Optimal and Robust Control Advanced Topics With Matlab 2nd Edition 2nd Edition Fortuna Download
Optimal and Robust Control Advanced Topics With Matlab 2nd Edition 2nd Edition Fortuna Download
https://ebookbell.com/product/optimal-and-robust-control-
advanced-topics-with-matlab-2nd-edition-2nd-edition-
fortuna-35085744
https://ebookbell.com/product/optimal-and-robust-control-advanced-
topics-with-matlab-frasca-5145358
Solutions Manual For Optimal And Robust Control Advanced Topics With
Matlab Luigi Fortuna
https://ebookbell.com/product/solutions-manual-for-optimal-and-robust-
control-advanced-topics-with-matlab-luigi-fortuna-56831866
https://ebookbell.com/product/linear-systems-optimal-and-robust-
control-alok-sinha-2475394
https://ebookbell.com/product/stabilization-optimal-and-robust-
control-theory-and-applications-in-biological-and-physical-
sciences-1st-edition-aziz-belmiloudi-auth-2019630
Optimal And Robust Scheduling For Networked Control Systems Stefano
Longo
https://ebookbell.com/product/optimal-and-robust-scheduling-for-
networked-control-systems-stefano-longo-4180368
https://ebookbell.com/product/optimal-and-robust-estimation-with-an-
introduction-to-stochastic-control-theory-second-edition-2nd-frank-l-
lewis-2388190
https://ebookbell.com/product/cooperative-control-of-multiagent-
systems-an-optimal-and-robust-perspective-jianan-wang-11025628
https://ebookbell.com/product/robust-and-optimal-control-a-twoport-
framework-approach-1st-edition-miching-tsai-4634722
Optimal And Robust State Estimation Finite Impulse Response Fir And
Kalman Approaches 1st Edition Yuriy S Shmaliy
https://ebookbell.com/product/optimal-and-robust-state-estimation-
finite-impulse-response-fir-and-kalman-approaches-1st-edition-yuriy-s-
shmaliy-51980664
Optimal and Robust
Control
Optimal and Robust
Control
Advanced Topics with MATLAB®
Second Edition
Luigi Fortuna
Mattia Frasca
Arturo Buscarino
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks
does not warrant the accuracy of the text or exercises in this book. This book’s use or discussion of
MATLAB® software or related products does not constitute endorsement or sponsorship by The
MathWorks of a particular pedagogical approach or particular use of the MATLAB® software.
Reasonable efforts have been made to publish reliable data and information, but the author and pub-
lisher cannot assume responsibility for the validity of all materials or the consequences of their use.
The authors and publishers have attempted to trace the copyright holders of all material reproduced
in this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so
we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.copyright.
com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. For works that are not available on CCC please contact mpkbookspermis-
[email protected]
Trademark notice: Product or corporate names may be trademarks or registered trademarks and are
used only for identification and explanation without intent to infringe.
DOI: 10.1201/9781003196921
Publisher’s note: This book has been prepared from camera-ready copy provided by the authors.
Preface xiii
Symbol List xv
2 Fundamentals of Stability 11
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Controllability Canonical Partition . . . . . . . . . . . . . . 37
3.3 Observability Canonical Partition . . . . . . . . . . . . . . . 39
3.4 General Partition . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5 Remarks on Kalman Decomposition . . . . . . . . . . . . . 47
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
vii
viii Contents
Index 303
Preface
The main aim of this book is to provide for undergraduate and graduate
students, as well as researchers, who already possess the main concepts of
automatic control and system analysis, a self-contained resource collecting
advanced techniques for linear system theory and control design. Selected
theoretical backgrounds are also presented in the book, together with many
numerical exercises and MATLABr examples.
We intend to offer a complete and easy-to-read handbook of advanced
topics in automatic control, including techniques such as the Linear Quadratic
Regulator (LQR) and H∞ control. Large emphasis is also given to Linear
Matrix Inequalities (LMIs) with the purpose of demonstrating their use as a
unifying tool for system analysis and control design.
In presenting the different approaches to control design, the books explic-
itly takes into account the problem of the robustness of the obtained closed-
loop control. Robustness, in fact, represents the capability of a control system
to guarantee the stability in the presence of uncertainty, due to the model itself
or to the use of approximated models, and as such is deemed as particularly
important in view of the practical implementation of the control techniques.
