Professional Documents
Culture Documents
1
>@?BADCFEHG>(IKJLAKM<NPORQ"G
Norwegian University of Science and Technology
S QUTWVXJLNYOHZBMOH[D\/QU?]ORM
University of Leicester
^`_baLcedgfihkjml+npoP_qcqo
rksutwvxsuy{z}|~y{"&yi"7itz~yUP~|i2Utuiy
The solution manual was prepared by Yi Cao, Vinay Kariwala, Eduardo Shigueo Hori and
Sigurd Skogestad
This version prepared October 2, 2007
Note: b k X&B}k{kX
(2.31)
!u b
(b !"{
is defined by
(2.36)
where .
H9 Y}{ kx }{ x }` ]B Y m]{
Solution. According to (2.31), the following equation can be obtained:
} ]{ }{ kx Y}{ }` ] }{` ]]
Hence,
leads to
k] ] H B]B
3 ]]
which gives the solution of
and
{@B{x ]@B{]{
9Y R"@B{]{
{
] ] {
B 9 given in (5.96) for a first-
Exercise 2.3 Derive the approximation for U
order delay system.
b u
by #
! %
"
$ (
& ' ) , and
b
Note: A first-order delay system can be represented
* !+
) (5.96)
9 6 0 9 6 89:0 .
;
F <
= < ?
> @
Solution.
to be @ , then A< @ and
@" C B A<
B B
Cascade rule As shown in Fig. 3.1(a), let the signal from
D , or,
G Fm HD B . Let I , we can obtain, D A& GB F& I(E B . B
, thus, , i.e. .
Feedback rule As shown in Fig.3.1(b), we have, D9 B
Exercise 3.3 Use the MIMO rule to show that (2.19) corresponds to the negative feedback
system in Figure 2.4.
G &
F E E and exiting
9
GF the
GFk , from K to E is GFk L and from M to E is <GF7 . So
Solution. In Fig. 2.4, from point the loop transfer function is loop gives
the term . Therefore, the transfer function from J to is
Exercise 3.5 Compute the spectral radius and the five matrix norms mentioned above for the
matrices in (3.29) and (3.30).
Solution.
:
Matrix (3.29):
N 5O
P
Q Q(R 2 {
Q Q(SUTWV i
Q Q(X
Q Q XZY
Q Q X {
N x
Matrix (3.30):
[O
P
Q QWR : Q Q SUT(V Q Q X Q Q X Q Q X\Y x
Exercise 3.7 Design decentralized single-loop controllers for the plant (3.65) using (a) the
diagonal pairings and (b) the off-diagonal pairings. Use the delay (which is nominally 5
.
seconds) as a parameter. Use PI controllers independently tuned with the SIMC tuning rules
(based on the paired elements).
m] B 0 . x 0` 0 . x 0 . x 0W
and for the off-diagonal pairings (the index refers to the output)
0 .`qWf x 0W 0 .q `f x mx 0 . x 0`
For improved robustness, the level controller (E ) is tuned about 3 times slower than the
pressure controller (E ), i.e. use 0
-. and 0 . . This gives a crossover frequency
of about
/.
in the fastest loop. With a delay of about 5 s or larger you should find, as
expected from the RGA at crossover frequencies (pairing rule 1), that the off-diagonal pairing
is best. However, if the delay is decreased from 5 s to 1 s, then the diagonal pairing is best, as
expected since the RGA for the diagonal pairing approaches 1 at frequencies above 1 rad/s.
Figures 3.7a and b present the simulations for these pairings.
ac
i.e. select and where
c N/ c ac O
c
;
,
and try the following values:
{-;f , {{ .
1.
2.
3.
H
2.5
2
1.5 y
1
1
y
0.5
0 y
2
0.5
0 50 100 150 200
(a) Diagonal pairing
2.5
2
1.5 y
1
1
y
0.5
0 y
2
!
0.5
0 50 100 150 200
(b) Offdiagonal pairing
2
y1
0
y
y
4
0 200 400 600 800 1000
(a) Diagonal pairing
2
y1
0
y
y
X
4
0 200 400 600 800 1000
(b) Offdiagonal pairing
c . What is the condition number of the controller in the three cases? Discuss
the results. (See also the conclusion on page 251).
Solution. The simulation results of designs (a) and (b) are shown in the following figures.
2.5 2.5
2 2
1.5 1.5
y1 y1
1 1
0.5 0.5
y2 y2
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
TIME (min) TIME (min)
Design (b) is the best if we want a fast settling time. By increasing we increase the loop
gain in the weak direction.
(Y (Y
The condition number of the controller:
(Y ] {{ (BP {{ {
Design (a): and ;
(Y Y
(
Design (b): and ;
Design (c): and .
From this, design (a) is always robust with respect to uncertainty, whereas designs (b) and (c)
are potencially sensitive to uncertainties.
9^] in (3.93). The response using
5 C
Exercise 3.12 Consider again the distillation process
the inverse-based controller in (3.95) was found to be sensitive to input gain errors.
Y
We want to see if the controller can be modified to yield a more robust system by using
3 s
the Glover-McFarlane
)3 C
loop-shaping procedure. To this effect, let the shaped plant be
Y
c , i.e. , and design an controller for the shaped plant
c 5 C{
: XZ i c
(see page 370 and Chapter 9), such that the overall controller becomes . (You
will find that which indicates good robustness with respect to coprime factor
uncertainty, but the loop shape is almost unchanged and the system remains sensitive to input
uncertainty.)
