T −1
0 
WII =  =
  1 1
Wyu Wyy   D DT ( I − D DT ) −1 D  DT ( I − D DT ) −1 D  − 2  D21 ( I − D11T D11 ) −1 D21
T 2
 
 21 11 11 11 12  12 11 11 12 

 Fu 
(
RII = DG2 T J q1 ,q2 DG2 , FII =   = − RII−1 BGT2V + DGT2 J q1 ,q2 CG2 )
 Fy 
 (
AG2 − BG2 RII−1 DG2 J q1 ,q2 CG2
T
) − BG2 RII−1 BG2
T

HV =  
( )
T
−1 T
 −CG2 ( J q1 ,q2 − J q1 ,q2 DG2 RII DG2 J q1 ,q2 )CG2 − AG2 − BG2 RII−1DG2 J q1 ,q2 CG2
T
T 46

Robust and Optimal Control - A Two-port Framework Approach

Step 3: K = CSDr (, ) = LFTl ( P , ),   BH 

 A + H z C1 + H y C2 
 ( B2 + H z D12 )Wuu 
 +( B2 + H z D12 ) Fu − H yWyy 
 11 ( s ) 12 ( s )  
S − H yWyu

 ( s )  ( s )  =  − H y Fy 
 21 22 
 Fu Wuu 0 
 Fy Wyu Wyy 

 A + H z C1 + H y C2 + ( B2 + H z D12 ) Fu −H y ( B2 + H z D12 )Wuu 
s
 
 P (s) =  Fu 0 Wuu 
 −Wyy−1 Fy Wyy−1 −1
−Wyy Wyu 

47
Robust and Optimal Control - A Two-port Framework Approach

Equivalent Framework
G1
z z
z 1  w  G11 G12 
  P11 P12 
y  
y   G21 G22  w
1  u
  P21 P22 
 
z u
 11 12 
  y 
 21  22 
K y

The controller K  = CSDr (, ) = ( 11 + 12 )(  21 +  22 ) ,   BH 

−1

u y
u u  P (s)
y u
 (s) (s)
y y
(s)
 A + H z C1 + H y C2 + ( B2 + H z D12 ) Fu − H y ( B2 + H z D12 )Wuu 

s

 P (s) =  Fu 0 Wuu 
 −1
−Wyy Fy −1
Wyy −1
−Wyy Wyu 
s  A + H C + H C + (B + H D )F −H y 
 K 0 ( s) = 
z 1 y2 2 2 z 12 u

F 0 
 u  48
Robust and Optimal Control - A Two-port Framework Approach

A unified approach to control synthesis

Objective: Constraints on coupled CSD

Find a such that matrices (Method II)
 I G12  11 12 
G1 =    RH  ,  =    RH 
Stabilization problem LFT ( P, K ) is stable  0 G22   0 I 

 I G12  11 12 

(i) G1 =    RH  ,  =   RH 
min LFT ( P, K ) 0 G22   0 I 
H2 control problem 2

(ii) G22 co-inner & 11 inner

H∞ control problem LFT ( P , K ) 

1 (i) G1 Dual J-lossless
(ii)  J-lossless

49
Robust and Optimal Control - A Two-port Framework Approach

H-infinity Solution of
Special SCC
Formulations

50
Robust and Optimal Control - A Two-port Framework Approach

Formulation of Formulation of
Corresponding Plant Corresponding Plant
System System

A B1 B2   A B1 B2 
PDF ( s) =  C1 D12 
Disturbance
POE ( s) =  C1 D11 I 
s s
D11 Output Estimation
Feedforward C2 I 0  C2 D21 0 
(OE)
(DF) with (A − B1C2 ) is Hurwitz with (A − B2C1 ) is Hurwitz

 A B1 B2 
 C1 D11 D12 
A B1 I 0 
Full Information
PFI ( s) = 
Full Control  
s s

  I  0 0  PFC ( s) =  C1 D11 0 I 

(FI)       
 0  I  0 
(FC) C2 D21 0 0

A B1 B2  A B1 I 
POI ( s) =  C1 D11 0
State Feedback Output Injection
PSF ( s ) = C1 D12 
s s
D11
(SF)  I 0 0  (OI) C2 D21 0

