Professional Documents
Culture Documents
Ebook Robust Engineering Designs of Partial Differential Systems and Their Applications 1St Edition Bor Sen Chen Online PDF All Chapter
Ebook Robust Engineering Designs of Partial Differential Systems and Their Applications 1St Edition Bor Sen Chen Online PDF All Chapter
https://ebookmeta.com/product/systems-immunology-and-infection-
microbiology-1st-edition-bor-sem-chen/
https://ebookmeta.com/product/evolutionary-equations-picards-
theorem-for-partial-differential-equations-and-applications-1st-
edition-christian-seifert/
https://ebookmeta.com/product/fundamentals-of-data-engineering-
plan-and-build-robust-data-systems-joe-reis/
Partial Differential Equations 3rd Edition Emmanuele
Dibenedetto
https://ebookmeta.com/product/partial-differential-equations-3rd-
edition-emmanuele-dibenedetto/
https://ebookmeta.com/product/fourier-analysis-and-partial-
differential-equations-1st-edition-jose-garcia-cuerva/
https://ebookmeta.com/product/building-integrated-photovoltaic-
thermal-systems-fundamentals-designs-and-applications-1st-
edition-huiming-yin/
https://ebookmeta.com/product/building-integrated-photovoltaic-
thermal-systems-fundamentals-designs-and-applications-1st-
edition-huiming-yin-2/
https://ebookmeta.com/product/practical-applications-of-
evolutionary-computation-to-financial-engineering-robust-
techniques-for-forecasting-trading-and-hedging-iba/
Robust Engineering
Designs of Partial
Differential Systems and
Their Applications
Robust Engineering
Designs of Partial
Differential Systems and
Their Applications
Bor-Sen Chen
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
publisher 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 storage 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
mpkbookspermissions@tandf.co.uk
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: 10.1201/9781003229230
Typeset in Times
by codeMantra
Contents
Preface.................................................................................................................. xi
Author ................................................................................................................. xv
PART I Background
xi
xii Preface
xv
Part I
Background
1 Introduction to Partial
Differential Systems
∂y ∂y ∂2 y ∂2 y
F x1 , x 2 , , x N , y, , , , 2 , , = 0 (1.1)
∂x1 ∂x2 ∂x1 ∂ x1 ∂ x 2
∂2 y ∂2 y ∂2 y ∂y ∂y
2 + cb + cc 2 + cd + ce
ca + cf y = c (1.2)
∂x1 ∂ x1 ∂ x 2 ∂x 2 ∂x1 ∂x 2
where the coefficients ca, cb, cc, cd, ce, and cf are all allowed to be functions of
(x1,x2), and the inhomogeneity or forcing function c = c(x1,x2).
The classification of the second-order 2-dimensional PDE equation (1.2) is
based upon the possibility of reducing the equation (1.2) by a coordinate transfor-
mation to canonical or discriminant form at a point, such as [1]
DOI: 10.1201/9781003229230-2 3
4 Robust Engineering Designs of Partial Differential Systems
∂2 y ∂2 y
− =0 (1.4)
∂x12 ∂x 22
has discriminant 4 > 0, and is hyperbolic; the heat equation
∂2 y ∂y
− =0 (1.5)
∂x12 ∂x 2
has discriminant 0 = 0, and is parabolic; the Poisson equation [3]
∂2 y ∂2 y
+ = −c (1.6)
∂x12 ∂x 22
has discriminant −4 < 0, and is elliptic.
Obviously, it is necessary to specify a boundary condition if the solution of
PDE is to be determined. The two most important kinds of boundary conditions
are [1–3]
∂y( x1 , x 2 , , x N )
= 0, ∀x1 , x 2 ,, x N ∈∂Ω (1.7)
∂nbase
∂
where is the outward normal derivative, nbase is the normal vector to the
∂nbase
boundary ∂Ω. In considering time t, the initial condition y(t = 0, x1 , x2 ) is neces-
sary as well. The mathematical problem to solve such PDE is said well-posed if it
satisfies the following fundamental properties [1–3]
The first property is an obvious logical condition. The same can be said about
the uniqueness property. In order to really reflect the physical problem that has a
Introduction to Partial Differential Systems 5
unique solution, the mathematical problem must have a unique solution. For the
last property, if the solution is to have physical significance, a small change in the
initial data must produce a small change in the solution. The data in a physical
problem are normally obtained from experiment, and are approximated in order
to solve the problem by numerical or approximate methods. It is essential to know
that the process of making an approximation, such as the finite difference approx-
imation, to the data produces only a small change in the solution. These proper-
ties for the oncoming analysis discussed in the following sections are assumed
for simplicity. Surely, when simulating the PDE examples by using the numerical
approximate method, i.e., the central finite difference scheme, these properties
are guaranteed.
∂y( x ,t)
= D∆y( x , t) + f ( y( x , t)) (1.8)
∂t
∂2 y1 ( x ,t) ∂2 y1 ( x , t ) ∂2 y1 ( x ,t)
+ ++
∂x1 2
∂x2 2
∂x N2
∂2 y2 ( x ,t) ∂2 y2 ( x, t ) ∂2 y2 ( x , t )
+ + +
D∆y( x , t) = [dij ]ny × ny ∂x12 ∂x22 ∂x N2 (1.9)
∂2 yny ( x ,t) ∂2 yny ( x, t ) ∂2 yn y ( x , t)
+ + +
∂x12 ∂x22 ∂ x N2
∂
where is the outward normal derivative, n is the normal vector to the boundary
∂n
∂Ω, and the initial condition is y( x , 0) y0 ( x ).