Solution. The following MATLAB script could be used to do this exercise. (To use these
commands it is necessary to have the old -toolbox)
G0 = [87.8 -86.4; 108.2 -109.6];
dyn = nd2sys(1,[75 1]); Dyn = daug(dyn,dyn);
G = mmult(Dyn,G0);
% Inverse-based controller
H
dynk = nd2sys([75 1],[1 1.e-6],0.7);
Dynk = daug(dynk,dynk);
Kinv = mmult(Dynk,minv(G0));
% TIME simulation
% Nominal
GK = mmult(G,K); I2=eye(2); S = minv(madd(I2,GK));
T = msub(I2,S);
kr=nd2sys(1,[5 1]); Kr=daug(kr,kr); Tr = mmult(T,Kr);
y = trsp(Tr,[1;0],100,.1);
u = trsp(mmult(K,S,Kr),[1;0],100,.1);
% With 20% uncertainty
Unc = [1.2 0; 0 0.8]; GKu = mmult(G,Unc,K);
Su = minv(madd(I2,GKu)); Tu = msub(I2,Su);
Tru = mmult(Tu,Kr);
yu = trsp(Tru,[1;0],100,.1);
uu = trsp(mmult(K,Su,Kr),[1;0],100,.1);
subplot(211);vplot(y,yu,--);title(OUTPUTS)
subplot(212);vplot(u,uu,--);title(INPUTS);
xlabel(TIME (min));
E
Exercise 3.14 Cascade implementation. Consider further Example 3.21. The local
E
feedback based on is often implemented in a cascade manner; see also Figure 10.11. In
this case the output from enters into and it may be viewed as a reference signal for .
Derive the generalized controller and the generalized plant in this case.
L
J 3 + - + -
+ B
+ E +
+ E
.
1. Let in , we get,
/ Y
F
2. From
H
F GF7GF7 (
, we get,
F
GFk (
b F
GF ( . Thus, k
The common factor is .
3. Since , we have,
and F.
As conclusion, we get,
F FW k
Exercise 3.18 Consider, the^] performance specification
Q , Q Y . Suggest a rational
transfer function weight and sketch it as a function of frequency for the following two
cases:
1. We desire no steady-state offset, a bandwidth better than rad/s and a resonance peak
{
(worst amplification caused by feedback) lower than .
x
]
2. We desire less than % steady-state offset, less than % error up to frequency rad/s, a
bandwidth better than rad/s, and a resonance peak lower than . Hint: See (2.105) and
(2.106).
c a c a , and , d c e(
} } Comment: In the latter case we
,
Solution.
require that the magnitude
should
] (fromincrease c 10 when the frequency increases by
by a factor
approximately a factor 3 rad/s to 10 rad/s).
Exercise 3.20 What is the relationship between the RGA-matrix and uncertainty in the
individual elements? Illustrate this for perturbations in the
-element of the matrix
]
N O (3.114)
Solution. The inverse of the RGA-element directly gives the relative change in the element
that gives singularity.
& N
{ O
RGA-matrix has 1,1-element of -80, so G becomes singular if 1,1-element is perturbed from
10 to 10(1+1/80)=10.125.
]
Exercise 3.22 Compute
Q Q(X , Q QxX , Q Q(XZY , Q Q R , Q Q(V and Q QxSUTWV for
$
the following matrices and tabulate your results:
F N
O N
O e N
O a N $ O
b !
Show using the above matrices that the following bounds are tight (i.e. we may have equality)
Q Q R
for matrices ( ):
Q Q V 7 Q Q V
Q Q X 7 Q Q X
Q Q X\Y 7 Q Q XZY
Q
Q(R Q Q SUT( V
Solution. The result table is as follows:
/U /\ //\ `x `8
1.
1 1 1 1.4142 1 2
2.
1 1 1 1 1 1
3.
2 2 2 2 1 4
4.
1 1.4142 2 1.4142 1 2
5. 2 1.4142 1 1.4142 1 2
Q From Y
the
Q table,
QWR , theQ above
Q(R bounds can be checked:
Q Q
Q Q(X :@" , , e YP Q e QxX `
Q
Q a Q(XZY a , e Y Q e QxX\Y
Q e QWR Q Q e a a
c
Exercise 3.24 Do the extreme singular values bound the magnitudes of the elements of a
matrix? That is, is greater than the largest element (in magnitude), and is smaller
than the smallest element? For a non-singular matrix, how is related to the largest
Solution. The answer for the first question is yes, because F for which ,
matrix,
second question, the answer is no. As an example, consider
is smaller than inverse of the
is 0. For a non-singular
largest element in , because
but the smallest element
B ] Q QxV .
H
{
Exercise 3.26 Find two matrices and such that P
P P which proves
that the spectral radius does not satisfy the triangle inequality and is thus not a norm.
P P
Solution. Let N O , and N
O . P R
9 , but 9
P P 9 .
3.28 Write as an LFT of
G F@4 ( , i.e. find such that
.
Exercise
G m
F G Fm ( gives as an LFT of , but we want the inverse. We get
, so
Solution.
5
b NF F O
and find . controllers in (4.94) can be written as
Exercise 3.30 Show that the set of all stabilizing
Solution. In (4.94), 3
.
~ w
% G Fm B
G Fm G G ]
G ~ G
O
We get
bN U U (3.115)
B
= ,
=
Exercise 4.1 We want to find the normalized coprime factorization for the scalar system in
9 f 9 f
(4.22). Let and
9
be as given in (4.23), and substitute them into (4.24). Show that after
^] P ] , (4.24) can be written as ] ^] ] ^]
some algebra and comparing of terms one obtains: , and .
. Thus:
9 8] e F-] x
Solution. Since
] e 9 /9 ] Y9
89 9
By comparing the terms of the denominator and numerator, it can be obtained that:
x-9 9 9 f{
{-9 9 /9 9 f
9^] k ^]WF< with just one state,
Exercise 4.3 (a) Consider a SISO system
i.e. is a scalar. Find the zeros. Does
9
^ ] have any zeros for ? (b) Do and
9 / . When the zero moves to infinity, i.e. no
have the same poles and zeros for a SISO system? Ditto, for a MIMO system?
