ESAIM: Proceedings
Elasticite, Viscoelasticite et Contr^ole Optimal Huitiemes Entretiens du Centre Jacques Cartier URL:http://www.emath.fr/proc/Vol.2/
ESAIM: Proc., Vol. 2, 1997, 215{224
MODELING AND OPTIMIZATION OF A NON-SYMMETRIC PLATE.
L.J. ALVAREZ-VAZQUEZ AND J.M. VIA~NO
Abstract. In the rst part of this paper we present a limit model for non-symmetric elastic plates derived from three-dimensional elasticity by use of an asymptotic method and we prove existence and uniqueness of a solution of this model. In the second part we try the shape opti- mization of the plate, we obtain the existence of an optimal prole and we propose its numerical solution by means of penalty methods.
Key words:
Non-symmetric plates, asymptotic analysis, shape optimiza- tion, penalty methods.Mathematics subject classication:
73K10, 73K40, 73C02, 35C20.1. Introduction
The use of asymptotic methods for obtention and mathematical justica- tion of plate models in the framework of the theory of elasticity has been shown to be a successful technique during the last decades. First fundamen- tal contributions in this direction were obtained by Ciarlet - Destuynder 7]
with the justication of the classical bi-harmonic model of Kirchho - Love for the bending of symmetric plates. The application of this method to dif- ferent situations (linear and nonlinear elasticity, composite and anisotropic materials, static, dynamic and thermoelastic cases, homogenization and so on) has provided important contributions. A complete analysis of plate models with exhaustive bibliographic references may be found in Ciarlet 6].
Asymptotic methods have also reached an important development in the case of shells (Destuynder 11], Ciarlet - Paumier 9], Ciarlet - Miara 8]) and rods (Trabucho - Via~no 16], Cimetiere et al. 10]).
All of the above works have been exclusively devoted to the study of symmetric plates. The case of non-symmetric plates has so far remained, as far as we know, completely unreported. The rst aim of this paper is the obtention of a mechanical model for non-symmetric plates in the frame- work of linear elasticity. In section 2 we describe the physical problem, we obtain the mathematical model by use of an asymptotic analysis of the three- dimensional problem and we prove existence and uniqueness of a solution for such a model.
In section 3, we study the optimal design problem of minimizing the weight of the non-symmetric plate under some geometric and technological constraints, with some imposed bounds on the deection. In structural op- timization, these problems can be formulated as state constrained optimal control problems governed by an elliptic dierential equation, the control being a small parameter appearing in the coecients of the dierential op- erator (cf. Casas 3], Haug - Arora 13], Hlavacek - Bock - Lovisek 14], 15], Haslinger - Neittanmaki 12]). In this section we pose such a problem and obtain the existence of, at least, an optimal prole.
Article published by EDP Sciences and available at http://www.edpsciences.org/proc or http://dx.doi.org/10.1051/proc:1997002
In order to carry out the numerical solution of the optimal shape prob- lem, we propose in the last section the use of penalty techniques, obtaining necessary optimality conditions and some convergence results.
2. Modeling of a non-symmetric plate
Let
"
be a positive real parameter and!
be a domain inR
2 with coordi- nates axisx
1x
2:
Leth
2W
21(!
) be a \thickness" function verifyingh
(x
1x
2)>
0 for all (x
1x
2)2!:
(1) We consider the non-symmetric elastic plate of thickness"h
occupying the reference conguration " dened by"=f(
x
1x
2x
"3) : (x
1x
2)2!