6 Robust Engineering Designs of Partial Differential Systems
y( x , t ) = 0, ∀x ∈∂Ω (1.10)
∂ y( x , t )
ii. Neumann boundary condition (N): = 0, ∀x ∈∂Ω
∂n
Remark 1.1.1
∂y
=0
∂t
∂y
Therefore, for the linear diffusion equation = ∆y, we have
∂t
∆y = 0
which is the Laplace equation. The solution depends on the boundary conditions
∂2 y
in equation (1.10). In the 1-dimensional case, , which means the solution
∂x 2
is of the form ax + b, where a, b are constants. The solution can be computed
Introduction to Partial Differential Systems 7
∂y
= −∆y
∂t
This is an unstable ill-posed process. By changing the sign of the time differen-
tiation ∂t → − ∂t, one can view this also as reversing in time the forward diffu-
sion process. Any chemist knows that diffusion is a smoothing process since the
concentration of a substance tends to flatten out. Going backward, the situation
becomes more and more chaotic. Hence, one would not expect well-posedness
of the backward-in-time problem for the diffusion equation. That is, going from
a blurred signal to a sharp one (not a physical process). To avoid the possible
ill-posedness of the concerned PDS like equation (1.8), the positive semidefinite
diffusion coefficient matrix D is commonly used.
When applying a control force to the PDS equation (1.8), we have
∂y( x ,t)
= D∆y( x , t) + f ( y( x , t)) + Bu( x , t) (1.11)
∂t
∂y( x ,t)
= D∆y( x , t) + f ( y( x , t)) + Bu( x , t) + En n( x , t ) (1.12)
∂t
∂y( x ,t)
= D∆y( x , t) + f ( y( x , t)) + Bu( x , t) + En n( x , t ) (1.13)
∂t
z( x , t) = h( y( x , t)) + Em m( x , t) (1.14)
where h(⋅) is a smooth measure function with h(0) = 0 and the rest of settings are
the same as the PDS equation (1.12). z( x , t) [ z1 ( x ,t ), z2 ( x , t), , znz ( x, t )]T ∈ nz
represents the measurement output affected by the measurement noise
m( x , t) [m1 ( x ,t ), m2 ( x , t), , mnm ( x , t )]T ∈ L2 (Ω × + ; nm ) with the influence
matrix Em ∈ nz × nm . The control designs for the aforementioned PDSs in equa-
tions (1.11) and (1.12), and the state observation and control designs in equations
(1.13) and (1.14), are quite interesting topics.
8 Robust Engineering Designs of Partial Differential Systems
Ay = f , in Ω (1.15)
B( y) = 0,on ∂Q (1.16)
yn = y0 ( x1 , x 2 ) + ∑ a φ (x , x )
i=1
i i 1 2 (1.17)
Introduction to Partial Differential Systems 9
which is called the Galerkin solution of equations (1.15) and (1.16), where
φi ( x1 , x 2 ) ∈ L2 (Ω; ny ) are known functions, y0 is introduced to satisfy the bound-
ary conditions, and the coefficients ai are to be determined. Substituting equation
(1.17) into (1.15) gives a nonzero residual Rn
n
In this method, the unknown coefficients ai are determined by solving the follow-
ing equations [3,4]
Rn ,φ j n
L2 ( Ω; y )
= 0, for j = 1, 2,, n (1.19)
∑a
i=1
i Aφi ,φ j
(n
L2 Ω; y )=− Ay0 ,φ j
( n
L2 Ω; y ) (1.20)
yn = ∑ a φ (x , x )
i=1
i i 1 2 (1.21)
0,i ≠ j
φi , φ j
( n
L2 Ω; y ) = 1,i = j
(1.22)
∑aφ i i Aφi ,φ j = f ,φ j
( n
L2 Ω; y ) (1.23)
i =1 ( n
L2 Ω; y )
which is, by equation (1.22),
ai = f ,φi (1.24)
( n
L2 Ω; y )
10 Robust Engineering Designs of Partial Differential Systems
yn = ∑
i=1
f , φi
(n
L2 Ω; y ) φi ( x1 , x2 ) (1.25)
Evidently, the Galerkin solution (1.21) is just the finite Fourier series solution.
When n goes to infinity, we will obtain the perfect approximation
n
y = lim yn = lim
n→∞ n→∞ ∑
i=1
f , φi
(n
L2 Ω; y ) φi ( x1 , x2 ) (1.26)
Therefore, the perfect Galerkin solution like equation (1.26) results in the
infinite-dimensional (n → ∞) problem. That is, we have to use infinitely many
equations in the linear combination form to obtain the perfectly exact solution
of PDEs or PDSs. This Galerkin method makes it possible simulate solutions
of PDEs and PDSs with sufficient large number n. To guarantee the well-
posedness of the PDEs or PDSs with the Galerkin method, both the property of
boundedness
Ayn ,φ j ≤ c1 Ayn φj
( n
L2 Ω; y ) ( n
L2 Ω; y ) ( n
L2 Ω; y ) , ∀j (1.27)
2
Ayn ,φ j ≥ c2 yn (1.28)
( n
L2 Ω; y ) ( n
L2 Ω; y )
for some constants c1 > 0 and c2 > 0, are necessary. These two conditions imply
well-posedness by the Lax-Milgram theorem.