K=[1 (s+3)/(s+4)]
The resulting 2x2 transfer function has no (MIMO) zeros, but the 1x1 transfer function
is
x
2 s2 + 10 s + 10
----------------------
s3 + 7 s2 + 14 s + 8
same; the two poles of plus the pole of .
which has two zeros at -3.62 and -1.38, respectively. The poles for both and are the
Thus, zero at
] .
(b) From B
^ ] to @ ^] , the polynomial system matrix is:
]
k
^ ] YN O
9^ ] F c . Thus, there are no zeros in this
and the corresponding transfer function is
transfer function. c
Exercise 4.7 For what values of does the following plant have RHP-zeros?
vsl l F vsl ll
l jv vsl l l v (4.1)
]] m
Solution.
] ] ]R m
^
]
] , the sum of the 2nd and 4th rows of ^ ] is equal to the 3rd row. Thus A& .
If
So the plant has RHP-zero if
.
Exercise 4.9 Use (A.7) to show that the signal relationships (4.83) and (4.84) may also be
^] ^] Y j
written as
(4.85)
From this we get that the system in Figure 4.3 is internally stable if and only if
^ ] is stable.
7?? H7
:
]
G Fk K GF7 ( K
Solution. Signal relationships (4.83) and (4.84) are:
B
E 9G F7 ~ K GFk (~ K
(4.83)
(4.84)
GFk GF7
Which can be written as:
N E O N 9GFk G Fk (( O N KK O
B
9GF G F ( and GFH Fk G FH ( :
GF G F Fm GF GF (
Noting that
N 9GF G F (( O N KK O N GF7 ( ( G F7 ( O
Since 3
F7 and using (A.7):
F k F
N k
N k F O
O
^] Y ]Y 8 ]R ]R ^ ] ]BPk- ] -]
Exercise 4.11 Given the complementary sensitivity functions
^] ^ ]
functions, I and I ?
what can you say about possible RHP-poles or RHP-zeros in the corresponding loop transfer
Solution. Assume
1) Compute
c c~ . Since has a RHP-zero, we conclude that I is
internal stability:
c } (and
c also with] a pole ] ).
at
2) has a RHP-zero, so I has a RHP-zero at
unstable with a RHP-pole at s=1.2
.
Exercise 4.13 Show that the IMC-structure in Figure 4.5 is internally unstable if either
or
B X J E where G F9
is unstable.
G 7
F
Solution. In Figure 4.5, it can be derived that
i.e. equation (4.89). Thus,
( . According to Lemma 4.6 if either or
,
is
unstable then the IMC system is internally unstable.
Exercise 4.15 Given a stable controller . What set of plants can be stabilized by this
controller? (Hint: interchange the roles of plant and controller.)
G F< ( G F
Solution. The set of the plants can be parameterized as:
, F
=
" = :
=
Solution. 1. Optimal control with state feedback yields a loop transfer function with a pole-
zero excess of 1 so (5.5) does not apply.
2. There are no RHP-zeros when all states are measured so (5.9) does not apply.
Q QY Q QY Q
QY Q QY
Exercise 5.3 Consider again the plant (5.24) from Example 5.2. Compute the bounds
on ,
,
controlling this plant?
and using Table 5.1. Do you expect any difficulties in
4 5
f f
*27
V
As the minimum bounds C and
98
are larger than the typical
maximum allowed value of about 2, it is expected that this plant will be difficult to stabilize
and control from Q QWY
;& a practical point of view. QE Q
Also, , which may not be acceptable if it is required to keep
input disturbances less than 1 (this would require .
Q WQ Y Q
for all
Solution. When A
is real, from , A we have:
, H A HB
When A is imaginary, we get:
H L
M 8&9 9 xi
M NK A POS H 4 }R
Q UT
b
A 4WV bM 89 9 xx
A as M X
In all these cases, H
I .
Exercise 5.9 Consider the case of a plant with a RHP-zero A where we want to limit the
YGE c 4 c Y GE 4
sensitivity function over some frequency range. To this effect let
,
^ ]
Y
Y GE c Y GE c (5.59)
B x
This weight is equal to at low and high frequencies, has a maximum value of about
x{
at intermediate frequencies, and the asymptote crosses at frequencies H and H .
]{
Thus we require tight control,
, in the frequency range between H[Z H
and H < H .
a) Make a sketch of
B ,
(which provides an upper bound on ).
A
A {
b) Show that the RHP-zero cannot be in the frequency range where we require tight
, A
9 A x x x { { x{
control, and that we can achieve tight control either at frequencies below about (the usual
6 9A
case) or above about . To see this select and evaluate for various values of
, AP
H
, e.g.
9
. (You will find that
A { 9 {
{
( ) for (corresponding to the requirement H
(corresponding to the requirement [
H Z ))
A
< ) and for
\^]7
?H_:`z_a7b
1
10
M
1/|Wp(w)|
0
10
1
10
wb/1000 wb/100 wb/10 wb
Frequency (rad/s)
]{ B,
Solution. a)The asymptotic magnitude Bode-plot is shown as follows (see Figure 5.9):
condition
, A A
b) In the above plot betweem H and H
, cannot be in the frequency range from H
]{
. Hence to satisfy the
to H where
,
The values of A via
we require tight control.
9 H A for are shown in the following table:
9 , A
0.1 0.59350
0.5 0.94788
1.0 1.3508
10.0 4.7966
100.0 4.7966
1000.0 1.3508
2000.0 0.94788
10000.0 0.59350
% H a similar ] analysis
of the
Exercise 5.10 For purely imaginary poles located at %
shows that we must at least require
weight (5.68) with
% . Derive this
bound. k {;f
cd >e g hg
\m)on drq
g hg
k j
a tFu e
}
}$
k jl p s k
Solution. }
i G9E j } j
Exercise 5.11 For a system with a single real RHP-zero and v RHP-poles % A X and tight
{
control at low frequencies (
in (5.50) derive the following
'z{ generalization of (5.52):
H , X
Axwy X:(+- * AA %% X
(5.74)
{ A
(Hint: Use (5.13).) Note that for a plant with a single RHP-pole and RHP-zero the
bound (5.74) with is feasible (upper bound on H is positive) for % .