Many books on LQR control and H∞ control have been proposed since
1980. The LMI technique has become well-known in the control community,
and MATLABr toolboxes to solve advanced control problems have been de-
veloped. However, these subjects are often presented for a specialist audience
in materials that are excellent resources for researchers and PhD students.
This book, on the contrary, is oriented to illustrate these topics in an easy
and concise way, using a language suitable for students, yet maintaining the
necessary mathematical rigor.
This book is, therefore, a compendium of many ordered subjects. For spe-
cific proofs, the reader is often referred to the proposed literature. Many ex-
amples and MATLABr based exercises are included here to assist the reader
in understanding the proposed methods. The book can be considered as a
palimpsest of advanced modern topics in automatic control, including an ad-
vanced set of analytical examples and MATLABr exercises. The topics in-
cluded in the book are mainly illustrated with reference to continuous-time
linear systems, even if some results for discrete-time systems are briefly re-
called.
The book is organized into chapters structured as follows. The first chapter
is an introduction to advanced control, the second discusses some fundamental
concepts on stability and provides the tools for studying uncertain systems.
xiii
xiv Preface
The third presents the Kalman decomposition. The fourth chapter is on sin-
gular value decomposition of a matrix, given the importance of numerical
techniques for systems analysis. The fifth and sixth chapters are on open-loop
balanced realization and reduced order models. The seventh chapter presents
the essential aspects of variational calculus and optimal control and the eighth
illustrates closed-loop balancing. The properties of positive-real, bounded-real
and negative-imaginary systems are the subject of the ninth and tenth chap-
ter. In the eleventh and twelfth chapter, the essential aspects of H∞ control
and LMI techniques commonly used in control systems design are dealt with.
The thirteenth chapter is devoted to discuss the class of stabilizing controllers.
The fourteenth chapter reviews some of the problems already discussed in the
book by introducing an approach based on the steady-state solution of a non-
linear dynamical system. Finally, the fifteenth chapter briefly discusses some
fundamental aspects of time-delay systems. The book also includes numerous
examples and exercises, considered indispensable for learning the methodology
of the topics dealt with, and a list of essential references.
This book is targeted at electrical, electronic, computer science, space and
automation engineers interested in advanced topics on automatic control. Me-
chanical engineers as well as engineers from other fields may also be inter-
ested in the topics of the book. The contents of the book can be learned
autonomously by the reader in less than a semester.
Symbol Description
xv
xvi Symbol List
CONTENTS
1.1 Uncertainty and Robust Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The Essential Chronology of Major Findings in Robust Control 9
In this chapter the main concepts related to robust control are introduced.
In particular, the key notions of robustness and uncertainty are presented. In
the formalization of uncertainty, both the structured and unstructured cases
are dealt with. In this preliminary part, the representation of a linear time
invariant system in term of the realization matrix is also given. Some examples
of uncertainty generated with MATLAB examples are reported. The essential
chronology of the recent history of robust control is also outlined.
DOI: 10.1201/9781003196921-1 1
2 Optimal and Robust Control: Advanced Topics with MATLAB r
FIGURE 1.1
Unitary feedback control scheme.
FIGURE 1.2
Feedback control scheme.
the design was related to nominal parameter values). The first goal of robust
control is to ensure that, even with changing parameter values, the system
asymptotic stability is guaranteed.
Above, there was an example of parametric uncertainty about the coeffi-
cients of the transfer function of a linear time-invariant system. Now consider
a nonlinear time-invariant system (with one input and output, for example),
described by the equations:
ẋ = f(x) + g(x)u
(1.1)
y = h(x)u
with x ∈ Rn (state variables), u ∈ R (system input) and y ∈ R (system
output), g : Rn → Rn , f : Rn → Rn and h : Rn → Rn . With robust control,
parametric uncertainty in the model is highlighted by using the parameters
α, β and γ:
ẋ = Ax + Bu
(1.3)
y = Cx + Du
with x ∈ Rn , u ∈ Rm , y ∈ Rp , A ∈ Rn×n , B ∈ Rn×m , C ∈ Rp×n , D ∈ Rp×m ,
coefficients of A, B, C and D vary in certain intervals. What is required for
robust control, once the controller is designed for nominal parameter values,
is that asymptotic stability is guaranteed even when the parameters are not
nominal.
4 Optimal and Robust Control: Advanced Topics with MATLAB r
R̃ = T−1 RT (1.5)
which is clearly a similarity relation.
