Aulomalic ConlroI Laboralory, ITH Zrich

www.control.ethz.ch
Cnntrn! 5ystcms II

Prnb!cm 5ct 10
George Xiaojing Zhang

[email protected]

Ovcrvicw
1. MIMO slabiIily crileria in frequency domain
SmaII Gain Theorem
SeclraI Radius Condilion
Nyquisl Crilerion
2. GeneraIized Ianl
Hov lo derive lhe Ianls
Robusl slabiIily
Part 1: MIMO 5tabi!ity !"# %&'(
Linear syslem is (intcrna!!y) stab!c if il
!"#$% Conlains no hidden unslabIe modes and lhe
in|eclion of bounded exlernaI signaIs al any Iace in
lhe syslem resuIls in bounded oulul signaIs
measured anyvhere in lhe syslem
: aII eigenvaIues of & Iie in oen LHI
: AII oIes of minimaI reaIizalion are in
oen LHI !'()*(+,(-#. 0(+)(-1(2%
G(s)

A B
C D

MIMO 5tabi!ity
SlabiIily of slandard feedback Ioo
Ioo lransfer funclion
5.9 Stability analysis in the frequency
domain [4.9]
Generalization of Nyquists stability test for SISO
systems.
5.9.1 Open- and closed-loop characteristic
polynomials [4.9.1]
e q
6
- - -
y
L
-
+ r
Figure 50: Negative feedback system
Open Loop:
L(s) = C
ol
(sI A
ol
)
1
B
ol
+D
ol
(5.69)
Poles of L(s) are the roots of the open-loop
characteristic polynomial

ol
(s) = det(sI A
ol
) (5.70)
5-36
L(s) :
5pcctra! Radius Cnnditinn
Assume 24(5,# ,667 4+(-2$#+ $8-)416- . Then
lhe c!nscd-!nnp systcm is stab!c if
If syslem gain is Iess lhan 1 in aII direclions
(eigenvaIues) and for aII frequencies, lhen aII
signaI devialions viII evenluaIIy die oul.
Sucienl condilion
Conservalive
L(s)
(L(j)) := max
i
|
i
(L(j))| < 1 .
5ma!! Gain Thcnrcm
Assume 24(5,# ,667 4+(-2$#+ $8-)416- . Then
lhe c!nscd-!nnp systcm is stab!c if
Any norm salisfying is OK
Oflen lake maximum singuIar vaIue
More conservalive lhan seclraI radius condilion
L(s)
kL(j)k < 1 .
kABk kAkkBk
(L)
Gcncra!izcd (MIMO) Nyquist
: - oen-Ioo unslabIe oIes in
CIosed-Ioo syslem is slabIe i Nyquisl Iol of
1. Makes anli-cIockvise encircIemenls of origin
2. Does nol ass lhrough origin
IoIes on -axis musl be incIuded inside "-conlour
L(s)
P
ol
det(I + L(s))
P
ol
j
Excrcisc 1
1. easy
2. easy
3. Use MATLA
Part 2: Gcncra!izcd P!ants !"# *&+(
a. Whal lhey are
b. Why lhey are usefuI (robusl slabiIily)
c. Hov lo derive lhem
Equivalent representation of Figure 64 where the
error signal to be minimized is z = y r and the
input to the controller is v = r y
m
c
c c c q
q
-
? ?
- -
?
6

- - -
P
u v
w

K
G
+
+
+
+
+
+
-
-
z
n
r
d
Figure 65: General control conguration equivalent to
Figure 64
6-29
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
Gcncra! Cnntrn! Cnnguratinn
9: exogenous inuls
:: exogenous oululs
8: conlroI signaIs
;: measured oululs
Equivalent representation of Figure 64 where the
error signal to be minimized is z = y r and the
input to the controller is v = r y
m
c
c c c q
q
-
? ?
- -
?
6

- - -
P
u v
w

K
G
+
+
+
+
+
+
-
-
z
n
r
d
Figure 65: General control conguration equivalent to
Figure 64
6-29
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
Gcncra! Cnntrn! Cnnguratinn
GoaI is lo minimize some norm of lhe lransfer
funclion from 9 lo :, e.g.
Iind a 0 lhal counleracls inuence of 9 on :
No modeI uncerlainly (yel).
H