0< x
"3< "h
(x
1x
2)g:
(2) We denote by@
"v
=@
v
the derivative@v=@x
and by@
3"v
the derivative@v=@x
"3:
Here and along the whole work we use, as it is customary in elas- ticity theory, the summation convention on repeated indices, supposing that the Latin indices range over f123gand Greek ones overf12g:
We study the physical problem corresponding to the mechanical behav- ior of a non-symmetric elastic plate, supposed to be clamped on the lateral surface and submitted to body and surface forces. We assume the constitu- tive material of the plate to be a homogeneous isotropic elastic material of Saint Venant-Kirchho's type with Young's modulus
E
and Poisson's ratio:
Then, in the linearized elasticity framework, the displacement eldu
"and the Piola-Kirchho stress tensor " are the solution of the following mixed variational formulation (see Ciarlet 5]):u
"2V
(") "2(") :Z
"
(1 +
E
"ij;E
"kkij)"ijdx
";Z
"
"ij(u
")"ijdx
"= 08"2(") (3)Z
"
"ij"ij(v
")dx
"=Z
"
f
"iv
"idx
"+Z
;
"
+
;
"
;
g
"iv
"ida
" 8v
"2V
(") (4) where;"0 =f(
x
1x
2x
"3) : (x
1x
2)2@!
0< x
"3< "h
(x
1x
2)g (5);"+=f(
x
1x
2x
"3) : (x
1x
2)2! x
"3 ="h
(x
1x
2)g (6);";=f(
x
1x
2x
"3) : (x
1x
2)2! x
"3 = 0g (7)V
(") =fv
"(v
"i)2H
1(")]3 :v
"i= 0 on ;"0g (8) (") =L
2(")]9S =f"("ij)2L
2(")]9 : "ij="jig (9) and "(u
") = ("ij(u
")) is the linearized strain tensor "ij(u
") = 12(@
"iu
"j+@
"ju
"i):
(10) We consider the reference symmetric plate of constant thickness 1 occupying the volume where=f(
x
1x
2x
3) : (x
1x
2)2!
0< x
3<
1g=!
(01) (11) and we dene;0 =f(
x
1x
2x
3) : (x
1x
2)2@!
0< x
3<
1g=@!
(01)(12);+=f(
x
1x
2x
3) : (x
1x
2) 2! x
3= 1g=!
f1g (13);;=f(
x
1x
2x
3) : (x
1x
2) 2! x
3= 0g=!
f0g:
(14)We can dene the change of variable from " to the xed domain :
":x
(x
1x
2x
3)2;!"(x
1x
2x
3) =(
x
1x
2"x
3h
(x
1x
2))(x
1x
2x
"3)x
"2":
(15) Now we scale the dierent elds appearing in the variational formulation (see Ciarlet - Destuynder 7]) and we deneu
("
) and ("
) byu
("
)(x
) =u
"(x
")u
3("
)(x
) ="u
"3(x
") (16) ("
)(x
) ="(x
")3("
)(x
) ="
;1"3(x
")33("
)(x
) ="
;233" (x
"):
(17) Also we assume that the applied forces are such thatf
"(x
") =f
(x
)f
3"(x
") ="f
3(x
) (18)g
"(x
") ="g
(x
)g
"3(x
") ="
2g
3(x
) (19) wheref
i 2L
2()g
i 2L
2(;+ ;;) are independent of":
Then, we ob- tain that (u
("
)("
)) is the only solution of the following scaled variational problem posed in :u
("
)2V
() 2() :; Z
h
hij(u
("
))ijdx
+Z
h
f1 +E
("
);E
("
)gdx
+"
2Z
h
f21 +E
3("
)3;E
(33("
)+("
)33)gdx
+"
4Z
h E
1 33("
)33dx
= 0 8 2() (20)Z
h
ij("
)hij(v
)dx
=Z
hf
iv
idx
+Z
;
h
("
)g
iv
ida
8v
2V
() (21) whereV
() =fv
(v
i)2H
1()]3 :v
i = 0 on ;0g (22)() =
L
2()]9S (23)h
("
) = 1 +"
2(@
1h
)2+"
2(@
2h
)2]1=2 (24) and h(v
)(hij(v
)) is the generalized strain tensor dened by h(v
) = 12@
v
+@
v
;x
3h
;1@
h@
3v
;x
3h
;1@
h@
3v
] (25) h3(v
) = 12@
v
3+h
;1@
3v
;x
3h
;1@
h@
3v
3] (26) 33h(v
) =h
;1@
3v
3:
(27) In order to obtain the convergence of the scaled three-dimensional problem as"
tends to zero we are going to show several technical results. In what follows, j j0D and jj jjmD denote theL
2(D
)-norm andH
m(D
)-norm, respectively.Let
E
h() be the spaceE
h() =fv
(v
i)2L
2()]3 : hij(v
)2L
2()g endowed with the normjj
v
jjh=fjv
j20+X3ij=1
j
hij(v
)j20g1=2 for allv
2E
h():
(28) We have the results whose complete proofs can be seen in 1].Theorem 1.