To verify if a system is minimal, it is possible to calculate the eigenvalues of
matrix R which are represented by λ̄1 , λ̄2 , . . . , λ̄n+m , whereas the eigenvalues
of A by λ1 , λ2 , . . . , λn . These quantities are system invariants, that is, they do
not depend on the state-space representation. The system is minimal if none
of eigenvalues of A coincide with those of R.
There is another type of uncertainty which is more difficult to deal with
than parametric uncertainty. Called structural uncertainty or unstructured un-
certainty, it concerns the model structure. Consider again the example of the
1
process described by P = s2 (s+1) , structural uncertainty takes into account
the possibility that the modelling did not account perhaps for an additional
pole and an additional zero in the process transfer function:
( αs0 + 1)
P =
s2 (s + 1)( αs + 1)
at −1. The conclusion is that the consequences of uncertainty on system order, having
overlooked some dynamics (which could also be a parasite dynamic which triggers in
certain conditions), can be very important.
once the minimum and maximum values of the various coefficients are as-
signed, parametric uncertainty is completely characterized:
am
1 ≤ a1 ≤ a1
M
...
am
n ≤ an ≤ an
M
(1.6)
bm
0 ≤ b0 ≤ bM
0
...
bm
m ≤ bm ≤ bm
M
reduced order models which can allow to deal with the control problem in a
simpler way. It seems clear that this operation introduces an error which can
be evaluated and taken into account in the robust controller design.
Now, let us turn to the case of the linear time-invariant systems described
by equations (1.3). In the next example, we will discuss parametric and struc-
tural uncertainty in the state matrix A and illustrate how the position of the
eigenvalues of this matrix is affected by this uncertainty.
(a) (b)
FIGURE 1.3
Stability of linear systems in presence of parametric uncertainty: (a) distribu-
tion of the parameters k1 and k2 ; (b) eigenvalues of the state matrix A (1.7)
when k1 and k2 are drawn from a normal distribution.
(a) (b)
FIGURE 1.4
Stability of linear systems in presence of parametric uncertainty: (a) distribu-
tion of the parameters k1 and k2 ; (b) eigenvalues of the state matrix A (1.7)
when k1 and k2 are drawn from an uniform distribution in [−0.5, 0.5].
FIGURE 1.5
Stability of linear systems in presence of unstructured uncertainty as in equa-
tion (1.8): eigenvalues of the state matrix A.
CONTENTS
2.1 Lyapunov Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Positive Definite Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Lyapunov Theory for Linear Time-Invariant Systems . . . . . . . . . . . 16
2.4 Lyapunov Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Stability with Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Further Results on the Lyapunov Theory . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.1 Hystorical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.2 Lyapunov Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
This chapter deals with the problem of stability, the main requirement of a
control system. In going over some basics in system analysis, the emphasis
is on some advanced mathematical tools (e.g., Lyapunov equations), which
will be useful below. Particular attention is given to Lyapunov theory for the
case of linear time-invariant systems. The Lyapunov linear matrix equation
is introduced in detail and the concept of positive definite matrix is also dis-
cussed. Some criteria to verify this property are reported and the solution
of Lyapunov equations via vectorization is presented with numerical exam-
ples. The Kharitonov criterion to test the stability of uncertain system is also
introduced. At the end of the chapter, several worked examples are included.
DOI: 10.1201/9781003196921-2 11
12 Optimal and Robust Control: Advanced Topics with MATLAB r
So, nonlinear systems may have more than one equilibrium point. In
addition, each of these equilibrium points has its own stability characteristics.
For example, think of a pendulum. It has two equilibrium points as shown in
Figure 2.1. Only the second of the two equilibrium points is stable. If the pen-
dulum were to start from a position near equilibrium point (a), it would not
return to its equilibrium position, differently from what happens at point (b).
For this reason, in nonlinear systems, stability is a property of equilibrium
points, and not of the whole system. The stability of an equilibrium point
can be studied through the criteria introduced by the Russian mathematician
Aleksandr Mikhailovich Lyapunov (1857–1918).
FIGURE 2.1
Equilibrium points of a pendulum.
1. V (0) = 0
2. V (x) > 0 ∀x ∈ Ω, x 6= 0
finding Lyapunov functions, while for linear systems there is. Effectively, Lya-
punov second criterion for linear systems can be simplified. Before introducing
the criterion, firstly, positive definite matrices are defined.
Given matrix P ∈ Rn×n , a quadratic form can be associated to it: V (x) =
x Px, which is clearly a scalar function of vector x ∈ Rn .