Here, we take a kind of Poisson equation for example to find the approxi-
mate solution by using the Galerkin method. Consider the following Poisson
equation:
∂2 y ∂2 y
+
∂ x12 ∂ x 22
= −1,in Ω = {( x , x ) x
1 2 1 < a, x 2 < b } (1.29)
yM ( x1 , x 2 ) = ∑ ∑
m=1,3,5 n=1,3,5
amnφmn ( x1 , x2 ) (1.31)
Introduction to Partial Differential Systems 11
where
mπx1 nπx 2
φmn ( x1 , x 2 ) = cos cos (1.32)
2a 2b
M M
m2π2 n2π2
RM = AyM + 1 = −
m =1
∑∑ n =1
4 a 2 +
4 b
2 amnφ mn +1
(1.33)
abπ 2 k 2 l 2 16ab
= a − 2 (−1)( k + l ) /2−1 = 0
2 + 2 kl (1.34)
4 a b π kl
or
2 ( k + l ) /2 −1
8ab (−1)
akl = 2 (1.35)
π (b 2 k 2 + a 2 l 2 )
2 M M
(−1)( m + l ) /2−1φmn
8ab
yM ( x1 , x 2 ) = 2
π ∑ ∑
m=1,3,5 n=1,3,5
(b 2 m 2 + a 2 n 2 )
(1.36)
∂y
= y + ay + bu (1.37)
∂t
where y ∈ D ( A ) is a constant function, u ∈ is a constant control input, a ∈,
and b ∈. With the Hilbert-Schmidt theorem, the solution of equation (1.37) can
be represented as [3]
∞
y= ∑ y (t)φ (x)
l=1
1 l (1.38)
12 Robust Engineering Designs of Partial Differential Systems
0, k ≠ l
φk ( x ),φl ( x ) = (1.39)
1, k = l
L2 ( Ω; )
and constitute a set of bases for L2 (Ω; ). λl is ordered so that Re {λl +1 } ≤ Re {λl },
where Re {λl } denotes the real part of λl . Hence, we have
∂y ∂yl (t )
,φl ( x) = , ul (t ) u,φl ( x ) (1.40)
∂t L2 ( Ω; ) ∂t L2 ( Ω;)
and
∆y,φl ( x ) L2 ( Ω; )
= λl yl (t) (1.41)
∂yl (t )
= λl yl (t ) + ayl (t ) + bul (t ), for l = 1, 2, (1.42)
∂t
Then, we collect all yl (t ) to construct the following ordinary differential system
(ODS)
dimensional problem of the ODS, based on the property of the eigenvalues λl and
the orthonormal eigenfunctions φl ( x ), the system state of the PDS equation (1.37)
should be partitioned into a finite-dimensional slow subsystem and an infinite-
dimensional fast residual subsystem. Hence, we define
y( x ,t ) = yc ( x , t ) + yr ( x ,t) (1.44)
where
nc ∞
yc ( x ,t ) ∑l=1
yl (t )φl ( x ) and yr ( x ,t ) ∑ y (t)φ (x)
l = nc +1
l l (1.45)
Ac Acr Bc 0
A= and B = (1.48)
Arc Ar 0 Br
FIGURE 1.1 An example of finite difference grids of the central finite difference scheme
on the 2-dimensional spatial domain Ω.
grids (Figure 1.1, e.g., the central finite difference scheme on the 2-dimensional
spatial domain Ω).
At each node yk ,l , each derivative is approximated by an algebraic expression
which references the adjacent nodes, i.e., yk −1,l , yk +1,l , yk ,1− l and yk ,1+ l . A system of
algebraic equations is obtained by evaluating the previous step for each node and
the system is solved for the dependent variable. For clarity, we illustrate the prin-
ciple of the FDM on the 1-dimensional spatial domain Ω, for example. The system
of algebraic equations with the sums and differences of algebraic function values
is based on the famous Taylor Series from elementary calculus [5]
∞
( x − h) ( x − h) 2
∑ (x −n!h)
n
y( x ) = y( h ) + y '(h) + y ''(h) + = y( n ) (h) (1.49)
1! 2! n=0
where y( n ) denotes the derivative taken n times of y with respect to the space vari-
able x, n! denotes the factorial of n, and h is denoted as an increment in x and is
assumed positive. Now, we replace x and h in the previous formula (1.49) by x + h
and x, respectively. Then, we have
∞
h2
∑ hn! y
n
h
y( x + h) = y( x ) + y '( x ) + y ''( x ) + = (n)
(x) (1.50)
1! 2! n=0
For the treatment of many problems, it is convenient to take only the first two
terms of the right hand side of the previous equation,
where the expression O(h 2 ) represents the error of the approximation, which is
proportional to h 2 , and O(h 2 ) is the famous big O notation that describes the limit-
ing behavior of a function when the argument h 2 tends towards a particular value
or infinity. From this relationship, we can easily define what is known as the first
dy
order, forward finite difference scheme to , i.e.,
dx
Introduction to Partial Differential Systems 15
dy y( x + h) − y( x )
= + O ( h) (1.52)
dx h
dy
Likewise, we can define the first order, backward finite difference scheme to
dx
dy y( x ) − y( x − h)
= + O (h) (1.53)
dx h
by
h2
y( x + h) = y( x ) + hy '( x ) + y ''( x ) + O(h 3 ) (1.55)
2
h2
y( x − h) = y( x ) − hy '( x ) + y ''( x ) + O(h 3 ) (1.56)
2
By subtracting equation (1.56) from (1.55), we can obtain the first order, central
dy
finite difference scheme to , i.e.,
dx
dy y( x + h) − y( x − h)
= + O(h 2 ) (1.57)
dx 2h
The central finite difference scheme obviously has the best approximation since
the increment h is always chosen as a small positive number which is definitely
far less than one (0 < h 1). Therefore, the central finite difference scheme is
normally used to simulate solutions of PDEs and PDSs; thus, the illustration of
the followings focuses on the FDM with the central finite difference scheme. As
dy
to the second order, central finite difference scheme to , we take up to the forth
dx
order as follows:
h2 h3
y( x + h) = y( x ) + hy '( x ) + y ''( x ) + y '''( x ) + O(h 4 ) (1.58)
2 6
h2 h3
y( x − h) = y( x ) − hy '( x ) + y ''( x ) − y '''( x ) + O(h 4 ) (1.59)
2 6
and
d2y
Now, the second-order central finite difference scheme to 2 can be obtained
dx
as follows:
d 2 y y( x + h ) − 2 y( x ) + y( x − h )
= + O(h 2 ) (1.61)
dx 2 h2
Hence, for the numerical solutions of PDEs and PDSs on the 1-dimensional spa-
tial domain Ω, one can utilize the following approximation:
∂ y( x , t ) y( x , t + ht ) − y( x , t )
= + O(ht ) (1.62)
∂t ht
∂ y( x , t ) y( x + hx , t ) − y( x − hx , t )
= + O(hx2 ) (1.63)
∂x 2hx
∂2 y( x , t ) y( x + hx ) − 2 y( x ) + y( x − hx )
= + O(hx2 ) (1.64)
∂x 2 hx2
where hx1 and hx2 are the increment of the time and space variables x1 and x 2,
respectively. Moreover, for the numerical solutions of PDEs and PDSs on the
2-dimensional spatial domain Ω, one can use equations (1.63) and (1.64) to obtain
the following:
( ) (
∂2 y (( x1 , x 2 ) , t ) y ( x1 + hx1 , x 2 + hx2 ) , t + y ( x1 − hx1 , x 2 − hx2 ) , t
=
)
∂ x1 ∂ x 2 4 hx1 hx2
−
( ) (
y ( x1 + hx1 , x 2 − hx2 ) , t + y ( x1 − hx1 , x 2 + hx2 ) , t )
4 hx1 hx2
( hx + hx2 )4
+O 1 (1.65)
hx1 hx2
∆y =
( ) (
y ( x1 + hx1 , x 2 ) , t − 2 y (( x1 + x 2 ) , t ) + y ( x1 − hx1 , x 2 ) , t )
2
h
x1
+
( )
2
( )
y ( x1 , x 2 + hx2 ) , t − 2 y (( x1 + x 2 ) , t ) + y y ( x1 , x 2 − hx2 ) , t t ) + O(⋅) (1.66)
h
x2
where hx1 and hx2 are the increment of the space variables x1 and x 2, respectively.