{ A
This confirms the approximate bound % derived for stability with acceptable low-
frequency performance and robustness on page 196.
, A d X:, ( - * A % XX Q, QY
A %
/ 021 3
4o5
*7
, A }| A , , H
Considering the performance weight (5.50):
|
| A H (5.50)
|
So, we have that:
|
|
|
|
A , , H | d , ( * A % XX
| X:- A %
|
| A H
| |
|
A , H | , (+* A % X
| , H | X:- X
| A || A %
)
|
| |
When A is real, all variables are real and positive, so (with
z{
H Axwy X:, (+- * A % XX
A %
Exercise 5.12 Perform closed-loop simulations with the SIMC PI controller and the
proposed PID controller for the room heating process. Also compute the robustness
parameters (GM, PM, C and ) for the two designs.
9^] Y x{ 8{_ ] c L ^ ] Y x{ ]]
Solution. The model for the heating room is:
SIMC:
and 0W { s
The SIMC and PID tunings used were:
PID: H
, 0`
{{ s and 0~ f s
\^]7
?H_:`z_a7b
The closed-loop simulations with SIMC PI and PID controllers are presented in Figure
5.12.
x- f{ K19 { { f{
The robustness parameters are:
For SIMC PI: , , C , and
f{ {{ K19 f/f f{-f { ]
;
For PID: , , C , and .
1.8 3.5
1.6
3
1.4 SIMC PI
2.5
1.2
2
1
0.2 0.5
0 200 400 600 800 1000 0 200 400 600 800 1000
Time Time
]{
Exercise 5.14 (a) The effect of a concentration disturbance must be reduced by a factor of
at the frequency rad/min. The disturbances should be dampened by use of buffer tanks
and the objective is to minimize the total volume. How many tanks in series should one have?
What is the total residence time?
(b) The feed to a distillation column has large variations in concentration and the use of
E
one buffer tank is suggested to dampen these. The effect of the feed concentration on the K
L^] Y _ % c ]
product composition is given by (scaled variables, time in minutes)
K E
That is, after a step in the output will, after an initial delay of min, increase in a ramp-like
fashion and reach its maximum allowed value (which is ) after another minutes. Feedback
control should be used and there is an additional measurement delay of minutes. What
should be the residence time in the tank?
(c) Show that in terms of minimizing the total volume for buffer tanks in series, it is optimal
to have buffer tanks of equal size.
(d) Is there any reason to have buffer tanks in parallel (they must not be of equal size
because then one may simply combine them)?
(e) What about parallel pipes in series (pure delay). Is this a good idea?
Solution.
`
(a) From (5.116) we must require
0 M x
This gives
M 0
With rad/min we find
{ {
B
{
f {
B
]5B
0
0
and we find that to minimize we should have 6 tanks in series with a total residence time
min. (In practice, we would prefer fewer tanks because the number of tanks
increases the cost, and probably use only 2 or at most 3 tanks).
^] Y ]
(b) Assume the transfer function of the tank is
0
The total delay in the feedback loop is
.
(
f [min]. To make the system controllable,
we need
L8 . d 8.
& ]
This leads to 0 [min].
0
(c) Consider the case with two buffer tanks with total residence time . Let the residence
@
^] Y R @ ] 0 @ ] 0 ]R 0 @H@ ]
time in one of the tanks be . Then
Exercise 5.16 What information about a plant is important for controller design, and in
particular, in which frequency range is it important to know the model well? To answer this
problem you may think about the following sub-problems:
(a) Explain what information about the plant is used for Ziegler-Nichols tuning of a SISO
PID-controller.
9
9 ^] &
(b) Is the steady-state plant gain
important for controller design? (As an example
^ ] & 3
& x
9 P B c
consider the plant with and design a P-controller such
that . How does the controller design and the closed-loop response depend on the
steady-state gain ?)
\^]7
?H_:`z_a7b
Solution. (a) and
9
, that is, only information at frequency is used.
,
(b) As just noted the steady-state information is not needed for Ziegler-Nichols tuning, and
L ]{k 6 ]{ 6
it is generally not important for feedback design. This is illustrated by the example where
9 9
for any value of
effected by . Main effect is on steady-state offset which is
B$
. Thus response hardly
and may vary
between (for ) and (for ).
where and are in C, is in kg/s, and the unit for time is seconds. Derive the scaled
model. Is the plant controllable with feedback control? (Solution: The delay poses no problem
(performance), but the effect of the disturbance is a bit too large at high frequencies (input
E ^] f- ] B8] B ^] f{ ] B-]R K ^]
L , when } } , i.e. at high
e D } is too large which will cause input saturation. Thus the plant is
So,
frequencies the disturbance
not controllable.
, F
=
" = = =
=
_ c 6 .-]
^ c] F
Exercise 6.3 To illustrate further the above arguments, determine the sensitivity function
Solution. Using F
G Fk ( and _ c 6 . ] , it can be obtained that
]R 9 9
F ]P 9 .-9 N 9uH . ] ]R 9 O
] 6 ), the elements of all have the magnitude of Bm .-9 .
Thus at low frequencies (
To maintain 10% offset at low frequencies, let
]
8
. 9 B
]
9
. It leads to
8 . . The
9uHb .]
bandwidth is obtained by letting:
|| @9.-9 || {
9 9 . 9 || u .-9 . Since 9 || 8 . , .89 2 , i.e. 9 . 6 .-9 .
This gives
9 , i.e. the bandwidth is equal to
9
Thus the equation for bandwidth can be simplified as
.
, X A B9 X~
diagonal performance weight matrix
X 8 8
, X
X
, ,
If
X E
8 is given by (5.50) and 8
than ), show that tight control of EX (arbitrarily poor control of all outputs other
at low frequencies imposes the following limitation on
H :
8 H X A
B9 X
?