6.4 General control problem

formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
Exercise 4 Generalized Plant, Cascade Implementation
It is common in practical applications to design cascaded control loops, including a fast reacting inner
loop and a safety or governing outer loop. For example, the control of an active suspensions system in
a car is usually designed like that: four decoupled fast inner loops control the active damping of each
wheel of the car, while the governing outer loop controls the overall behavior of the car.
A simple example of such a cascaded control loop is shown in Figure 3. In this case the output from
K
1
enters into K
2
and it may be viewed as a reference signal for y
2
.
r
-
+
-
c -
K
1
-c
+
-
-
K
2
-
u
2
G
2
-c
?
d
2
+
+
- q
y
2
6
G
1
-c
?
d
1
+
+
-
y
1
q
6
Figure 3: Common case of cascade control where the primary output y
1
depends directly on the extra
measurement y
2
.
For this example, derive the generalized controller K and the generalized plant P, using (y
1
r) as the
exogenous output. Proceed by rst dening writing down which signals are contained in the exogenous
inputs w, the exogenous outputs z, the control signals u, and the sensed outputs v (see Figure 4). Then
nd the corresponding transfer functions of the generalized controller u = Kv and the generalized
plant [z; v] = P[w; u].

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 4: General control conguration for the case with no model uncertainty.
3 of 3
Examp!c: Task
Examp!c: Dctcrminc ,
Exercise 4 Generalized Plant, Cascade Implementation
It is common in practical applications to design cascaded control loops, including a fast reacting inner
loop and a safety or governing outer loop. For example, the control of an active suspensions system in
a car is usually designed like that: four decoupled fast inner loops control the active damping of each
wheel of the car, while the governing outer loop controls the overall behavior of the car.
A simple example of such a cascaded control loop is shown in Figure 3. In this case the output from
K
1
enters into K
2
and it may be viewed as a reference signal for y
2
.
r
-
+
-
c -
K
1
-c
+
-
-
K
2
-
u
2
G
2
-c
?
d
2
+
+
- q
y
2
6
G
1
-c
?
d
1
+
+
-
y
1
q
6
Figure 3: Common case of cascade control where the primary output y
1
depends directly on the extra
measurement y
2
.
For this example, derive the generalized controller K and the generalized plant P, using (y
1
r) as the
exogenous output. Proceed by rst dening writing down which signals are contained in the exogenous
inputs w, the exogenous outputs z, the control signals u, and the sensed outputs v (see Figure 4). Then
nd the corresponding transfer functions of the generalized controller u = Kv and the generalized
plant [z; v] = P[w; u].

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 4: General control conguration for the case with no model uncertainty.
3 of 3
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
v =

r y
1
y
2

u =

u
2

Examp!c: Dctcrminc #
Exercise 4 Generalized Plant, Cascade Implementation
It is common in practical applications to design cascaded control loops, including a fast reacting inner
loop and a safety or governing outer loop. For example, the control of an active suspensions system in
a car is usually designed like that: four decoupled fast inner loops control the active damping of each
wheel of the car, while the governing outer loop controls the overall behavior of the car.
A simple example of such a cascaded control loop is shown in Figure 3. In this case the output from
K
1
enters into K
2
and it may be viewed as a reference signal for y
2
.
r
-
+
-
c -
K
1
-c
+
-
-
K
2
-
u
2
G
2
-c
?
d
2
+
+
- q
y
2
6
G
1
-c
?
d
1
+
+
-
y
1
q
6
Figure 3: Common case of cascade control where the primary output y
1
depends directly on the extra
measurement y
2
.
For this example, derive the generalized controller K and the generalized plant P, using (y
1
r) as the
exogenous output. Proceed by rst dening writing down which signals are contained in the exogenous
inputs w, the exogenous outputs z, the control signals u, and the sensed outputs v (see Figure 4). Then
nd the corresponding transfer functions of the generalized controller u = Kv and the generalized
plant [z; v] = P[w; u].