O(⋅) is proportional to the complicated function constructed from hx21 and hx22 . One
can use the results in equations (1.62)–(1.66) to obtain the numerical solutions of
PDEs and PDSs. Since the parabolic PDEs and PDSs are concerned in this book,
the followings focus on the approximation equations (1.62) and (1.66).
Introduction to Partial Differential Systems 17
For the parabolic PDEs and PDSs on the 2-dimensional spatial domain Ω
(Figure 1.1), we first impose a mesh of grid points; hence, we have the interior
( )
grid points x1( k ) , x 2(l ) = ( khx1 , lhx2 ), where k = 1, 2,, N x1 , and l = 1, 2,, N x2 . With
the result of equation (1.66), but without considering O(⋅), we have
∆y =
(( ) ) (( ) ) ((
y x1( k +1) , x 2(l ) , t − 2 y x1( k ) + x 2(l ) , t ) + y x1( k −1) , x 2(l ) , t ) )
2
h x1
+
(( ) ) (( ) ) ((
y x1( k ) , x 2(l +1) , t − 2 y x1( k ) + x 2(l ) , t + y x1( k ) , x 2(l −1) , t ) ) (1.67)
2
hx2
For simplicity, one always defines hx to be the distance between mesh points, i.e.,
hx1 = hx2 = hx . This yields the simplification
1
∆yk ,l =
hx2
[ yk +1,l + yk −1,l + yk ,l +1 + yk ,l −1 − 4 yk ,l ] (1.68)
( )
where yk ,l y x1( k ) , x 2(l ) , t = y ( khx1 ,lhx2 ). If the higher order term is considered,
then we could find out that the local truncation error for ∆yk ,l is
1 2 ∂4 y ∂4 y
hx + (1.69)
12 ∂x14 ∂x 24 x = x ( k ) , x
1 1
(l )
2 = x2 , y = y ( x1( k ) ,x2(l ) )
( )
which is usually written as O hx2 such that the error is proportional to the space
( )
constant square O hx2 . That is, the local truncation error O hx2 , which will be ( )
discussed later, can be neglected if one chooses a sufficient small grid size hx .
Note that five points (points to the left, right, above and below of a central point,
( )
x1( k ) , x 2(l ) ) are involved in the approximation of ∆yk ,l ; hence, the FDM with the
central finite difference scheme is also called the five point formula. Obviously,
the FDMs with the forward, central, and backward finite difference schemes are
spatial-discretized methods.
Recall that whatever types of PDEs or PDSs, i.e., the hyperbolic, parabolic,
and elliptic types of PDEs or PDSs, can be numerically simulated by employ-
ing the FDM mentioned above. In order to guarantee the well-posedness of the
PDEs or PDSs with the FDM, we give the following discussion. Suppose y( x1 , x 2 )
represents the exact solution of a PDE or PDS L ( y ( x1 , x 2 )) = 0 with independent
variables x1 as well as x 2, and yk ,l is the exact solution of the corresponding spatial-
discretized algebraic equation F ( yk ,l ) = 0, which is constructed from equations
(1.62)–(1.66). Then, the FDM is said to be convergent if yk ,l tends to y( x1 , x 2 ) as
(
hx1 and hx2 tend to zero. The difference d k ,l y x1( k ) , x 2(l ) − yk ,l Yk ,l − yk ,l is the )
truncation error or the so-called discretization error. This error can be gener-
ally be minimized by decreasing the grid sizes hx1 and hx2 . Moreover, this error
depends on the number of terms in the truncated series equations (1.62)–(1.66),
18 Robust Engineering Designs of Partial Differential Systems
which are used to approximate each partial derivative. Another kind of error is
introduced when a PDE or PDS is approximated by the spatial-discretized alge-
braic equation. If the solution yk ,l is replaced by the solution y( x1 , x 2 ) at the grid
points (see Figure 1.1), then the value F (Yk ,l ) is called the local truncation error
at Pk ,l . The FDM and the PDE or PDS are said to be consistent if F (Yk ,l ) tends to
zero as hx1 and hx2 tend to zero. In general, the spatial-discretized algebraic equa-
tions cannot be solved exactly because the numerical computation is carried out
only up to a finite number of decimal places. Consequently, another kind of error
is introduced in the solution yk ,l during the actual process of computation. This
error is called the round-off error, and also depends upon the type of computer
used. In practice, the actual computational solution is yk ,l *, but not yk ,l , so that the
difference rk ,l yk ,l − y*k ,l is the round-off error at Pk ,l . In reality, the round-off
error depends mainly on the actual computational process and the FDM itself;
therefore, this error cannot be made small by allowing hx1 and hx2 to tend to zero.