H
()] =E
h() and the norms jj jjh and jj jj1 are equivalent.We introduce the space of displacements of Kirchho-Love
V
hKL() =fv
(v
i)2V
() : hi3(v
) = 0g:
(29) We have the following characterization for this space.Theorem 2. The space
V
hKL() is characterized byV
hKL() =fv
(v
i) :v
3(x
1x
2x
3) =3(x
1x
2)32H
02(!
)v
(x
1x
2x
3) =(x
1x
2);x
3h
(x
1x
2)@
3(x
1x
2)2H
01(!
)(30) Remark 3. We write (30) in the following shortened formV
hKL() = fv
(v
i) :v
3 =3v
=;x
3h@
33 2
H
02(!
) 2H
01(!
)gand we note that the following mapping is an isomorphism
j
: (123)2H
01(!
)]2H
02(!
);!;!
j
() = (1;x
3h@
132;x
3h@
233)2V
hKL(:
) (31) Theorem 4. The semi-normj jh dened byv
2V
();!jv
jh=fX3ij=1
j
hij(v
)j20g1=2 is a norm overV
() equivalent to jj jj1:
Corollary 5. The mapping
v
2V
hKL();!jv
jh=f X2=1
j
h(v
)j20g1=2 is a norm overV
hKL() equivalent to the norm jj jj1:
Corollary 6. The mapping jj jjKL dened by (see (30))
v
2V
hKL();!jjv
jjKL=fjj1 jj21!+jj2 jj21!+jj3jj22!g1=2 is a norm overV
hKL() equivalent to the norm j jh and, consequently, to the norm jj jj1:
The rst step in order to obtain convergence of the sequence (
u
("
)("
)) consists of the following a priori estimates.Theorem 7. There exist
"
0>
0 and a constantC
=C
(h!
) such that for all 0< " < "
0 we havejj
u
("
)jj1C
j("
)j0C
j
"
3("
)j0C
j"
233("
)j0C:
Corollary 8. The sequencesf
u
("
)g">0 andf("
)g">0 verify the following weak convergencesu
("
)* u
inV
()(
"
)*
inL
2()"
3("
)*
0 inL
2()"
233("
)*
0 inL
2()where
u
is the unique element inV
hKL() solution of the problemZ
1;
E
2h
f(1;)h(u
)h(v
) +h(u
)h(v
)gdx
=Z
hf
iv
idx
+Z
;
g
iv
ida
for allv
2V
hKL() (32) and is given by =E
1;
2f(1;)h(u
) +h (u
)g:
(33) Corollary 9. The following strong convergences hold true:u
("
)!u
inV
()(
"
)! inL
2()"
3("
)!0 inL
2()"
233("
)!0 inL
2():
Remark 10. We can make the limit problem (32) more explicit by taking
u
= (1;x
3h@
132;x
3h@
233) (34)v
= (1;x
3h@
132;x
3h@
233) (35) for 2H
01(!
) and332H
02(!
) and by lettingg
i+(x
1x
2) =g
i(x
1x
21)g
i;(x
1x
2) =g
i(x
1x
20):
Then the limit problem takes the following form:1;
E
2fZ
!
h
(1;)()() +()()];Z
!
h
22 (1;
)(()@
3+()@
3) +(()@
3+()@
3)]+
Z
!
h
33 (1;
)@
3@
3+@
3@
3]g=Z
!
h
Z
1
0
f
i]i;Z
!
h
2Z
1
0
x
3f
]@
3+Z
!
g
i;+g
i+]i;Z
!
hg
+@
3 for all 2H
01(!