A
8
8
8f
Solution. Q Q Y
could consider
Q QY
Unfortunately, the result does not hold as originally stated: Instead of one
(which is OK mathematically but makes less sense from a
physical point of view).
Alternatively (and better), the exercise should start with For a system with a I
I ^ ] B
single.... The solution then goes as follows:
If A
has a RHP-zero at with input direction , then for internal stability of the
A B9 B9
feedback system the following interpolation constraint applies:
Then, if we multiply B
by , we have that
B B
F Kaa G, X ), we have that
Kaa G, X 9B Y
As is diagonal (
8 8 8 8
So X , X A B9 X Y X , X A B9 X
8 8 8 8
, X
Considering that is given by (5.50) and
,
, , from
8
X , X A B 8 X
8 8
,
A , H H BF 8 X
we have that
4
With
(no steady-state offset):
A , H B A
A
Then
, H A X
B9
8
9^] YN O
Exercise 6.6 Consider the plant
c
(6.39)
with (
{ in (6.38) if we want to control E perfectly.
(c) Output is most difficult since the zero is mainly in that direction; we get strong interaction
9^] Y B] Y x N i c O
Exercise 6.8 Analyze input-output controllability for
cc
Compute the zeros and poles, plot the RGA as a function of frequency, etc.
x , x , and ]{ ; 2 zeros: { and (RHP-
zero). The diagonal elements of the RGA are:
Solution. The system has 6 poles: R } } and off-diagonal
elements of the RGA are:
. The amplitudes of
these }
} elements are shown
as functions of frequency in Fig. 6.8(a). It is shown in the figure that all elements of the RGA
1
10
H^}
H}} }
0
10
1
10 2 1 0 1 2
10 10 10 10 10
N m ]
F C N ]] 2
Exercise 6.10 Let
m O C O N O
(a) Perform a controllability analysis of
9^] .
{
0
10
1
10
2
10
3
10
4
10
5
10 2 1 0 1 2
10 10 10 10 10
(b) Let b@ @ B K
and consider a unit disturbance ) KK
R . Which direction
(value of A A
) gives a disturbance that is most difficult to reject (consider both RHP-zeros
and input saturation)?
(c) Discuss decentralized control of the plant. How would you pair the variables?
9^] YN c c O
Solution. (a) The transfer function matrix is:
The system has two stable poles: -10 and -1 and one RHP-zero: A<
c . The RGA matrix is:
c
N c~ c~c
O
c~ c~
Thus, the magnitudes of the RGA elements are large (about 10) at low frequencies and small
at high frequency (approximating to a unit matrix). But the approximation is at 2
2 A,
9
i.e. outside the bandwidth which by the RHP-zero, A< .
2is{limited b
b N O and the most difficult output direction is B
f{ { . Let K@ L 9 Q .LThen, L K@ {W 9 . So,
(b) At steady-state,
m
9 m
in the most difficult direction, B , i.e. H ` 9 f { { . This
Q
gives .
< P m]
N m] O
(c) At the steady state,
Remember to also consider the sensitivity to input gain uncertainty.
have the same RGA matrix which is constant with
. These large RGA values indicate both and are difficult to
Solution. and
approximation of #c
be controlled. Additionally
_ 6 H can be expressed as: and the
L
Exercise 6.14 Analyze input-output controllability for
9^] YN xx xx O L
^ ]
YN c O N c O
c c
Which disturbance is the worst?
k
Disturbance K is in the same direction of B , thus is easy to be rejected. The disturbance
Solution. This is a ill-conditioned plant. The RGA, at all frequencies.
Exercise 6.16 Find the poles and zeros and analyze input-output controllability for
9^] PN B8]] -] B8]] -] O
.
Here c is a constant, e.g.
Remark. A similar model form is encountered for distillation columns controlled with the
-configuration. In which case the physical reason for the model being singular at steady-
state is that the sum of the two manipulated inputs is fixed at steady-state, .
the computations we select
. The plant has two poles at ] and one
LHP-zero at
Solution. For ]
9^] Y ]B N/ ]R ]R O
, e.g. consider the realization
9
]R
Here
^ ] has a zero at since the matrix N ]] O has a zero for ] .
]
(alternatively, we may have a realization with only one pole at
(scalar),
s F ] e.g.
N )), but we then have no zero at ). In any case, the plant
b O
is unstable and needs to be stabilized.
The plant is singular at steady-state (with infinite RGA and condition number). This may
seem to indicate that control is very difficult, but this is not the case. The reason is that the
]
singularity is caused by the largest singular value being infinite at steady-state, whereas the
]
smallest singular value is at so it is non-zero (if the smallest singular value was
zero at then tight control at steady-state would be impossible).
In summary, acceptable control for this plant is possible provided the bandwidth is
sufficiently high (higher than about frequency ).
< F
<
=
9^] Y ]R
Exercise 7.2 Suppose that the nominal model of a plant is
B8 ]H
and the uncertainty in the model is parameterized by multiplicative uncertainty with the weight
, ^] P ]{ ]H
Call the resulting set . Now find the extreme parameter values in each of the plants (a)-(g)
P ,
below so that each plant belongs to the set . All parameters are assumed to be positive. One
o o \
approach is to plot
in (7.25) for each
( , etc.) and adjust the
parameter in question until just touches .
. < _ c x
0
B
(a) Neglected delay: Find the largest for (Answer: ).
(b) Neglected lag: Find the largest for
(Answer:
c
c
).
(c) Uncertain pole: Find the range of for (Answer: to ).
L
(d) Uncertain pole (time constant form): Find the range of for
(Answer:
c
to ).
q }C }C
(e) Neglected resonance: Find the range of for
c } c
(Answer:
5 i c x
to ).
S
V (Answer: ).