Hence, the total error involved in the finite difference analysis at the point Pk ,l is
analysis states that a FDM is stable if the errors made at one time step ht of the
calculation do not cause the errors to increase as the computations are continued.
On the contrary, if the errors grow with time, the FDM is unstable. The stabil-
ity analysis is based on the decomposition of the errors into Fourier series. For
examples, consider 1-dimensional and 2-dimensional heat equations as
∂y( x ,t) ∂2 y ( x , t )
=α (1.71)
∂t ∂x 2
and
∂y (( x1 , x 2 ) , t ) ∂2 y (( x1 , x 2 ) , t ) ∂2 y (( x1 , x 2 ) , t )
= α1 + α2 (1.72)
∂t ∂x1 2
∂ x 22
where α ∈ + , α 1 ∈ + , and α 2 ∈ +. Then, the stability conditions of equations
(1.71) and (1.72) are
α ht 1 αh α h 1
≤ , and 12 t + 22 t ≤ (1.73)
hx2 2 hx1 hx2 2
respectively, where h 2x1 and h 2x2 have been defined in equations (1.65) and (1.66).
As to the effect of diffusion on the concerned PDS equation (1.8), we choose a
conservative stability condition as
D ht ht ht 1
λmax h 2 + h 2 ++ h 2 ≤ 2 (1.74)
x1 x2 xN
where λmax
D
is the maximum eigenvalue of the diffusion coefficient matrix D. With
the stability conditions (1.73) and (1.74) and the consistency based on hx1 → 0 and
hx2 → 0, the convergence can be guaranteed via the Lax’s equivalence theorem.
Surely, the FDM with the central finite difference scheme under the stability con-
dition (1.74) is utilized in our simulation results for the concerned PDSs through-
out this book.
After introducing the solutions of PDEs and PDSs via using the FDMs, we
briefly demonstrate the treatments of the diffusion effect on the parabolic PDSs
for the control design purposes. With the central finite difference scheme, the
similar methods will be used in the followings to deal with the diffusion effect.
The primary idea is to collect the system states at all grid nodes on spatial domain
as a spatial state vector. Then, the PDSs can be transformed into the ODSs with
the local truncation errors. Accordingly, the control designs for the PDSs can
be formulated as ones for the ODSs that can be realized via the traditional alge-
braic matrix techniques. Taking a simple 2-dimensional linear parabolic PDS for
example, we consider the PDS
∂y
= D y + ay + bu (1.75)
∂t
20 Robust Engineering Designs of Partial Differential Systems
where the only difference between equations (1.75) and (1.37) is that
x = [ x1 , x 2 ] ∈Ω ⊂ 2 with D ∈ +. With the typical grid mesh shown
T
∂ 2 y( x , t ) y (t ) + yk −1,l (t ) − 2 yk ,l (t )
∂ x 2 = k +1,l
x = x k ,l
hx2
yk ,l +1 (t ) + yk ,l −1 (t ) − 2 yk ,l (t )
+ + O k ,l (hx2 ) (1.76)
hx2
( )
where Ok ,l hx2 is referred to as the local truncation error, which converges to
zero if the grid size hx is chosen to be as small as possible by applying the Lax’s
equivalence theorem in Theorem 1.2.1, i.e., Ok ,l hx2 → 0 as hx → 0. Therefore, ( )
the PDS equation (1.75) can be represented as follows:
yk ,l (t ) = ayk ,l (t ) + buk ,l (t )
D
+
hx2
[ yk +1,l (t ) + yk −1,l (t ) − 2 yk ,l (t ) + yk ,l+1 (t ) + yk ,l−1(t ) − 2 yk ,l (t )] (1.77)
where uk ,l (t ) is the control effort at the grid node x k ,l. To simplify the control
design, we define a spatial state vector Y(t) to collect the states yk ,l (t ) at all grid
nodes in Figure 1.1, and Y (t ) ∈ NY is given as follows:
T
Y (t ) = y1,1 (t ), , yN x1 ,1 (t ), , yk ,l (t ), , y1, N x2 (t ), , yN x1 , N x2 (t ) (1.78)
y j (t ) = ay j (t ) + bu j (t ) + DTjY (t ) (1.79)
where Tj ∈ 1× NY expresses the interaction from the other grid nodes to the jth
node as follows:
= ∑ (T
j2 =1
x1 , j , j2 + Tx2 , j , j2 ) y j2 (t ) (1.80)
Introduction to Partial Differential Systems 21
−2
I , for j2 = (l − 1) N x1 + k
hx2
1
I , for j2 = (l − 1) N x1 + k − 1
Tx1 , j , j2 hx2
1
I , for j2 = (l − 1) N x1 + k + 1
hx2
0, otherwise
(1.81)
1
I , for j2 = (l − 2) N x1 + k
hx2
1
I , for j2 = lN x1 + k
Tx2 , j , j2 hx2
−2
I , for j2 = (l − 1) N x1 + k
hx2
0, otherwise
By applying some matrix multiplication techniques to (1.79), the PDS with the
state Y (t ) can be constructed; thus, the control designs for the PDS (1.75) can be
realized via the traditional algebraic matrix techniques when hx → 0. Note that if
the state y( x ,t) of the PDS (1.75) is not a constant function, i.e., y( x ,t) ∈ ny , then
Y (t ) ∈ ny NY , Ok ,l (hx2 ) ∈ ny , Tj ∈ ny × ny NY , Tx1 , j , j2 ∈ ny × ny , and Tx2 , j , j2 ∈ ny × ny .