51
Robust and Optimal Control - A Two-port Framework Approach

Output Estimation (OE) problem (a dual part to DF)

A B1 B2 
  
I  with (A − B2C1 ) is Hurwitz
P ( s ) P ( s ) s
POE ( s ) =  11 12
 =  C1 D11
 P21 ( s ) P22 ( s )  C 0 
 2 D21
z' z
A 0 − B1 B2 0  I − P11 
 P*1,OE ( s) P*2,OE ( s)  =  C1 0 
s
y' 0 − P 
I − D11
w
I  21 

C2 0 − D21 0 − I 
z' u
 P12 0
 P*1,OE P*2,OE  = M −1
OE
G1,OE G2,OE  y' P
 22 − I  y K

 A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 
s

G1,OE ( s ) =  Wzz C1 Wzz −Wzz D11   RH 
 Wyz C1 + Wyy C2 Wyz −Wyy D21 
 
 A + H z C1 + H y C2 B2 + H z −H y 
s

G2,OE ( s ) =  Wzz C1 Wzz 0   RH 
 Wyz C1 + Wyy C2 Wyz −Wyy 

So far the procedure is identical to the general approach ! 52
Robust and Optimal Control - A Two-port Framework Approach

rcf:
 A + H zC1 + H yC2 B2 + H z −H y 
 WzzC1 Wzz 0   I 0 0 
    OE ( s ) 
 WyzC1 + WyyC2 −Wyy    −C1 Wzz −1 0 
 OE ( s ) 
Wyz
   −1 −1 −1  
 0 I 0   C2 Wyy WyzWzz −Wyy 
 0 0 I 

 A − B2C1 B2Wzz −1 + H zWzz −1 − H yWyy −1WyzWzz −1 H yWyy −1 

s
 
OE ( s )=I =  0 I 0 
 0 0 I 
 
 A − B2C1 B2Wzz −1 + H zWzz −1 − H yWyy −1WyzWzz −1 H yWyy −1 
s
 
 OE ( s ) =  −C1 Wzz −1 0   GH 
 C2 Wyy −1WyzWzz −1 −Wyy −1 


 I 0 −1
where OE =  and  OE = G2, OE .Since (A-B2C1) is Hurwitz, OE derived
 0 I 
for this OE problem is actually an identity matrix. Then the H-infinity controllers can

be generated by K = CSDr (G2,−1OE ,  ) = CSDl (G2,OE ,  ),   BH 

 z = CSDl (G1,OE , 0) w = −G11−1G12 w 53

Robust and Optimal Control - A Two-port Framework Approach

 A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 
s A+ H C + H C − B1 − H z D11 − H y D21 
s
  Hz
G1,OE ( s ) =  −Wzz D11  =
z 1 y 2
Wzz C1 Wzz  [G11 ( s ) G12 ( s )]  Wzz C1 Wzz −Wzz D11 
 Wyz C1 + Wyy C2 Wyz −Wyy D21   
 

 I − H zWzz −1   A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 0   A + H y C2 0 − B1 − H y D21 − H zWzz −1 

  = 
0 Wzz −1   Wzz C1 Wzz −Wzz D11 I   C1 I − D11 Wzz −1 

−1
 A + H y 2 C2
s − H zWzz −1 
 G11 ( s ) =  
 C1 Wzz −1 
 A + H y C2
−1
s B1 + H y D21 
z T (s)w
 T ( s ) = CSDl (G1,OE , 0) = −G11 ( s ) G12 ( s ) =  
 C1 D11 

Coprime factorization

因此可求得閉迴路系統T的poles會等於eig ( A H y C2 )

54
Robust and Optimal Control - A Two-port Framework Approach

 A + H z C1 + H y C2 B2 + H z −H y 
s
 
G2,OE ( s ) =  Wzz C1 Wzz 0 
 Wyz C1 + Wyy C2 Wyz −Wyy 


 I −( B2 + H z )Wzz −1   A + H z C1 + H y C2 B2 + H z −H y 0   A + H y C2 − B2C1 0 − H y −( B2 + H z )Wzz −1 
  = 
 0 W zz
−1
 Wzz C1 Wzz 0 I   C1 I 0 Wzz −1 