5 cc (Answer: ). These
(f) Neglected dynamics: Find the largest integer for
(g) Neglected RHP-zero: Find the largest 0 for
results imply that a control system which meets given stability and performance requirements
, \ xx .
for all plants in , is also guaranteed to satisfy the same requirements for the above plants
]^ ]7 (and with everything else
L ]
]
). (Answer: Same as above).
(h) Repeat the above with a new nominal plant
the same except
Solution.
(a)
_ D . When . x just touches , .
(b)
Y D D . When 0 `f just touches , .
(c)
D . When 5 7 , .
(d)
Y D D . When 7 , .
(e)
D D D} }C C } D D ~ } . When
i f , . Note the
} } C x C , i.e. the middle term is
^]] not ^]B{{ than the answer given in thec book} is right.
answer given in the book is not right. If
c
(f)
i5 D D V . When
aS
7 .
]
,
(g) Y 5 D . When 0 , , .
(h) If we change s for all plants to are the same as (a) to (g).
Thus the answer is the same ~c asthenabove.
9
Q , Q Y
Exercise 7.4 Represent the gain uncertainty in (7.54) as multiplicative complex uncertainty
with nominal model
, (rather than used above).
9 V
9 V
(a) Find and use the RS-condition to find
8 . Note that no iteration
9 V
9 V
is needed in this case since the nominal model and thus is independent of .
(b) One expects
8
to be even more conservative than
8 since this uncertainty
description is not even tight when is real. Show that this is indeed the case using the
numerical values from Example 7.10.
I , for , and ,
,
(Hint: Use
R I I .)
, , o , , k k ,
I , d I k , I . So, if , I and , I then I I Y9 YF , and if
Solution.
, v and , then I Y Y .
^] Y - ]R- _ ]c
Exercise 7.7 Consider a true plant
9^ ]
. and sketch the additive uncertainty weight when the nominal model is
(a)- ]Derive
H
7 {
^ ] Y 9 8] 9
(b) Derive the corresponding robust stability condition.
(c) Apply this test for the controller and find the values of that yield stability.
Is this condition tight?
Solution. _ c ] 6 ]B ` ]7 and ,H^ ] 7
c
Exercise 7.9 Consider again the system in Figure 7.18. What kind of uncertainty might
,
and represent? c
Solution.
may seem to represent some additive uncertainly, but actually it
c c
represents inverse multiplicative (pole) uncertainty. (This may be
the point where the signal goes to
to just after the block for ).
h seen by moving
9 X 9 V 9 V
(7.103)
9 9 X (7.102)
as v&
(a) Derive
J 9and
(7.103). (Hint: The gain variation in (7.101) can be written exactly
(b) Show that the form in (7.102) does not allow for v
.)
9 .
(c) Discuss why (b) may be a possible advantage.
9
9 m 9 V
9 X m J 9 3_ : 9 9 V , and if 3_ :H m
Solution.
(a) Let v
. If then v then
v . Solving the equations:
9V J 2 99 XX
9 V H J 2
gives that
9 V 9V
J 9 V 9 V
9 V 9 V
9X 9 V 9 V
{
9 ^ ] Y 9 v ^] and letting , J leads to (7.102) and (7.103).
Inserting v into v
9X 9 R J I . But this is impossible when m .
(b) Since
Thus v
9
, v
is not allowed.
implies
(c) Usually the gain CANNOT be zero physically, so using an uncertainty description where
9
this is not possible may be an advantage. However, note that the inverse gain form may allow
for v being infinite which is also impossible physically.
<
,F
" =
= = =
&
v
Exercise 8.1 The uncertain plant
where
in (8.7) may be represented in the additive uncertainty form
r is a single scalar perturbation. Find and .
,
N , O 2m
Solution.
N , , O N , ,
Solution.
O
F
Exercise 8.4 Show in detail how in (8.29) is derived.
G Fm (~
Solution.
N O N O GF7 (~ k F
N O N O
N 9 GF< GFm Y O
where @
GFH the identities F( F7 @ GF ( , and
Fm GF ( . Using
, it finally yields the desired matrix .
Exercise 8.8 Find for the uncertain system in Figure 7.18.
Solution. , , ,
, F , F , k
Exercise 8.10 Find the interconnection matrix for the uncertain system in Figure 7.18.
What is ?
, GF , GFk
3N , GFk (( , GFk (( O
Solution.
, G7F ( .
So,
Exercise 8.12 -structure for combined input and output uncertainties. Consider the
block diagram in Figure 8.8 where we have both input and output multiplicative uncertainty
Q WQ Y and Q Q(Y . Collect the perturbations into g
(8.33)
where
-structure in Figure 8.3 Show that
and rearrange Figure 8.8 into the
N O N O N O (8.34)
Solution. Since !
GF7 ( and , it yields:
N O N O G F7 (~ k
GF7 G F7
N 9GF7 ( G F7 (( O
N O N O N O
\H_]
?
GF X X G Q F&Y
Exercise 8.13 Consider combined multiplicative and inverse multiplicative uncertainty at
Q X
the output, v , where we choose to norm-bound the
combined uncertainty,
. Draw a block diagram of the uncertain plant,
and derive a necessary and sufficient condition for robust stability of the closed-loop system.
Solution. A block diagram of the uncertain plant is shown in Fig. 8.13. From the diagram it
+ + + +
-
Fig. 8.13 Combined output uncertainties for Exercise 8.13
X
N
can be obtained that
O
Q WQ Y {
According to (8.64) the RS condition is:
Exercise 8.15 (continued
from Example 8.7). (b) For in (8.98) and a diagonal show
U P
_ %
that using the lower bound
) F (which is always
exact). (Hint: Use 2 (the blocks in are unitary scalars, and we may fix
in (8.98) and a diagonal show that
one of them equal to ).) (c) For
using the upper bound r ~ (which is exact in this case since has
two blocks).