Accordingly, the control designs for the PDSs can be realized based on control
design of DDS with state Y(t) as the previous analysis [4,5,9–11]. More details
will be given in the following chapters. The main drawback of using the cen-
tral finite difference scheme to deal with the diffusion effect on the parabolic
PDSs for the control designs is that the number of grid nodes N Y needs to be
sufficient large. That is, the tradeoff problem between the grid size hx for the
central finite difference scheme and the control performance occurs. If the grid
size h x is small enough so that Ok ,l (hx2 ) → 0, then a great deal of computational
resources is necessary because of the sufficient large N Y. Otherwise, the effect
of the local truncation error Ok ,l (hx2 ) needs to be considered to avoid unexpected
control performance. That is, a great deal of computational resources is neces-
sary as well.
∫Ω
yT ( x )∆y( x ) dx =
∫ Ω
yT ( x )∇ 2 y( x ) dx
∂y( x )
=−
∫ ∇ y(x)∇y(x) dx + 2 ∫
Ω
T
∂Ω
yT ( x) ds (by Green Theorem) (1.82)
∂n
∂y( x )
where ∇y( x) ⋅ n =
∂n
By the Dirichlet and Neumann boundary conditions in equation (1.10) [2,3]
∫Ω
yT ( x )∆y( x ) dx = −
∫ ∇ y(x)∇y(x) dx
Ω
T
(1.83)
cp
∫ Ω
yT ( x )y( x ) dx ≤
∫ ∇ y(x)∇y(x) dx
Ω
T
(1.84)
for some Poincaré constant cp, which can be chosen as the first nonzero eigen-
value of the Laplace operator ∆.
Then we obtain
∫ Ω
yT ( x )∆y( x ) dx ≤ −c p
∫ Ω
yT ( x ) y( x ) dx (1.85)
Based on equation (1.85), the Laplace operator ∆y( x ) in the system design of
PDSs could be treated like the conventional ordinary differential systems (PDSs).
These Poincaré inequality techniques will be applied in Chapters 5, 12, and 13 for
signal processing and control design of PDSs [12–14].
Part II is divided into three chapters. Based on finite difference scheme and
stochastic spatial state space model, Chapter 3 introduces a robust H∞ filter design
of linear stochastic PDSs with a set of sensor measurements [8]. With the help
of fuzzy spatial state space method and finite difference scheme, in Chapter 4,
a robust H∞ filter design is also introduced for nonlinear stochastic PDSs in sen-
sor signal processing [9]. In Chapter 5, based on Green theorem and Poincaré
inequality, a robust H∞ synchronization design is given to PDSs with diffusion
coupling [12].
Part III focuses on robust control system design and is divided into eight chap-
ters. Based on Galerkin method, a robust observer-based stabilization control
design is proposed in Chapter 6 for large structural systems like supported beam
problem under mode truncation, parameter perturbations and actuator saturations
[6]. Chapter 7 introduces a robust observer-based control design of large flexible
structures based on mode state space approach and frequency domain robustness
measure method. The robust observer-based controller could stabilize the con-
trolled mode state space model and tolerate the coupling spillovers from residual
mode state space model [7]. In Chapter 8, based on finite difference scheme and
spatial state space model, a robust H∞ stabilization design for stochastic linear
PDSs under spatio-temporal disturbance and sensor measurement noises. Based
on Galerkin method and fuzzy state space modeling, a robust observer-based
control design is proposed for nonlinear PDSs in Chapter 9. T-S fuzzy PDS can
interpolate several local linear partial systems to approximation any nonlinear
PDS by fuzzy bases interpolation [4]. Then the schemes in observer-based control
design in Chapter 7 could be employed for robust H∞ stabilization of nonlinear
PDSs. In Chapter 10, based on fuzzy interpolation schemes, FDM and spatial
state space model, a robust H∞ reference tracking control design of nonlinear
distributed parameter time-delayed systems is introduced for practical application
[5]. The proposed optimal robust H∞ reference tracking control design problem
of nonlinear PDSs can be transformed to a linear matrix inequalities (LMIs) –
constrained optimization problem. Based on FDM, spatial state space model
and fuzzy interpolation method, a robust H∞ stabilization design is proposed in
Chapter 11 for nonlinear stochastic PDSs. A robust estimator-based controller
is proposed for robust stochastic H∞ stabilization design to efficiently attenuate
the effect of random external disturbance and measurement noise in the spatio-
temporal domain [11]. In Chapter 12, based on divergence theorem and Poincaré
inequality to transform diffusion matrix inequality to bilinear matrix inequal-
ity (BMI), a robust fuzzy H∞ estimator-based stabilization design is proposed
for nonlinear parabolic PDS with different boundary conditions [13]. Based on
the proposed decoupling method, the robust fuzzy H∞ estimator-based stabilizing
design problem for nonlinear parabolic PDSs can be effectively solved by a set of
LMIs instead of BMI. Since design-cost of fuzzy controller for nonlinear PDSs is
very expensive, some low design-cost fuzzy controllers are proposed for nonlin-
ear parabolic PDSs [14]. Based on Green theorem and Poincaré inequality, robust
fuzzy H∞ stabilization area and point controllers are proposed as low design-cost
controller of nonlinear PDSs in Chapter 13.
24 Robust Engineering Designs of Partial Differential Systems
1.4 CONCLUSION
In this chapter, we first describe PDSs. Then some numerical and approximation
techniques of Galerkin method, FDM and Poincaré inequality method for PDEs
and PDSs are introduced to treat of Laplace operator, which is one of the most dif-
ficult problems in the signal processing and control engineering designs of PDSs
in the following chapters. Finally, an overview of each chapter in three parts of
the book about signal processing and control designs of the partial differential
engineering systems is also outlined in this chapter.