−1
s  A + H y C2 − B2C1 −( B2 + H z )Wzz −1 
G (s) =  
11
 C1 Wzz −1 

−1
 A + H y C2 − B2C1
s Hy 
 K ( s ) = CSDl (G2,OE , 0) = −G ( s)G12 ( s) = 
11
 C1 0 

Coprime factorization

Note that K = CSDl (G2,OE , ) = CSDr (G2,−1OE , ),   BH 

55
Robust and Optimal Control - A Two-port Framework Approach

P( s)

K( s )

I  A I B2C1 B1   A + H y C2 0 B1 + H y D21 
s  s   s  A + H y C2 B1 + H y D21  
LFTl ( P ( s ), K ( s )) =  H y C2 A + H y C2 − B2C1 D21H y  =  H yC2 A − B2C1 H y D21  =  
 C1 D11
 C C1 D11   C1 0 D11    
 1 
  A B1 B2 
 s
  s A H y C2 B1 H y D21
 P( s ) =  C1 D11 I  LFTl ( P ( s ), K ( s))
 C2 D21 0 
C1 D11
where 
 s A+ H C − B C Hy 
 K (s) =  y 2 2 1
A矩陣的Eigenvalue為A
  C1 0  H y C2 , A B2C1
56
Robust and Optimal Control - A Two-port Framework Approach

Output Injection (OI) problem (a dual part to SF)

z' z
A B1 I   I − P11 
 P ( s ) P12 ( s )  
D11 0 
S
0 − P 
POI ( s ) =  11 =  C1
y' w
  21 

 P21 ( s ) P22 ( s )  C D21 0 

 2 z' u
 A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21   P12 0

s
 y' P − I  y K
G1,OI ( s ) =  −Wzz D11  22
Wzz C1 Wzz   RH 
 Wyz C1 + Wyy C2 Wyz −Wyz D11 − Wyy D21 

 A + H z C1 + H y C2 I −H y  (I) ( A, B2 ) stabilizable and (C2 , A) detectable

s

G2,OI ( s ) =  Wzz C1 0 0   RH 
 A − j I B1 
 Wyz C1 + Wyy C2 0 −Wyy  rank  = q + n , rank D21 = q2
D21  2
 (II)
 C2
In order to make G1,OI  RH  is dual J-lossless, one has
 −
1

Wzz 0   ( I − D11 D11T ) 2 0 
WII =  = 1
Wyz Wyy   
1
−1 T − 2 −
T T T −1
  D21 ( I − D11 D11 ) D21  D21 D11 ( I − D11 D11 )  D21 ( I − D11T D11 ) −1 D21
T
 2 
H II =  H z H y  = −(YC T + B1 D*1T ) RII−1 ,

( ) 
T
A − B1 D*1 RII−1C −C T RII−1C
T

 
HY =  T
= 解得Y
T −1
( )
 − B1 ( I − D*1 RII D*1 ) B1 − A − B1 D*1 RII C 

T T
−1

 −( I − D11 D11 ) D11 D21 
T T
D  C 
RII =  T T 
, D*1 =  11  , C =  1 
 D D
21 11 D 21 21 
D  21 
D  C2 
57
So far the procedure is identical to the general approach !
Robust and Optimal Control - A Two-port Framework Approach

rcf:

 A + H z C1 + H y C2 I −H y 
 Wzz C1 0 0   I 0 0 
   OI ( s ) 
−1 
 Wyz C1 + Wyy C2  
0 −Wyy  y 2 H C + H W −1
y yy W C
yz 1 I − H W
y yy    ( s) 
  
0   C2 − Wyy−1Wyz C1 −Wy−y1   
OI
 0 I 0
 0 0 I 
 A + H z C1 + H y C2 I 0 
   11 ( s ) 0 
 OI ( s )  S 
W C
zz 1 0 0   0 I 
 ( s)  =  0 0 I =  
 OI   H C + H W −1W C I − H W −1   11 ( s ) − H yWyy 
−1

 y 2 y yy 
−Wy−y1 
 21 ( s )
y yy yz 1
 C2 − Wyy Wyz C1
−1
0 −1 
−Wyy 


Recall:  u   11 12   u   11 12   

 y  =     y  =   I  y
   21 22     21 22   

   12 
=  11 y

 21   22 

u =  11 + 12   21 +  22  y

−1

58
Robust and Optimal Control - A Two-port Framework Approach

rcf:

 A + H z C1 + H y C2 I −H y 
 Wzz C1 0 0   I 0 0 
   OI ( s ) 
−1 
 Wyz C1 + Wyy C2  
0 −Wyy  y 2 H C + H W −1
y yy W C
yz 1 I − H W
y yy    ( s) 
  
0   C2 − Wyy−1Wyz C1 −Wy−y1   
OI
 0 I 0
 0 0 I 
 A + H z C1 + H y C2 I 0 
   11 ( s ) 0 
 OI ( s )  S 
W C
zz 1 0 0   0 I 
 ( s)  =  0 0 I =  
 OI   H C + H W −1W C I − H W −1   11 ( s ) − H yWyy 
−1

 y 2 y yy 
−Wy−y1 
 21 ( s )
y yy yz 1
 C2 − Wyy Wyz C1
−1
0 −1 
−Wyy 


u = CSDr ( OI ( s ),  ) =  11 + 12  21 +  22  y

−1

And then central controller is given by 𝐶𝑆𝐷𝑟 (𝛱𝑂𝐼 (𝑠), 0)

u = CSDr ( OI ( s), 0) y = 12 22 −1 = (− H yWyy−1 )(−Wyy−1 ) −1 = H y y  x = H y y,
since B2 = I
 z = CSDl (G1,OI , 0) w = −G11−1G12 w 59
Robust and Optimal Control - A Two-port Framework Approach

 A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 
s A+ H C + H C − B1 − H z D11 − H y D21 
s
  Hz
G1,OI ( s ) =  −Wzz D11  =
z 1 y 2
Wzz C1 Wzz  [G11 ( s ) G12 ( s )]  Wzz C1 Wzz −Wzz D11 
 Wyz C1 + Wyy C2 Wyz −Wyz D11 − Wyy D21   


 I − H zWzz −1   A + H z C1 + H y C2 H z − B1 − H z D11 − H y D21 0   A + H y C2 0 − B1 − H y D21 − H zWzz −1 

  = 
0 Wzz −1   Wzz C1 Wzz −Wzz D11 I   C1 I − D11 Wzz −1 

−1
 A + H y 2 C2
s − H zWzz −1 
 G11 ( s ) =  
 C1 Wzz −1 
 A + H y C2
−1
s B1 + H y D21 
 T ( s ) = CSDl (G1,OI , 0) = −G11 ( s ) G12 ( s ) =   z T (s)w
 C1 D11 

Coprime factorization

因此可求得閉迴路系統T的poles會等於eig ( A H y C2 )

60
Robust and Optimal Control - A Two-port Framework Approach

B2 I

 A B1 I 
  D11 0 
s
s A+ H C B1 + H y D21   P( s ) =  C1
LFTl ( P( s ), K ) = 
y 2
 , where  C2
 C 1 D 11   D21 0 
K = H y

T ( s ) = LFTl ( P( s ), K ) = CSDl (G1,OI , 0)

可得出閉迴路系統poles為A H y C2 (意即A KC2 )

61
Robust and Optimal Control - A Two-port Framework Approach

Summary CSDr CSDl Same as the general case ! CSDl CSDr

Formulation of
Formulation of System Corresponding Plant Corresponding Plant
System

 A B1 B2   A B1 B2 
PDF ( s) =  C1 D11 D12  POE ( s) =  C1 D11 I 
s s
Disturbance
Output Estimation
Feedforward C2 I 0  (OE) C2 D21 0 
(DF)
with (A − B1C2 ) is Hurwitz with (A − B2C1 ) is Hurwitz

 A B2 
I 0 
B1
C A B1
D11 D12   s

PFI ( s) = 
s
Full Information Full Control
PFC ( s) =  C1 0 I 
1

I  D11
0 0 
0 0
(FI) (FC)
   I   0  C2 D21
 0     

 A B1 B2  A B1 I 
POI ( s) =  C1 D11 0
s

PSF ( s ) = C1 D11 D12 

State Feedback s Output Injection
(SF)
 I 0 0 
(OI) C2 D21 0

62