T
P = Ps + Pas
where the symmetric part is
1
Ps = (P + PT )
2
and the antisymmetric part is
1
(P − PT ).
Pas =
2
Clearly, only the symmetric part of a matrix determines if a matrix is
positive definite as:
V (x) = xT Px = xT Ps x
In fact, the asymmetric part does not contribute. This can be seen consid-
ering that:
V (x) = xT Px = p11 x21 + . . . + pnn x2n + (p12 + p21 )x1 x2 + (p13 + p31 )x1 x3 + . . .
. . . + (pn−1,n + pn,n−1 )xn−1 xn
On the other hand, from the definition of Ps one has:
p21 +p12
. . . pn1 +p
p11 2 2
1n
p +p
21 12 p22 pn2 +p2n
2 ... 2
Ps =
.. .. ..
. . .
p1n +pn1 pn2 +p2n
2 2 . . . p nn
VT V = I
VT = V−1
and
P = VΛVT
V (x) = xT Ps x = xT VΛVT x
Consider the state transformation x̃ = VT x, then:
p11 p12
D1 = p11 > 0; D2 = det > 0; . . . ; Dn = det P > 0 (2.3)
p21 p22
16 Optimal and Robust Control: Advanced Topics with MATLAB r
MATLABr Exercise
2.1
1 2 5
Consider P = 2 5 −1 . Since the matrix is symmetric, the Sylvester test can
5 −1 0
be applied to determine if it is positive definite or not.
Let us first define matrix P in MATLAB with the command:
>> P=[1 2 5; 2 5 -1; 5 -1 0]
and then compute D1 , D2 and D3 as follows:
>> D1=det(P(1,1))
>> D2=det(P(1:2,1:2))
>> D3=det(P)
One obtains: D1 = 1, D2 = 1 and D3 = −146. Since D3 < 0, P is not positive definite.
p11 p12
P= .
p12 p22
Solving the Lyapunov equation means solving a system of three equations with three
unknowns. By substituting in equation (2.4), we get:
−2 0 p11 p12 p11 p12 −2 0 −1 0
+ =
0 −5 p12 p22 p12 p22 0 −5 0 −1
4p11 = 1
⇒ 7p12 = 0
10p22 = 1
The three equations with three unknowns can be rewritten in matrix form:
4 0 0 p11 1
0 7 0 p12 = 0
0 0 10 p22 1
1
4
0
This leads to: P = 1 . Matrix P is positive definite, so the system is asymp-
0 10
totically stable.
V̇ (x) = xT AT Px + xT PAx =
Proof (Proof of the necessary part of Lyapunov’s II criterion for linear sys-
tems). The proof of the necessary part of the theorem, that is if the system is
asymptotically stable, then for every positive definite matrix Q there is a posi-
tive definite matrix P which is the solution of the Lyapunov equation (2.4), is
a constructive demonstration introducing the integral solution of the Lyapunov
equation.
Fundamentals of Stability 19
and, since x(t) = eAt x(0), also matrix eAt tends to zero:
lim eAt = 0
t→+∞
A2 t 2 A3 t 3
eAt = I + At + + + ··· ⇒
2! 3!
(A2 )T t2 (A3 )T t3
(eAt )T = I + AT t + + + ··· =
2! 3!
(AT )2 t2 (AT )3 t3 T
= I + AT t + + + · · · = eA t
2! 3!
R ∞ AT t At
The improper integral P = 0 e Qe dt is finite, since each term tends
to zero. Matrix P, because of the properties of Theorem 5, is the integral of a
positive definite matrix for each t and therefore is positive definite. Note that
matrix P is also symmetric (if Q is symmetric).
R∞ T
Given that P = 0 eA t QeAt dt is a positive definite symmetric matrix,
the solution of Lyapunov equation (2.4) can be verified.
By substituting P into the Lyapunov equation we obtain:
Z ∞ Z ∞
T T
AT eA t QeAt dt + eA t QeAt dtA =
0 0
Z ∞
T T
= (AT eA t QeAt + eA t QeAt A)dt =
0
Z ∞ i∞
d AT t At h T
= (e Qe )dt = eA t QeAt = −Q
0 dt 0
lim eAt = 0
t→+∞
with an > 0 is Hurwitz if all the principal minors of the so-called Hurwitz
matrix are greater than zero:
an−1 an−3 an−5 ... 0 0
an an−2 an−4 ... 0 0
0 an−1 an−3 ... 0 0
H= 0 an an−2 ... 0 0
.. .. .. .. .. ..