2 Fuzzy Solutions to
Partial Differential
Equations
2.1 INTRODUCTION
It is well known that a lot of phenomena of nature or physical systems can be
modeled by partial differential equations PDEs, such as heat equations, wave
equations, and so on. Hence, studies of PDEs have become one of the main
topics of modern mathematical analysis and have attracted much attention.
However, the exact solutions to the PDEs cannot be easily obtained except for
very simple or special cases. In recent years, many methods have been devel-
oped for solving some kinds of PDEs. For example, a spreadsheet program is
used for the numerical solution of the hyperbolic equation in [17]. Moreover,
some studies produce a solution in the form of an array that contains the value
of the solution at a selected group of points [18]. Others use finite-element
methods that are famous and widely adopted in the mechanical fields to solve
some specific PDEs [19–21]. In general, a finite-element solution can only offer
one discrete solution to approximate the exact solution of PDEs through mini-
mizing the functional of the finite-element method, but this solution is often
limited differentiable. In addition, the functional of variational finite-element
method must be given first. In general, it is not easy to find the functional of the
finite-element method for complex PDEs that cannot be easily derived from the
variational calculus [22,23]. Furthermore, how close this computed solution is
to the exact solution and whether it converges to the exact solution cannot eas-
ily be checked in practice. Recently, the fuzzy spline wavelets [24] were used
to approach the solution of differential equations. On the other hand, a fuzzy
transform technology [25] was applied to approach the numerical solutions of
PDEs based on finite-difference methods in [26].
Hence, the investigation of an effective and more correct method for solving
the partial differential systems (PDSs) or the systems of PDEs is an important
task! As mentioned earlier, one cannot easily obtain a solution for complex PDEs
(high order and nonlinear types). The main reason is that exact solutions are like
unknown black boxes, and the profile of the solution cannot be easily obtained
by human works or any mathematical analysis. Since the solution of PDEs can
be regarded as an unknown system, an approximated method that can handle
any unknown system is likely to find a suitable and accurate solution for the
treated PDEs. The numerical solution via a fuzzy transform method, as in [26],
DOI: 10.1201/9781003229230-3 25
26 Robust Engineering Designs of Partial Differential Systems
needs to solve a PDE via an algebraic equation through a fuzzy transform tech-
nique. However, for a nonlinear or high-dimensional PDE, the algebraic equation
becomes very complex. In this chapter, a fuzzy solution via adaptive algorithm is
proposed to approach the solution of PDE. Therefore, for the nonlinear or high-
dimensional cases, the solution of PDE can be easily obtained by the proposed
adaptive fuzzy algorithm.
In this chapter, we view the traditional problems of PDEs mentioned earlier
from a different perspective with the help of fuzzy logic systems. Fuzzy logic
systems have been widely used in system modeling to deal with the nonlinear
unknown systems for control or filter designs in recent years [27–31]. In most of
these fuzzy system designs [32–35], the fuzzy system was thought to be a univer-
sal approximator [36,37] for any nonlinear system. Fuzzy logic system has also
been proved to be a very good representation for a class of nonlinear dynamic
systems by the conventional schemes, and any nonlinear unknown system can be
approximated to any desired accuracy [32]. For this reason, an advanced method
that relies on the functional approximation capability of the fuzzy logic systems
via fuzzy interpolation scheme to solve the problems of PDEs will be proposed
in this chapter. This interpolation scheme employs a linear regression of fuzzy
logic system as the basic approximation element, whose parameters in conse-
quent parts are adjusted to minimize an appropriate error function via an adap-
tive algorithm. In addition, the upper bound of approximation error between the
exact solution and the proposed fuzzy solution is derived, and this result can be
easily extended to high-dimensional PDE cases. Moreover, the sufficient condi-
tion for the convergence of the proposed fuzzy solution in mesh points is also
provided.
Three attractive features of the proposed method are obtained as follows:
2.2 PROBLEM FORMULATION
2.2.1 DescriPtion of PDe Problems
Consider a class of a two-dimensional PDE formulated as follows [38]:
f ( x , y,φ ,φ x ,φ y ,φ xy ,φ xx ,φ yy ) = u( x, y) (2.1)
Fuzzy Solutions to Partial Differential Equations 27
∂φ
g x , y, φ , = 0 (2.2)
∂n ∂Ω
∂φ ∂φ
where is the normal derivative on ∂Ω, i.e., some functions of φ and evalu-
∂n ∂n
ated on boundary ∂Ω are set to zero.
Assumption 2.2.1
Remark 2.2.1
or
n∇φ ( x ) = C ( x ), (Neumann) (2.5)
where x = ( x1 , x n ) ∈ n ,φ ( x ) is the solution to be computed, and n is the out-
ward unit vector normal to the boundary.
To solve the PDE problem in equations (2.1) and (2.2), we denote two functions
β 0 and β1 as follows:
∂φ ∂φ
β1 x , y,φ , = g x , y,φ , (2.7)
∂n ∂n ∂Ω
28 Robust Engineering Designs of Partial Differential Systems
If one exact solution φ ( x , y) for the PDEs and the BCs can be found analytically,
β 0 (*) and β1 (*) will be zero. However, it is very difficult to find one exact solution
from the PDEs with the BCs mentioned earlier by mathematical analysis or tools.
For this reason, we will develop one approximated fuzzy solution φ ( x , y θ ) in
the sequel to satisfy the PDEs equation (2.1) with BCs in equation (2.2) as pre-
cisely as possible.
Problem of the PDE: Using Assumption 2.2.1, we consider the PDE equation
(2.1) with boundary conditions (2.2). The design objective is to find one suitable
and approximated fuzzy solution φ ( x , y θ ) such that β 0 (*) and β1 (*) are as small
as possible.