) 3 2H
02(!
):
Remark 11. If we assume an asymptotic expansion of the form (
u
("
)("
)) = (u
00) +"
2(u
22) +"
4(u
44) +:::
and if we substitute this expression into the scaled variational problem, we obtain that the rst term (
u
00) veriesu
0=u
0 =i03 =
x
3i0@
h
;Z x30
@
(h
i0 );g
i;;Z x30
hf
i:
Remark 12. We can return to the reference conguration "taking into ac- count that
u
is a limit approximation ofu
("
):
Performing the inverse change of variable and the inverse scaling, we obtain that the limit displacement eldu
0" in " is given byu
0"(x
") =u
(x
) =(x
1x
2);x
3h
(x
1x
2)@
3(x
1x
2) ==
(x
1x
2);"
;1x
"3@
3(x
1x
2) ="(x
1x
2);x
"3@
3"(x
1x
2)u
03"(x
") ="
;1u
3(x
) ="
;13(x
1x
2) =3"(x
1x
2)where
"= "3=
"
;13:
Letting
g
i"+(::
) =g
"i(::"h
(::
)) andg
i";(::
) =g
"i(::
0) we haveg
" ="g
andg
3"="
2g
3:
Then, we obtain that"2H
01(!
) and3"2H
02(!
) are the unique solution of the following problem which is the obtained model1;
E
2fZ
!
"h
(1;)(")() +(")()];Z
!
"
2h
22 (1;
)((")@
3+()@
3")+((")@
3+()@
3")]+
Z
!
"
3h
33 (1;
)@
3"@
3+@
3"@
3]g=Z
!
Z "h
0
f
"i+g
i";+g
i"+]i;Z
!
Z "h
0
x
"3f
"+"hg
"+]@
3 for all 2H
01(!
) 3 2H
02(!
):
3. Optimization of a non-symmetric plate
We consider an homogeneous, isotropic, elastic non-symmetric plate oc- cupying the reference conguration
h=f(
x
1x
2x
3)2IR
3 : (x
1x
2)2!
0< x
3< h
(x
1x
2)gclamped on its lateral surface and submitted to body forces of density
f
h (f
hi) on hand surface forces of densityg
h+(g
ih+) on ;h+andg
h;(g
ih;) on ;h; where;h+ =f(
x
1x
2x
3)2IR
3 : (x
1x
2)2! x
3=h
(x
1x
2)g;h;=f(
x
1x
2x
3)2IR
3 : (x
1x
2)2! x
3 = 0g:
We denote by
u
h (u
hi) : h;!IR
3the displacement eld corresponding to the above conditions. In the sequel we identify the functionsu
hf
h andg
h with the functionsu
(h
) : ;!R
3f
(h
) : ;!R
3 andg
(h
) : ; ;!R
3 such thatu
(h
)(x
1x
2x
3) =u
h(x
1x
2x
3h
(x
1x
2))f
(h
)(x
1x
2x
3) =f
h(x
1x
2x
3h
(x
1x
2))g
+(h
)(x
1x
21) =g
h+(x
1x
2h
(x
1x
2))g
;(h
)(x
1x
20) =g
h;(x
1x
20):
For simplicity, we assume from now on that the applied forces are such that
f
(h
) =f
andg
(h
) =g
wheref
andg
are independent ofh:
As it was shown in the previous section the displacement eld
u
(h
) can be approximated by the only solution of the following variational problem:u
(h
)2V
hKL()a
h(u
(h
)v
) =l
h(v
) for allv
2V
hKL() where for alluv
2V
hKL(),a
h(uv
) =Z
1;
E
2h
f(1;)h(u
)h(v
) +h(u
)h(v
)gdx
(36)l
h(v
) =Z
hf
iv
idx
+Z
;
g
+iv
ida
+Z
;
g
i;v
ida:
(37)The optimal design problem deals with minimizing the weight of the non- symmetric plate h
or equivalently its volume (since the material is homo- geneous), that we denote byJ
(h
) =R!h:
The plate must be designed in such a way that allowed deections be less than a given bound
e:
Asu
3(h
)2H
02(!