. . . . . .
0 0 ... ... a1 0
0 0 ... ... a2 a0
AX + XB = −C (2.5)
Fundamentals of Stability 21
where A, B and C are matrices of Rn×n (in the most general case, they are
non-symmetric), and X is a n × n unknown matrix.
Even here, the equation can be vectorized:
x11 c11
x21 c21
x31
M = − c31
.. ..
. .
xnn cnn
2 2
where M is a suitable matrix in Rn ×n (if B = AT and C is a symmetric
matrix, the number of variables reduces to k = n(n+1) 2 and M ∈ Rk×k ).
The existence of a solution is related to the invertibility of matrix M. This
matrix can be obtained with Kronecker algebra. Let us recall therefore some
of the definitions and properties of Kronecker algebra.
The Kronecker product of two matrices A and B of Rn×n is a matrix of
n2 ×n2
R defined by:
a11 B a12 B . . . a1n B
a21 B a22 B . . . a2n B
A⊗B= .
.. ..
.. . .
an1 B an2 B . . . ann B
The matrix eigenvalues can be calculated from those of matrix A and B.
Let us indicate with λ1 , λ2 , . . . , λn the eigenvalues of A and with µ1 , µ2 , . . . , µn
the eigenvalues of B. Then, the n2 eigenvalues of A ⊗ B are:
λ1 µ1 , λ1 µ2 , . . . , λ1 µn , λ2 µ1 , . . . , λn µn
It can be shown that the vectorization matrix is given by:
M = I ⊗ A + BT ⊗ I
with I ∈ Rn×n .
Also the eigenvalues of matrix M are simply linked to those of matrices A
and B. The eigenvalues of M = I ⊗ A + BT ⊗ I are indeed given by all the
possible n2 combinations of sums between an eigenvalue from A and one from
B:
λ1 + µ1 , λ1 + µ2 , . . . , λ1 + µn , λ2 + µ1 , . . . , λn + µn
M is invertible when there are no null eigenvalues:
λi + µj 6= 0, ∀i, j
This leads to answer the question whether a Lyapunov equation has a solu-
tion or not. Formally, the condition above can be expressed with the following
theorem.
22 Optimal and Robust Control: Advanced Topics with MATLAB r
AX + XB = −C
n×n n×n
with A ∈ R ,B∈R and C ∈ Rn×n has only one solution if and only if
λi + µj 6= 0, ∀i, j
where λ1 , . . . , λn are the eigenvalues of A and µ1 , . . . , µn are those of B.
An immediate consequence of this theorem is that the Lyapunov equation
AT P + PA = −Q has only one solution if A has no eigenvalues on the imagi-
nary axis and if there are no pairs of real eigenvalues with equal module and
opposite sign (for example, λ1 = −1 and λ2 = +1), all of which conditions
are verified by the hypothesis of asymptotic stability.
We have seen that vectorizing the Lyapunov equation requires inverting a
matrix of size n2 × n2 (or a matrix of size n(n+1)
2 × n(n+1)
2 ). Even for a system
of order n = 10 this yields a matrix of ten thousand coefficients. This method
is more laborious to calculate than other iterative methods used routinely for
solving Lyapunov equations. Below is the sketch of an algorithm based on a
Schur decomposition for solving Lyapunov equations of type AT P+PA = −Q.
A Schur decomposition able to obtain A = UĀUT with UT U = I and higher
triangular matrix Ā is considered. At this point, all the Lyapunov equation
terms are multiplied by UT left and U right. Thus, ĀT P̄ + P̄Ā = −Q̄ is
obtained with Q̄ = UT QU and P̄ = UT PU. Here, triangular matrix A can
be used to iteratively solve the equation and then to find the solution P from
P = UP̄UT .
1. Define matrix A:
>> A=[-1 0 1 0; 2 0 3 5; 1 10 0.5 0; -2 3 -5 -10];
and identity matrix:
>> MatriceI=eye(4);
c11
c21
c
2. Construct the vector 31
.
..
cnn
Fundamentals of Stability 23
>> c=MatriceI(:)
3. Construct matrix M
>> M=kron(eye(4),A’)+kron(A’,eye(4))
4. Find the solution vector by inverting matrix M
>> Psol=-inv(M)*c;
and obtain matrix P with an appropriate size (4 × 4)
>> P=reshape(Psol,4,4)
0.4727 0.2383 −0.1661 0.1690
0.2383 1.7516 −0.3188 0.8961
One obtains P = −0.1661 −0.3188 −0.0466 −0.1292
.