9b
Solution. (b) The matrix
is singular and its non-zero eigenvalue is then given by
its trace (Fact 1 in Appendix A.2.1). We then get
YrU P N _ _ O rU _
2
The sum of two complex numbers is maximized when they are in the same direction, and since
P
`K%
we have freedom to select the direction (phase) of the first term, we get .
. Since
(c) Use is a singular matrix we have from (A.37) that
Y[ N L
K4 O K4 K
(8.100)
U P N :KK O
N 1K O
K _
U N _ 22 1KK O
U _ 2 K
K F _ F
, ! , we have
P P Y K (which is enerally smaller).
The second equality has been proven in Example 8.7. When
essentially no degrees of freedom and
When is full, is a general unitary matrix, and which is generally larger.
So the answer for the last question is also NO.
If (8.94) were true for any structure of then it would imply P
P .8.20
Exercise
2x
Show by a counterexample that this is not true.
Solution. Select N ] O , N O . Then P 9m ] , < ] , and
P , so P 97 P .
9^ ] in (8.108) which is ill-conditioned with P
Exercise 8.22 Consider the plant
at all frequencies (but note that ). With an inverse-
based controller
^ ] 9
^ ]
c , compute for RP with both diagonal and full-block
the RGA-elements of are all about
input uncertainty using the weights in (8.133). The value of is much smaller in the former
case.
Solution. The
curves for diagonal and full-block input uncertainty are shown in Fig. 8.22.
The value of for diagonal uncertainty is much smaller than that of the case discussed in
Section 8.11.3.
Exercise 8.24 Explain why the optimal -value would be the same if in the model (8.144) we
changed the time constant of [min] to another value. Note that the -iteration itself would
be affected.
Solution. The optimal -curve as a function of frequency is flat, i.e. the optimal is almost
a constant ( -value) over a wide rang of frequencies. In the model (8.144), the change of
the time constant can be treated as a frequency scaling change, while the optimal -value is
independent of frequency. Thus it would be the same.
\H_]
?
4.5
3.5
2.5 Full
ssv
1.5
1
Diag
0.5
0 3 2 1 0 1 2
10 10 10 10 10 10
Frequency
Fig. 8.22. curves for diagonal and full uncertainty in Exercise 8.22
Exercise 9.1 Show that the closed-loop objectives 1 to 6 can be approximated by the open-
loop objectives 1 to 6 at the specified frequency ranges.
( 2 GFP( ( 6 (
Solution.
1. When
Y , . So the close-loop objective 1, making
(
( GFu ( 6
small, can be approximated by the open-loop objective 1, making large.
6
( 6
2. When , . So the closed-loop objective 2, making
( 2 GF7 ( 6 6 F
small, can be approximated by the open-loop objective 2, making
(
small.
3.
4. (
g GF7 (
. So
. Thus,
small
small small.
.
( Y
5. See 4.
( GF7 ( 6 (
6. . Thus, small small.
formulate a standard problem, draw the corresponding control configuration and give
expressions for the generalized plant .
Solution. The control configuration is shown in Fig. 9.6. From the figure, the generalized
plant can be derived as:
)
F k
Exercise 9.8 Design an
Y
loop-shaping controller for the disturbance process in (9.75)
using the weight ^] -]
in (9.76), i.e. generate plots corresponding to those in Figure 9.18. Next,
c
3 F
repeat the design with (which results in an initial which would yield
closed-loop instability with ). Compute the gain and phase margins and compare
the disturbance and reference responses. In both cases find and use (2.45) to compute the
maximum delay that can be tolerated in the plant before instability arises.
_
c
+
-
<
Fig. 9.6 Control configuration for
optimization problem
Y ^] - ]
Solution. The loop shapes, disturbance response and reference response of the system using
] 3
the loop-shaping controller designed with are shown in Fig. 9.8. It is
] { c
shown that the initial would yield closed-loop instability with (dashed-line). The
gain margin of this design is , phase margin is . rad/s. Thus the maximum
Exercise 9.10 Show that the Hanus form of the weight in (9.109) simplifies to (9.108)
delay is PM s which can be tolerated in the plant before instability.
Thus,
GFm ~ Y Y Y Y
NY
s Y s N YY Y Y O
O
k 1
Loop shapes
4
10
Gs
GsKs
Magnitude
2
10
0
10
2
10 2 1 0 1 2
10 10 10 10 10
Frequency
Disturbance response Reference response
2 2
1.5 1.5
1 1
0.5 0.5
0 0
0.5 0.5
1 1
0 1 2 3 0 1 2 3
Time Time
< <
K B J B .
measurements.
B
b
J B K B K.
Solution. For this system, we have that
J and L m] . eC
and F v
Also, 9
So,
E B f K% E mB B ] K% E f B K
The state-space model of the system can be written as:
E : m B, L f]
So, the linearized models for the three measured variables are:
E : f , L
L
,
E:
Following the singular value procedure (page 400):
] B `
] J such that a unit
1. The input is scaled by the factor
deviation in each input from its optimal value has the same effect on the cost function .
_ X
X maximum
2. The
&
_ L X . Then, forA ] E ,_ 2 8 C
L setpoint error due to variations in disturbances
_
d
d
m
B
f
is given as
"
eC 8 f
and similarly,
e e and
.
8 C 8 C is M
.
4. The expected variation (span) for A9 E is
: o
5. The scaled gain matrices and the worst-case losses are
A< E d] J ;
A< E : \ | } 9
d{B{ J ;
A< E : |} df
J
|} }
These scaled gain matrices can be plotted for different values of r (see Figure 10.2).
4
y1
3.5 y
2
y3
3
2.5
1.5
0.5
0
0 5 10 15 20
r
E E J
So, we can conclude that it is better to choose
i EE iswhen J 8 f{ . If 8f{ J { x ,
then we can choose both or . For , the best choice.
Q QWY . (Hint: Rearrange (5.31) on page 178 using the definition of pole
achievable value of
Exercise 10.4 Show that for a system with a single unstable pole, (10.23) represents the least
vectors.)