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 4: General control conguration for the case with no model uncertainty.
3 of 3
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
w =

d
1
d
2
r

z =

y
1
r

(:: given!)
Examp!c: #, - Cnnguratinn
Exercise 4 Generalized Plant, Cascade Implementation
It is common in practical applications to design cascaded control loops, including a fast reacting inner
loop and a safety or governing outer loop. For example, the control of an active suspensions system in
a car is usually designed like that: four decoupled fast inner loops control the active damping of each
wheel of the car, while the governing outer loop controls the overall behavior of the car.
A simple example of such a cascaded control loop is shown in Figure 3. In this case the output from
K
1
enters into K
2
and it may be viewed as a reference signal for y
2
.
r
-
+
-
c -
K
1
-c
+
-
-
K
2
-
u
2
G
2
-c
?
d
2
+
+
- q
y
2
6
G
1
-c
?
d
1
+
+
-
y
1
q
6
Figure 3: Common case of cascade control where the primary output y
1
depends directly on the extra
measurement y
2
.
For this example, derive the generalized controller K and the generalized plant P, using (y
1
r) as the
exogenous output. Proceed by rst dening writing down which signals are contained in the exogenous
inputs w, the exogenous outputs z, the control signals u, and the sensed outputs v (see Figure 4). Then
nd the corresponding transfer functions of the generalized controller u = Kv and the generalized
plant [z; v] = P[w; u].

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 4: General control conguration for the case with no model uncertainty.
3 of 3
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
v =

r y
1
y
2

u =

u
2

w =

d
1
d
2
r

z =

y
1
r

Equivalent representation of Figure 64 where the

error signal to be minimized is z = y r and the
input to the controller is v = r y
m
c
c c c q
q
-
? ?
- -
?
6

- - -
P
u v
w

K
G
+
+
+
+
+
+
-
-
z
n
r
d
Figure 65: General control conguration equivalent to
Figure 64
6-29
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
Gcncra! Cnntrn! Cnnguratinn
GoaI is lo minimize some norm of lhe lransfer
funclion from 9 lo :, e.g.
Iind a 0 lhal counleracls inuence of 9 on :
No modeI uncerlainly (yel).
H

Adding Unccrtainty.
9: exogenous inuls
:: exogenous oululs
8: conlroI signaIs
;: measured oululs
6.4.8 A general control conguration
including model uncertainty [3.8.8]
The general control conguration in Figure 63 may
be extended to include model uncertainty. Here the
matrix is a block-diagonal matrix that includes all
possible perturbations (representing uncertainty) to
the system. It is normalized such that

1.
-

- -
-

u v
z w
K
P

Figure 70: General control conguration for the case

with model uncertainty
6-42
u = K(s)v
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
Fu!! Cnnguratinn
6.4.8 A general control conguration
including model uncertainty [3.8.8]
The general control conguration in Figure 63 may
be extended to include model uncertainty. Here the
matrix is a block-diagonal matrix that includes all
possible perturbations (representing uncertainty) to
the system. It is normalized such that

1.
-

- -
-

u v
z w
K
P

Figure 70: General control conguration for the case

with model uncertainty
6-42
&,<624 (-= )6-4+6, 7+65,#<
914> 8-)#+4(1-4= )(- 5#
+#7+#2#-4#? 4>12 9(=@

z
v

P
11
P
12
P
21
P
22

w
u

Conlracl conlroIIer 0A2B inlo Ianl by Iover

Iinear fraclionaI lransformalion (LIT)
N
6.4.8 A general control conguration
including model uncertainty [3.8.8]
The general control conguration in Figure 63 may
be extended to include model uncertainty. Here the
matrix is a block-diagonal matrix that includes all
possible perturbations (representing uncertainty) to
the system. It is normalized such that