0.1690 0.8961 −0.1292 0.4980
5. Verify that matrix P solves the problem
>> A’*P+P*A
MATLAB can solve Lyapunov equations more efficiently with the lyap command. In
this case one can use:
>> P=lyap(A’,eye(4))
In conclusion, using command eig(P) to calculate the eigenvalues of P, it can be verified
that P is not a positive definite matrix and that the system is not asymptotically stable
which is the case after calculating the eigenvalues of A (λ1 = −13.0222, λ2 = 4.7681,
λ3 = −1.1229 + 0.8902i, λ4 = −1.1229 − 0.8902i).
am M
0 ≤ a0 ≤ a0
... (2.6)
am
n ≤ a M
n ≤ an
D1 (s) = am n m
n s + an−1 s
n−1
+ aM
n−2 s
n−2
+ aM
n−3 s
n−3
+ ...
M n M n−1 m n−2
D2 (s) = an s + an−1 s + an−2 s + an−3 sn−3 + . . .
m
(2.7)
D3 (s) = am n M
n s + an−1 s
n−1
+ aM
n−2 s
n−2
+ am
n−3 s
n−3
+ ...
D4 (s) = aM
n sn
+ am
n−1 sn−1
+ am
n−2 s n−2
+ aM
n−3 s n−3
+ ...
are stable, then the polynomial D(s) is stable for any parameter in the given
ranges.
D(s) = a3 s3 + a2 s2 + a1 s + a0 (2.8)
with uncertain parameters:
1 = am
3 ≤ a3 ≤ aM
3 =3
2 = am
2 ≤ a2 ≤ aM
2 =4
2 = am
1 ≤ a1 ≤ aM
1 =5
1 = am
0 ≤ a0 ≤ aM
0 =2
and let us study the stability by the Kharitonov criterion.
First of all, we recall that, for a third-order system, the direct application of the Routh
criterion yields the conclusion that stability requires two conditions: 1) all the coefficients
are positive, i.e., a0 , a1 , a2 , a3 > 0; 2) a2 a1 > a3 a0 . Therefore, to apply the Kharitonov
criterion we need to consider the following polynomials
D1 (s) = am 3 m 2 M
3 s + a2 s + a1 s + a0
M
D2 (s) = aM 3 M 2 m
3 s + a2 s + a1 s + a0
m
(2.9)
D3 (s) = am
3 s 3 + aM s2 + aM s + am
2 1 0
D4 (s) = aM 3 m 2 m
3 s + a2 s + a1 s + a0
M
am M m M
2 a1 > a3 a0
a2 a1 > a3 am
M m M
0 (2.10)
aM M m m
2 a1 > a3 a0
am m M M
2 a1 > a3 a0
Since am m M M
2 a1 < a3 a0 , we conclude that the system is not stable in the presence of the
considered parametric uncertainty.
Similarly, one can conclude that the system is stable for the following parametric un-
certainty
1 = am
3 ≤ a3 ≤ aM
3 =3
2 = am
2 ≤ a2 ≤ aM
2 =4
4 = am
1 ≤ a1 ≤ aM
1 =5
1 = am
0 ≤ a0 ≤ aM
0 =2
Fundamentals of Stability 25
D(s) = a5 s5 + a4 s4 + a3 s3 + a2 s2 + a1 s + a0 (2.11)
and uncertain parameters:
1 ≤ a5 ≤ 2
3 ≤ a4 ≤ 5
5 ≤ a3 ≤ 7
4
3
≤ a2 ≤ 52
1
2
≤ a1 ≤ 34
1 ≤ a0 ≤ 2
To apply the Kharitonov criterion the following characteristic polynomials have to be
considered:
Solution
First, determine the characteristic closed-loop polynomial:
uncertainty interval. Indicating Amin = 6, Amax = 10, Bmin = 16 and Bmax = 24, we
need to study the roots of the four polynomials:
5
k=120
4
k=10 k=30
3
2
1
jω
0
−1
−2
−3
−4
−5
−10 −8 −6 −4 −2 0 2
σ
FIGURE 2.2
Locus of the closed-loop eigenvalues for Exercise 2.5.