) we impose:jj
u
3(h
)jj1!e:
In addition, we impose to the thickness some constraints of technological nature:
i
) The plate is designed in order to be constructed later, and this con- struction would not be possible if the plate is too thick or too thin. This fact leads us to impose the constraint:0
< a
h
(x
)b <
1 for allx
2!:
ii
) We also need that the variation of the thickness be slow and progres- sive. Then, we requirejj
@
h
jj1!c
jj@
h
jj1!d:
These technological constraints lead us to work in the set of feasible thick- ness:
U
ad=fh
2W
21(!
) :a
h
(x
)b
jj@
h
jj1!c
jj@
h
jj1!d
g which is a convex, closed and bounded subset ofW
21(!
):
If we dene the set
K
h = fv
2V
hKL() : jjv
3 jj1!e
g which is a convex, closed subset ofV
hKL() we can formulate the following optimal design problem:(
P
)Find h
2U
adsuch that
:u
(h
)2K
hJ
(h
) =inf
fJ
(h
) :h
2U
adu
(h
)2K
hg:
Our rst aim is to study the existence of a solution for problem (
P
):
In order to do so, we considerJ
as a function dened in the spaceW
11(!
)J
:h
2W
11(!
);!J
(h
) =Z
!
h
2IR
and we dene the set
B
=fh
2U
ad :u
(h
) 2K
hg that we assume to be non empty. Then, problem (P
) can be written in the following form:(
P
)Find h
2B such that J
(h
) =inf
fJ
(h
) :h
2B
g:
Since
J
is continuous, to obtain the existence of a solution it should be enough to prove thatB
is compact inW
11(!
):
In order to prove this fact, we write problem (
P
) in a more suitable form.We recall that the denition of h implicitly assumes that
h
is in the subsetA
(!
) of the spaceW
11(!
) given byA
(!
) =fh
2W
21(!
) :h
(x
)(h
)>
0 for allx
2!
g:
We denote by
L
2(V
hKL()) the space of continuous bilinear forms onV
hKL() and byL
(V
hKL()V
hKL()0) the space of continuous linear forms fromV
hKL() on its dual. We also introduce the functionsT
:h
2A
(!
);!T
(h
) =l
h 2V
hKL()0F
:h
2A
(!
);!F
(h
) =a
h 2L
2(V
hKL())G
:a
2L
2(V
hKL());!G
(a
) =A
2L
(V
hKL()V
hKL()0)where
A
(y
)(z
) =a
(yz
) for allyz
2V
hKL():
Of course,G
is an isometric isomorphism.With each bilinear form
a
h =F
(h
) we associate the operatorA
h =G
(a
h) =G
(F
(h
)):
Sincea
h isV
hKL();elliptic for allh
2A
(!
), it fol- lows from Lax-Milgram lemma that for eachl
h 2V
hKL()0 there exists a unique elementu
(h
) 2V
hKL() such thatA
h(u
(h
)) =l
h:
Then,A
h is an isomorphism fromV
hKL() ontoV
hKL()0 and the following function is well dened and of classC
1:u
:h
2A
(!
)W
11(!
);!u
(h
) =A
;1h (l
h)2V
hKL()H
1()]3:
This is the key to prove thatU
ad andB
are compact inW
11(!
):
Conse- quently, ifl
h is of the form (37) then problem (P
) has at least one solution.Remark 13. For
hk
2A
(!
