Vous êtes sur la page 1sur 5

Institute for Systems Theory and Automatic Control Uni Stuttgart

08.07.2010
Prof. Dr.-Ing. F. Allgower
Robust Control
ssc
http://www.ist.uni-stuttgart.de/education/courses/robust/

Exercise 5
Solution:
Problem 1
Consider the following system



 
0
1
1
x =
x+
w
2 2
1


y = 2 1 x.

Compute the H -norm of the transfer function G(s) = C(sI A)1 B by computing
the eigenvalues of the Hamiltonian matrix H.
Hint: Find the smallest such that the Riccati matrix X is positive definite.
Solution:
Compute the eigenvalues of H and check if ((H)) 6= 0.
Compute the matrix X and check if X > 0.
%system
A = [0 1; -2 -2];
B1 = [1; 1];
C1 = [2 1];
for gamma = linspace(2,3);
R=1/gamma^2*B1*B1; Q=C1*C1; H=[A R;-Q -A];
eH = eig(H);
figure(1)
plot(eH,.);
hold on
if min(abs(real(eH)))>1e-7;
[n,n]=size(A); [T,D]=eig(H);
Z=[];
for j=1:2*n
if real(D(j,j))>0;
Z=[Z T(:,j)];
end
end
T11=Z(1:n,:); T21=Z(n+1:2*n,:);
X=T21*T11^(-1);
eX=eig(real(X));
figure(2)
plot(gamma,min(eX),b.,gamma,max(eX),b.);

hold on
end
end
%test
G_hinf=normhinf(ltisys(A,B1,C1,0))

Problem 2
Consider the following system
x = Ax + B1 w + B2 u
z = C1 x + |{z}
I u.
D12



A sI B2
Suppose that
has full rank for all s C0 C+ .
C1
I
a) Show that A B2 C1 is Hurwitz.
Solution:


0 B2
0
I


 

A sI B2
A B2 C1 sI 0
=
C1
I
C1
I

b) With G12 (s) = C1 (sI A)1 B2 + I show that G1


12 exists and is stable.
Solution:
with D full rank, the inverse is given by


A B2 C1 B2
1
G12 =
C1
I
stability was already shown is part a), since the original system is stable, the
inverse has no unstable zeros.
c) Show that
inf

F such that AB2 F Hurwitz

k(C1 + D12 F )(sI A B2 F )1 B1 k = 0.

Solution:
F = C1

Problem 3
We want to investigate robust stability for the following plant with structured uncertainties. Assume that the measurement entering the controller is delayed by 1 time unit.
The plant G0 is perturbed by a multiplicative output uncertainty G with kG k 1.
a) Assume = 0 and G = 0. Transform the system into the P K-structure. Compute the generalized plant.
Solution:


z1
Ws Ws Ws G0
w1
z2 = 0

0
Wks
w2
y
I
I
G0
u

s = tf(s);
G = ss(2*(1-2*s)/((25*s+1)*(20*s+1)));
%Weights Ws = (s/M+w0)/(s+w0*A), Wks=1
M
= 2.5; w0 = 0.005; A=1.e-1;
Ws = (s/M+w0)/(s+w0*A);
Wks = tf(0.6);
W_tau = (s+0.08)/(0.45*s+0.65);
%nominal closed loop
systemnames=G Ws Wks;
inputvar=[w1;w2;u];
outputvar=[Ws; Wks;w1-G-w2];
input_to_G=[u];
input_to_Ws=[w1-G-w2];
input_to_Wks=[u];
P=sysic;
%then some parameters (number of measurements and inputs, and bounds on gamma )
nmeas = 1; nu = 1; gmn=0.5; gmx=20; tol = 0.001;
[Knom,Pclnom,gopt] = hinfsyn(P,nmeas,nu,gmn,gmx,tol);
P_hinf=normhinf(lft(P,Knom))
b) Now, the system is perturbed by the uncertainties 6= 0 and G 6= 0. Model the
input delay as a multiplicative input uncertainty (see exercise 3). Transform the
system into the M -structure. Compute M and .
Solution:


 

w1
(1 + G0 K)1 G0 KW
(1 + G0 K)1
z1
=
(1 + G0 K)1 G0 KW (1 + G0 K)1 G0 K w2
z2
{z
}
|
M


1 0
=
02
%uncertain closed loop
systemnames=G W_tau Ws Wks;
inputvar=[wde_tau;wde_G;w1;w2;u];
outputvar=[w1-w2-wde_G-G; G; Ws; Wks; W_tau+w1-w2-wde_G-G];
input_to_G=[u];
input_to_W_tau=[wde_tau];
WS

z1

z2

WKS

w2
G

w1

e s

+
K

G0

Figure 1: Uncertain Control Loop.

input_to_Ws=[W_tau+w1-w2-wde_G-G];
input_to_Wks=[u];
Pun=sysic;
omega=logspace(-4,2,300);
N=lft(Pun,Knom,1,1);
Consider now the following realization
G0 =

2(12s)
(25s+1)(20s+1) ,

WS =

W =

s+0.08
0.45s+0.65 ,

WKS = 0.6

0.4s+0.005
s+0.0005

and the controller


K=

3.981s2 + 0.3583s + 0.007963


s3 + 9.203s2 + 1.27s + 0.0006328

c) Compute the H -norm of M . What can you say about robust stability (Matlab
normhinf)?
%Hinf-norm of M
M=N(1:2,1:2);
hinf_M=normhinf(M)
d) Compute the structured singular value of M wrt. to (Matlab: mussv). What
can you say about robust stability? What do you observe?
% structured singular value of M
M0=frd(M,omega);
blk = [1 1; 1 1];
[mu_M,info]=mussv(M0,blk);
mu_bound=norm(mu_M(1,1),inf,1e-6)

Problem 4
Solve the online H design problems.
http://www.ist.uni-stuttgart.de/education/courses/robust/online/

Problem 5
Note: This problem was already part of exercise 4
In the robust control toolbox the H -control problem is solved by the command hinfsyn.
Therefore the generalized plant P has to satisfy the conditions 1. and 2. of slide 3 in
part 10.
If one or both conditions fail one can extend the performance channels with we and
ze


w
w
=
we


z
z =
ze

to get with some small > 0 the extended generalized plant P


x = Ax + B1 w + B1e we + B2 u
z = C1 x + D11 w + D12 u
e
ze = C1e x + D12
u
e
y = C2 x + D21 w + D21
we + D22 u.
e , D e this extended generalized plant satisfies the
For appropriate choice of B1e , C1e , D12
21
above conditions. Based on P an (almost) optimal H -controller K has been designed.

a) Argue why K stabilizes P if and only if K stabilizes P .


Solution:
P,22 = P,yu = C1 (sI A)1 B2 + D22 = Pyu = P22


I
K

P22
I

exists and is proper and stable

K stabilizes P
b) Show that if K stabilizes P and achieves kP K k < , then K stabilizes P
and achieves kP K k < .
Solution:

Pzw Pzw Pzu


(i) P K = Pz w Pz w Pz u , Pzw = P
Pyw Py,w Pyu
 
A
(ii) max{
(A),
(B)}

(A) +
(B)
B


A B

(A) +
(B) +
(C) +
(D)
max{
(A), . . .
(C)}

C D
K stablizes P and kP K k kK K k <
c) Show that if there exists a K stabilizing P and achieves kP K k < , then K
stabilizes P and achieves kP K k < for some sufficiently small > 0.
Solution:
kP K k kPzw K k + kPzw K k + kPz w K k + kPz w K k
{z
} |
{z
} |
{z
}
|
0

for 0
gain depends linearily on since is not in the feedback loop

Vous aimerez peut-être aussi