ẋ = f (x) (2.13)
+
for t ∈ R and with initial conditions x(0) = x0 . Taking into account that
T T
x = x1 x2 . . . xn and f (x) = f1 (x) f2 (x) . . . fn (x) , these
equations may be rewritten as:
ẋ1 = f1 (x1 , x2 , . . . , xn )
ẋ2 = f2 (x1 , x2 , . . . , xn )
(2.14)
...
ẋn = fn (x1 , x2 , . . . , xn )
Since in general the solutions x(t) cannot be obtained in closed form, here
our interest is to provide some important information about the time evolution
of the state variables.
The equilibrium states are obtained by solving the algebraic nonlinear
system f (x) = 0. Suppose to indicate one of such states as x̄. The Lyapunov
theory provides a characterization of the behavior of the solution when a small
perturbation to the equilibrium state is applied.
Let us, first, recall (see Definition 1) that such an equilibrium is said to
be Lyapunov stable if ∀ε > 0 there exists δ = δ(ε) > 0 such that ∀x0 with
k x0 − x̄ k< δ one has that k x(t) − x̄ k< ε for ∀t > 0.
In addition, the stability is asymptotic if the equilibrium point is x̄ stable
and lim x(t) = x̄.
t→+∞
Here it is important to remark that the previous definitions represent local
properties. Let us now consider the case in which the property of asymptotic
stability holds globally, that is, it holds for any initial condition x0 , as formally
expressed by the following definition.
k1 = T f (tk , x(k))
k2 = T f (tk + T2 , x(k) + k21 )
(2.20)
k3 = T f (tk + T2 , x(k) + k22 )
k4 = T f (tk + T, x(k) + k3 )
This method can be extended to consider a recursive equation of the type:
m
X
x(k + 1) = x(k) + bi k i (2.21)
i=1
Example 2.3
Consider the following nonlinear dynamical system representing a rigid spacecraft:
ẋ1 = I23 x2 x3
ẋ2 = I31 x1 x3 (2.23)
ẋ3 = I12 x1 x2
with
I2 −I3
I23 = I1
I3 −I1
I31 = I2 (2.24)
I1 −I2
I12 = I3
where I1 > I2 > I3 are the principal moments of inertia of the spacecraft. For the sake
of example we consider I1 = 3, I2 = 2 and I3 = 1, such that I12 = 1, I23 = 1/3,
I31 = −1.
Let us now analyse numerically the stability property of the equilibrium point x̄ = 0. In
particular, let us calculate the effect of a small perturbation of the equilibrium point.
To this aim, we use the command:
>> [t,x]=ode45(’nonlinsys2’,[0:0.01:100],[0.1 0.2 0.3])
where the function nonlinsys2 is defined as follows:
function xdot=nonlinsys2(t,x)
I23=1/3;
I31=-1;
I12=1;
xdot=[I23*x(2)*x(3);I31*x(3)*x(1);I12*x(1)*x(2)];
end
The solution shows an oscillatory behavior that yields the conclusion that the equilib-
rium state x̄ = 0 is stable, but not asymptotically. This can be appreciated by plotting
the obtained waveform:
>> figure, plot(t,x)
The result is shown in Figure 2.3.
Example 2.4
Consider the following nonlinear dynamical system:
−6x1
ẋ1 = (1+x2 2 + 2x2
1) (2.25)
2(x1 +x2 )
ẋ2 = − (1+x2 )2
1
Fundamentals of Stability 31
0.5
x(t)
x1 x2 x3
-0.5
0 50 100
t(s)
FIGURE 2.3
Time evolution of the state variables x(t) of system (2.24) in Example 2.3.
Example 2.5
Consider now again the nonlinear dynamical system in equation (2.22) discussed in the
MATLAB Exercise 2.6.
Repeating the procedure of the previous exampes, we can verify that in this case the
equilibrium state x̄ = 0 is globally asymptotically stable.
Example 2.6
Consider now the nonlinear dynamical system given by:
equilibrium point x̄ = 0, otherwise it does not. This supports the conclusion that the
equilibrium point x̄ = 0 is locally asymptotically stable.
Notice that the theoretical calculations require the proper choice of a Lya-
punov function, which is a non-trivial step. The simulations are an empirical
method that often provides an immediate picture of the system behavior, but
not a rigorous proof.