1.
-

- -
-

u v
z w
K
P

Figure 70: General control conguration for the case

with model uncertainty
6-42
- -
-

z w

N
Figure 71: General block diagram for analysis with
uncertainty included
! !
! !
!
!
!
!
1
2
3
4
1 2
3 4
System with
Actuators, Sensors
and Controller
Outputs
Inputs
N
>
w
z
w z
(a) (b)
Figure 72: Rearranging a system with multiple per-
turbations into the N-structure
6-43
- 5tructurc: Rnbust PcrI. |5P 8.10{
- 5tructurc: Rnbust PcrInrmancc
N
N = F
l
(P, K) = P
11
+P
12
K(I P
22
K)
1
P
21
=

N
11
N
12
N
21
N
22

6.4.8 A general control conguration

including model uncertainty [3.8.8]
The general control conguration in Figure 63 may
be extended to include model uncertainty. Here the
matrix is a block-diagonal matrix that includes all
possible perturbations (representing uncertainty) to
the system. It is normalized such that

1.
-

- -
-

u v
z w
K
P

Figure 70: General control conguration for the case

with model uncertainty
6-42
- -
-

z w

N
Figure 71: General block diagram for analysis with
uncertainty included
! !
! !
!
!
!
!
1
2
3
4
1 2
3 4
System with
Actuators, Sensors
and Controller
Outputs
Inputs
N
>
w
z
w z
(a) (b)
Figure 72: Rearranging a system with multiple per-
turbations into the N-structure
6-43

z
v

P
11
P
12
P
21
P
22

w
u

N
11
N
12
N
21
N
22

u

- 5tructurc: Rnbust 5tabi!ity !"# +&-(

- -
-

z w

N
Figure 71: General block diagram for analysis with
uncertainty included
! !
! !
!
!
!
!
1
2
3
4
1 2
3 4
System with
Actuators, Sensors
and Controller
Outputs
Inputs
N
>
w
z
w z
(a) (b)
Figure 72: Rearranging a system with multiple per-
turbations into the N-structure
6-43
M
M

Consider onIy erlurbalion Ioo

M(s) = N
11
(s)
N =

N
11
N
12
N
21
N
22

N
11
N
12
N
21
N
22

u

M
M

Assume 'A2B and are slabIe. Then above

syslem is slabIe for aII erlurbalions i
(s)

kMk

< 1
(M(j)) ((j)) < 1 ,
- 5tructurc: Rnbust 5tabi!ity !"# +&-(
Excrcisc 2
1. Comule cIosed-Ioo lransfer funclions for
- conhguralion, i.e. hnd '
2. Use MalIab lo Iol A9>1)>CB singuIar vaIue
3. Use Robusl SlabiIily Theorem
M
Excrcisc 3
1. Derive generaIized Ianl, read from bIock
diagram
2. Irom D0-conhguralion lo E-conhguralion
a. cIose lhe Ioo manuaIIy
b. use Iover LIT
En|ny.
6.4 General control problem
formulation [3.8]

-
- -
K
P
sensed outputs control signals
exogenous inputs
(weighted)
exogenous outputs
(weighted)
u v
z w
Figure 63: General control conguration for the case
with no model uncertainty
The overall control objective is to minimize some
norm of the transfer function from w to z, for
example, the H

norm. The controller design

problem is then:
Find a controller K which based on the information
in v, generates a control signal u which counteracts
the inuence of w on z, thereby minimizing the
closed-loop norm from w to z.
6-27
6.4.8 A general control conguration
including model uncertainty [3.8.8]
The general control conguration in Figure 63 may
be extended to include model uncertainty. Here the
matrix is a block-diagonal matrix that includes all
possible perturbations (representing uncertainty) to
the system. It is normalized such that

1.
-

- -
-

u v
z w
K
P

Figure 70: General control conguration for the case

with model uncertainty
6-42
M

- -
-

z w

N
Figure 71: General block diagram for analysis with
uncertainty included
! !
! !
!
!
!
!
1
2
3
4
1 2
3 4
System with
Actuators, Sensors
and Controller
Outputs
Inputs
N
>
w
z
w z
(a) (b)
Figure 72: Rearranging a system with multiple per-
turbations into the N-structure
6-43
Teslal: 3 oul of 4
OH: Monday, 1-2m, ITL I 10.1
2: add uncerlainly
3: Iover LIT
4:
robusl slabiIily
1: generaI
conlroI conhguralion