%
9
^ ] with the RHP-pole at ] %
%P ^^ ]P] % 9 %u
Also, the stable version of mirroed across the imaginary
axis is:
c %
:
Then, we have
Q % Q Q d E]v B v ]]R % ^] Q - v
c % % c
or
%uY E v d B v
Q Q d Q Q
c %
d
QWY& c %u Q E v % Q d 8Q B v Q
Then,
Q
Exercise 10.5 For systems with multiple unstable poles, the variables can be selected
sequentially using the pole vector approach by stabilizing one real pole or a pair of complex
poles at a time. Usually, the selected variable does not depend on the controllers designed in
the previous steps. Verify this for each of the following two systems:
^ ] Y ^ ] d vsvslo v`o m -m v l-v ^ ] Y ^] d vsvsl o v
o v v`m -v
^ ] Y v2Ujv2 l
l v2Ujl-m
(Hint: Use simple proportional controllers for stabilization of %K
and evaluate the effect of
change of controller gain on pole vectors in the second iteration.)
^ ]
Solution. For , the absolute pole vectors are:
2
y
y 1
0 y
2
2
0 0.01 0.02 0.03 0.04 0.05
(a) G (s)
1
1.5
y
1
1
y
0.5
y2
0
0.5
0 0.01 0.02 0.03 0.04 0.05
(b) G (s)
2
Exercise 10.9 Draw the block diagrams for the two centralized (parallel) implementations
corresponding to Figure 10.10 (in the book).
Exercise 10.11 Process control application. A practical case of a control system like the
E
J E B
one in Figure 10.13 is in the use of a pre-heater to keep the reactor temperature at a given
x
value . In this case may be the outlet temperature from the pre-heater,
J B
the bypass flow
(which should be reset to , say % of the total flow), and
the flow of heating medium
(steam). Make a process flowsheet with instrumentation lines (not a block diagram) for this
heater/reactor process.
Consider the
q FCCofprocess
of a plant. Show that the possible pairings can be eliminated by requiring DIC. (b)
in Exercise 6.17 on page 257. Show that the six possible
pairings can be eliminated by requiring DIC.
f Wf
Solution. (a) Applying (10.78) to (A.83), where the RGA is:
f k
Y m{ { x `f
f f{ k f
iWk {
:
-
+ Plant
+
-
(e) Extra measurements
Plant
Pre-heater
Reactor
it can be seen that outputs 1 and 4 can only be paired with inputs 1 and 4 in 2 possible
combinations, and that output 2 and 3 can only be paired with inputs 2 and 3 also in 2 possible
combinations. Thus only 4 pairings give positive RGA required by DIC and other 20 pairings
can be eliminated.
{ f8f m
(b) The RGA is:
Y {x f8fB {{-f
it can be seen that outputs 1 can only be paired with input 1 and input 3 can only be paired
with output 3. Thus there is only one pairing that gives positive RGA required by DIC and
other 5 pairings can be eliminated.
{
= <
5s is
,
Exercise 11.1 The steady-state gain of a full order balanced system
. Show, by algebraic manipulation, that this is also equal to
the steady-state gain of the balanced residualization given by (11.7)(11.10).
Solution. Using (A.8) and noting in (A.8) is equal to in (11.7), we have
C N
O N O
C N O
Exercise 11.3 Is Theorem 11.3 true, if we replace balanced truncation by balanced
residualization?
] ] ^ ] ^] is a normalized left-coprime
9^] ^] x ^ .] If is axnormalized
Solution. Yes, it is still true. Balanced truncation and balanced residualization are related
by the bilinear transformation
9^]
factorization of then
^ ] x ^] is equivalent
left-coprime factorization
of . Applying Theorem 11.3 to
residualization with .
^ ] x ^] to using balanced
F =
,
F < ,
F
!
<
.
Exercise 12.1 Let
Show that
be a Hermitian matrix (
if and only if
) having the form
N O
(12.9)
Solution. For a complex hermitian matrix ,
, it implies that _ @ < @ 7 , for all
@
@ X @K @ F @ X
in ( is the set of complex numbers).
Let and .
_ @
@u @ x@# @ X X @% @ X @# @ X @ X @ X (@ X
Then,
<
9 0 .
as LMI problems. Verify your results with the Robust Control toolbox command wcgain using
numerical values
Transfer functions and can be easily converted to state-space representation:
Solution.
@K @ B
E @ B
0 , 9 , , and
For :
For :
, 9 , , and
Substituting these , , , and in (12.30) we get the LMI problem.
The results obtained are similar to using Robust Control toolbox command wcgain, i.e.,
both methods give almost the same upper bounds, as can be seen on the next Table.
u
wcgain
LMI 1.5 3.06
;
<
Exercise 13.1 Repeat the -optimal design based on
-iteration in Section 8.12.4 using
the model (13.19).
in (13.9). After 10 iterations,
the resulting controller with 25 states (denoted
Solution. Apply the MATLAB program given in Table 8.2 to
{{ (see Fig. 13.1(a)). The final -curves for RS, NPinandthe RPfollowing)
with controller are shown
gives a peak -value of
in Fig. 13.1(b). It is shown that all requirements are well satisfied. The time response of E
and E to a filtered setpoint change in E , J
B-] is shown in Fig. 13.1(c). both for
the nominal case (solid line) and for 20% input gain uncertainty (dashed line)
1.1
Iter. 1
1 Iter. 2
Iter. 3
Iter. 10
0.9
mu
0.8
0.7
0.6
0.5 3 2 1 0 1 2 3
10 10 10 10 10 10 10
Freqency
0.9
0.8 RP
NP
0.7 RS
0.6
0.5
mu
0.4
0.3
0.2
0.1
0 3 2 1 0 1 2 3
10 10 10 10 10 10 10
Freqency
y1
0.8
0.6
0.4
0.2
y2
0 10 20 30 40 50 60 70 80 90 100