Example 2.8
Let us now consider again system (2.25) and study the following Lyapunov function:
x21
V (x1 , x2 ) = + x22 (2.31)
1 + x21
In this case, one obtains:
6x1
T h 2x1
i − (1+x 2 )2 + 2x2
∂V 2x2 1
V̇ = f = (1+x2 2 2(x1 +x2 )
=
∂x 1) − (1+x21)
2 (2.32)
12x2 4x2
= − (1+x21)4 − 2
(1+x2 2
1 1)
Fundamentals of Stability 33
Since V̇ < 0 ∀x, we conclude that the equilibrium point x̄ = 0 is globally asymptotically
stable.
2.7 Exercises
1. Apply the vectorization method to solve the Lyapunov equation:
AT P + PA = −I
with
0 1 −1
A= 2 −5 −1
3 1 −2
s2 + 2s + 2
G(s) =
s4 + a1 + a2 s2 + a3 s + a4
s3
with 1 ≤ a1 ≤ 3, 4 ≤ a2 ≤ 7, 1 ≤ a3 ≤ 2, 0.5 ≤ a4 ≤ 2.
5. Given the polynomial p(s, a) = s4 +5s3 + 8s2 + 8s + 3 with a =
3 8 8 5 , find p(s, b) with b = (b− + − +
0 , b0 ) . . . (b3 , b3 ) so
that the polynomial class p(s, b) is Hurwitz.
2
6. Study the stability of the system G(s) = s4 +q1 ss3 +5s
+3s+2
2 +q s+q
2 3
with
parameters q1 ∈ [1, 3], q2 ∈ [5, 10], q3 ∈ [2, 18].
Discovering Diverse Content Through
Random Scribd Documents
The Project Gutenberg eBook of Hellaassa
This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.
Title: Hellaassa
Language: Finnish
HELLAASSA
Kirj.
Santeri Ivalo
Siis varsin mukava, oikein mainio talo tuo Hellas. Kummako siis jos
N.S:kin vietti vuosipäiväänsä tässä syrjäisessä, hauskassa paikassa.
— Mikäkö N.S. on? Niistä ei saa oikein selvää, kaikista noista
kirjaimista, joita ehtimiseen, milloin mihinkin kapakkaan Helsingissä
kutsutaan kokoon. Seura kai se oli jonkinlainen: Nouseva Sukupolvi,
Nuoret Suomalaiset, — mikä lie ollut. Ylioppilaita siihen vain kuului,
sen saattoi päättää eräänä syysiltana Hellaan etehisessä niistä
lyyryotsa-samettilakeista, joita siellä tungoksessa kipperästi riippui
täyteen ahdetuista naulapuista. Sillä siksi illaksi oli N.S. kutsuttu
kokoon Hellaaseen.
— Lisää olutta!
— Entäpä sisältö?
Mutta lähtö siitä tuli röökynälle. Paha metakka alkoi kuulua isosta
pöydästä.
— Entä sinä…
Kiire siinä oli Annilla tyydyttää kaikkien tarpeita. Taas kun hän
kantoi kaksin käsin lasikuormaa pöydälle rikkonaisten sijaan, niin
siinä ovella juuri hoiperteli eräs pitkä partaniekka häntä
hätyyttämään. Leuasta tarttui tyttöä kiinni ja suudella loiskautti.
Minkäs se teki tyttö siihen, kädet kun olivat kiinni. Mutta niin se
näytti hänen kasvoistaan päättäen, että itku oli herkässä
herahtamaan.
— Onhan tuo vähän surullista. Eikä tyttö näytä olevan vielä kuin
tuskin 17-vuotias.
— Välttämättömyydelle!
Älä sinä tyttö vielä itke itke sinä sitten vasta, kun toisella
jalalla rukkia poljet ja toisella liekutat lasta.
II.
— Mitä se Lassi valehtelee siinä? Eljas löi Lassia olalle, niin että
rillit taaskin tipahtivat.
Iloiseksi oli mennyt Annin katse, kun heidän tulevan näki. Sydän
se näet on kapakkatytölläkin, nuorella varsinkin; Eljas oli pulska
poika, ja häntä oli Anni muistellut N.S:n vuosipäivästä asti. Heti
viivähtikin Anni sen pöydän tienoilla, johon vastatulijat olivat
istahtaneet, ja alkoi juttelun.
— Näitäkö 50 pennin?
— Älähän toki Lassi enää, johan tässä nyt kerraksi, esteli Antti,
joka sääliskeli huonetoverinsa huomispäivää.
ebookbell.com