) we denem
hk =DT
(h
)(k
)2V
hKL()0a
hk =DF
(h
)(k
)2L
2(V
hKL()):
Then we have from (37)m
hk =Z
kf
iv
i:
In the same way we obtain from (36) that for all
uv
2V
hKL() given by (34),(35)a
hk(uv
) =E
1;2fZ
!
k
(1;)()() +()()];Z
!
hk
(1;)(()@
3+()@
3)+(()@
3+()@
3)]+Z
!
h
2k
(1;)@
3@
3+@
3@
3]g:
LetA
hk =G
(a
hk)2L
(V
hKL()V
hKL()0):
Then we havep
=Du
(h
)(k
) =A
;1hm
hk;A
hk(u
(h
))]:
This equation will be helpful for the numerical computation of
Du
(h
)(k
) in the next section.4. Penalty method
In order to obtain the numerical solution of problem (
P
) we will use penalty techniques, dening a family of problems (P
) which approximate problem (P
) and which have a suitable numerical solution. Besides, we will obtain necessary optimality conditions for the penalty problem (P
):
For
>
0 we dene the functionJ
:h
2A
(!
);!J
(h
) =J
(h
) + 12 jju
(h
);P
Kh(u
(h
))jj2KL2R
whereP
Kh :V
hKL() ;!K
h denotes the projection in the Hilbert spaceV
hKL() over the closed and convex subsetK
hwhich is a continuous function (Barbu-Precupanu 3], Cea 11]). Thus,J
is also a continuous function.We consider the following problem:
(
P
)Find h
2U
adsuch that J
(h
) =inf
fJ
(h
) :h
2U
adg:
Due to the compacity ofU
adand the continuity ofJ
the problem (P
) has, at least, one solution.If for each
>
0 we noteh
a solution of problem (P
) then we have the following result.Theorem 14.
i
) If 1<
2 thenJ
2(h
2)J
1(h
1)hinf2B
J
(h
) (38)jj
u
(h
1);P
Kh 1(
u
(h
1))jjKL jju
(h
2);P
Kh 2(
u
(h
2))jjKL (39)J
(h
2)J
(h
1)hinf2B
J
(h
):
(40)ii
) The family fh
:>
0g is bounded inW
21(!
) and there exist sequences fng fh
ng such that ifn
!1then n!0h
n !h
inW
11(!
):
Each of these limitsh
is a solution of problem (P
):
iii
) It is veried thatlim!0
J
(h
) = lim!0
J
(h
) = infh2B
J
(h
):
As an immediate consequence, we obtain the following convergence result.
Corollary 15. If problem (
P
) has a unique solutionh
and, for each>
0
h
denotes a solution of the problem (P
)then as !0 we haveh
!h
inW
11(!
):
In order to obtain optimality conditions for the problem (
P
) we rst prove the following result.Theorem 16. The function
J
:A
(!
);!R
is of classC
1 andDJ
(h
)(k
) =J
(k
) +a
hk(u
(h
)z
);m
hk(z
) whereu
(h
) andz
verifyA
h(u
(h
)) =l
h (41)p
=I
;P
Kh (u
(h
)) (42)A
h(z
) +p
= 0 (43) and whereA
h:V
hKL();!(V
hKL())0 is the adjoint operator ofA
h:
A point
h
2U
ad is said to be a stationary point for problem (P
) if and only if it veriesDJ
(h
)(h
;h
) 0 8h
2U
ad:
SinceU
ad is a convex set, we have that ifh
2U
adis a solution of (P
) thenh
is a stationary point for (P
) (cf. Cea 4]). This fact supplies us the following necessary optimality condition.Theorem 17.
h
2U
ad is a stationary point for (P
) if and only if there existp
u
z
2V
hKL() verifyingA
h(u
) =l
hp
=I
;P
Kh (u
)A
h(z
) +p
= 0J
(h
);J
(h
) +a
hh;h(u
z
);m
hh;h(z
)0 8h
2U
ad:
References
1] L.J. Alvarez-Vazquez and J.M. Via~no. Modeling and optimization of non- symmetric plates.To appear.
2] V.Barbuand T.Precupanu.Convexity and optimization in Banach spaces.Sijtho and Noordho, Bucharest, 1978.
3] E.Casas.Optimality conditions and numerical approximations for some optimal de- sign problems.Control Cibernet., 19, pp 73-91 (1990).