Rα : IF x is µix and y is µ uj
(2.8)
THEN φ ( x , y) = θ ij
where Rα (α = 1, , mx × n y ) are fuzzy rules, and x and y are the input vari-
ables to the fuzzy system. φ ( x , y) = θ ij is the solution proposed by the αth rule;
µix (i = 1, ,mx ) and µ uj ( j = 1, ,n y ) are the membership functions that charac-
terize the ith and jth fuzzy sets defined in the space of the variables of x and y
coordinates, respectively. These membership functions in these cases have been
chosen as follows:
1 x − aix 2
µix = exp − x (2.9)
2 σ i
1 y − ay 2
µ = exp −
u
j
j
y
(2.10)
2 σ j
where aix , a yj , σ ix , σ jy are the mean values and the standard deviations of the mem-
bership distributions, respectively. From the membership function shown in equa-
tions (2.9) and (2.10), it is easy to find out that they are nonlinear functions in C∞.
The overall fuzzy solution is given by
Another random document with
no related content on Scribd:
Indeed, the most prevailing incentive to our labor was to secure
the means of education for some male member of the family. To
make a gentleman of a brother or a son, to give him a college
education, was the dominant thought in the minds of a great many of
these provident mill-girls. I have known more than one to give every
cent of her wages, month after month, to her brother, that he might
get the education necessary to enter some profession. I have known
a mother to work years in this way for her boy. I have known women
to educate by their earnings young men who were not sons or
relatives. There are men now living who were helped to an education
by the wages of the early mill-girls.
In speaking of this subject, Mr. Thomas Wentworth Higginson
says,—
“I think it was the late President Walker who told me that in his
judgment one-quarter of the men in Harvard College were being
carried through by the special self-denial and sacrifices of women. I
cannot answer for the ratio; but I can testify to having been an
instance of this myself, and to having known a never-ending series
of such cases of self-devotion.”
CHARACTERISTICS (CONTINUED).
One of the most curious phases in the life of New England, and
one that must always puzzle the historian of its literature, is its
sudden intellectual blossoming half a century ago.
Emerson says, “The children of New England between 1820 and
1840 were born with knives in their brains;” and this would seem to
be true, since during or very near that time, were born the majority of
those writers and thinkers whose lives have been so recently and so
nobly rounded out,—Emerson, Bryant, Longfellow, Lowell, Whittier,
John Pierpont,—they whose influence cannot be overestimated in
bringing an ideal element into our hitherto prosaic New England life.
The seeds of this intellectual growth came suddenly, as if blown
from some far-off cultured land, and were sown broadcast. Some
found a resting-place in this little corner of New England, where were
gathered together these daughters of Puritan ancestors, and they,
too, feeling the intellectual impetus, were impelled to put in writing
their own crude thoughts. Their desire for self-improvement had
been to some extent gratified, and they now began to feel the benefit
of the educational advantages which had been opened to them. As
in “Mary Barton,” they “threw the shuttle with increasing sound,
although Newton’s ‘Principia’ lay open before them, to be snatched
at in work-hours, but revelled over at meal-time or at night.”
And the “literary” girls among us would often be seen writing on
scraps of paper which we hid “between whiles” in the waste-boxes
upon which we sat while waiting for the looms or frames to need
attention. Some of these studious ones kept note-books, with
abstracts of their reading and studies, or jotted down what they were
pleased to call their “thoughts.” It was natural that such a thoughtful
life should bear fruit, and this leads me to speak of The Lowell
Offering, a publication which was the natural outgrowth of the mental
habit of the early mill-girls, for many of the pieces that were printed
there were thought out amid the hum of the wheels, while the skilful
fingers and well-trained eyes of the writers tended the loom or the
frame.
The idea of organization for literary and educational purposes
was first proposed in 1837 by Miss Harriot F. Curtis, perhaps the
most progressive of all the mill-girls. She with her immediate
associates conceived the idea of forming a little society for mental
improvement. In The Lowell Offering of January, 1845, is the
following account of its formation written by Miss Maria Currier.
“IMPROVEMENT CIRCLE.
“In one of the corporations [the Lawrence] of this city, about eight
years ago, might have been seen, on a summer evening, a company
of four or five young females, who through the day had labored at
their several employments in some one of the factories connected
with the corporation. Perhaps they were not ambitious above others
of their sex.... But wishing to improve the talents which God had
given them, they proposed the formation of a society for mutual
improvement. An evening was appointed for the proposed purpose;
and having invited a few others to join them, they met at the time
appointed.... A president, vice-president, and secretary were chosen;
a constitution was drafted, and by-laws formed, to which each of the
members affixed her name.... At length a circle on a more extensive
scale was formed by a gentleman of this city, and a plan conceived
of bringing before the world the productions of inexperienced
females; of showing that intellect and intelligence might be found
even among factory operatives. It was then that The Offering was
published; and many of those who were present at the first meeting
of our Improvement Circle were contributors to its pages.”
THE
LOWELL OFFERING,
WRITTEN, EDITED, AND PUBLISHED
BY FEMALE OPERATIVES.
Our magazine is the only one which America has produced, of
which no other country has produced the like. The Offering is prima
facie evidence, not only of the American “factory-girls,” but of the
intelligence of the mass of our country. And it is in the intelligence of
the mass that the permanency of our republican institutions
depends.
And our last appeal is to those who should support us, if for no
other reason but their interest in “the cultivation of humanity,” and the
maintenance of true democracy. There is little but this of which we,
as a people, can be proud. Other nations can look upon the relics of
a glory come and gone—upon their magnificent ruins—upon worn-
out institutions, not only tolerated, but hallowed because they are old
—upon the splendors of costly pageant—upon the tokens of a
wealth, which has increased for ages—but we can take pride in
these. We have other and better things. Let us look upon our “free
suffrage,” our Lyceums, our Common Schools, our Mechanics’
Literary Associations, the Periodical of our Laboring Females; upon
all that is indigenous to our Republic, and say, with the spirit of the
Roman Cornelia, These, these are our jewels.
Terms: One dollar per year in advance. Postage: 100 miles and
under, 1½ cents. Over 100, 2½ cents.
Published at Lowell, Mass., monthly, by
MISSES CURTIS & FARLEY.