Sunteți pe pagina 1din 13

Available online at www.sciencedirect.

com

COMPOSITES
SCIENCE AND
TECHNOLOGY
Composites Science and Technology 68 (2008) 19621974
www.elsevier.com/locate/compscitech

Boundary conditions for unit cells from periodic microstructures


and their implications
Shuguang Li *
School of MACE, The University of Manchester, Sackville Street, Manchester M60 1QD, UK
Received 4 December 2006; received in revised form 27 March 2007; accepted 30 March 2007
Available online 14 April 2007

Abstract
The most important aspect of formulating unit cells for micromechanical analysis of materials of patterned microstructures is the derivation of appropriate boundary conditions for them. There is lack of a comprehensive account on the derivation of boundary conditions
in the literature, while the use of unit cells in micromechanical analyses is on an increasing trend. This paper is devoted to the generation
of such an account, where boundary conditions are derived entirely based on considerations of symmetries which are present in the
microstructure. The implications of the boundary conditions used for a unit cell are not always clear and therefore have been discussed.
It has been demonstrated that unit cells of the same appearance but subject to boundary conditions derived based on dierent symmetry
considerations may behave rather dierently. One of the objectives of the paper is to inform users of unit cells that to introduce a unit cell
one needs not only mechanically correct boundary conditions but also a clear sense of the microstructure under consideration. Otherwise
the results of such analyses could mislead.
2008 Published by Elsevier Ltd.
Keywords: B. Mechanical properties; C. Anisotropy; C. Elastic properties; C. Finite element analysis (FEA); C. Modelling; Unit cells

1. Introduction
Micromechanical analyses have been on an increasing
trend in order to understand the behaviour of modern
materials with sophisticated microstructures, e.g. bre or
particulate reinforced composites, textile composites, etc.
Unit cells are often resorted to in order to facilitate such
analyses. The introduction of a unit cell is usually based
on certain assumptions, such as a regular pattern in the
microstructure, which is sometimes a reasonable approximation, or an idealisation otherwise. A regular pattern
oers certain symmetries which can then be employed to
dene the unit cell and to derive the boundary conditions
for it for micromechanical analysis. Several accounts of
the systematic use of symmetries for the derivation of the
boundary conditions for unit cells have been presented by

Tel.: +44 161 306 3842.


E-mail address: shuguang.li@manchester.ac.uk

0266-3538/$ - see front matter 2008 Published by Elsevier Ltd.


doi:10.1016/j.compscitech.2007.03.035

the author in [13]. In the literature, there are many


accounts where simplistic boundary conditions have been
imposed to unit cells in an intuitive manner, sometimes,
rather casually without much justication. In [4], boundary
conditions have been so introduced that boundary eects
have been considered and a signicant eort has been made
to include additional cells to form larger unit cells in order
to reduce the boundary eects. This would be absolutely
unnecessary, had the boundary conditions been derived
appropriately. In many publications [49], to name but a
few, boundary conditions have been so assumed that
boundaries have to remain at or straight after deformation in order to deliver simple boundary conditions, which
cannot usually be fullled when the material is subjected to
the macroscopic shear deformation. Such simplistic boundary conditions are correct in a few special cases, e.g. square
unit cells with reectional symmetric microstructure when
they are subjected to a deformation corresponding to a
macroscopic direct strain. Even so, there are implications
on the permissible patterns of the microstructures, which

S. Li / Composites Science and Technology 68 (2008) 19621974

do not seem to have been given any attention hitherto.


Another confusing issue is how many boundary conditions
need to be prescribed at any given part of the boundary of
a unit cell. Sometimes, only one displacement has been prescribed but in other cases more than one are prescribed. In
[10], equilibrium or compatibility conditions were imposed
as a part of boundary conditions, which are in fact wrong,
as will be explained later in the next section. This paper is
devoted to the issues as raised above, in particular, to the
confusing issues associated with the use of reectional
symmetries.
Because of the nature of the symmetries employed, the
boundary conditions obtained for the unit cells are often
in a form of equations relating displacements on one part
of the boundary to those on another, referred to hereafter
as equation boundary conditions, because of the use of translational or rotational symmetries [13]. This may impose
restrictions on the applications of these boundary conditions and hence the unit cells. For instance, when nite elements are employed for the micromechanical analysis, as is
often the case, the mesh to be generated must possess identical tessellation between the parts of boundary which are
related through those equation boundary conditions. This
could sometimes be dicult to achieve for 3-D problems,
such as in particle reinforced or textile composites. The
constraints in equation form may not be available in some
FE codes. It is therefore desirable to avoid such equation
boundary conditions whenever possible.
2. The concept of natural boundary conditions
Some of the confusions in deriving boundary conditions
for unit cells result from the use of nite elements which is
usually based on a variational principle of some kind in
which some boundary conditions, called natural boundary
conditions, are satised automatically as a part of the variation process. The natural boundary condition is a mathematical terminology commonly used in variational
principles [11]. The spirit of a variational principle is that
the conditions for a functional to take its stationary value
are equivalent to the satisfaction of the Euler equations and
the natural boundary conditions corresponding to the
functional. In the context of the minimum total potential
energy principle, the Euler equations are equilibrium equations of elasticity and the natural boundary conditions are
the traction boundary conditions. Almost all commercial
FE codes are based on the minimum total potential energy
principle or its counterpart, the virtual displacement principle, in which traction boundary conditions will be satised
in the same way and at the same time as the equilibrium
equations are satised when the total potential energy is
minimised.
It should be emphasised that natural boundary conditions should not be imposed prior to the variation process,
especially when seeking an approximate solution, e.g. using
nite elements. It does not help to obtain a more accurate
result but, rather on the contrary, it may prevent the total

1963

potential energy reaching its minimum in the solution space


and hence lead to a wrong result. As an illustration, a
straight bar having Youngs modulus E, length L and
cross-section area A under linearly distributed axial force
p = kx where k is a known constant, as shown in Fig. 1, is
considered as a simple example. The axial displacement
and the corresponding total potential energy, which is the
minimum corresponding to the exact solution, are given
in Table 1. An approximate solution is also shown in Table
1. The approximate displacement function is assumed to be
a quadratic function. In order to satisfy the displacement
boundary condition at x = L, it has to take the form as
shown, where a and b are constant coecients to be determined. After determining a and b using the minimum total
potential energy principle, the displacement and the corresponding total potential energy can be obtained as shown
in the row labelled Approximate solution in Table 1. This
is the best approximation that one can obtain with a quadratic displacement function. It can be noted that the traction boundary condition in this solution is only satised
approximately. This approximation is consistent with the
level of approximation in the satisfaction of the equilibrium, for which the stress obtained at the xed end as also
included in Table 1 could serve as an indication. The rigorous measure of the approximation is the minimised value of
the total potential energy. However, if one imposes the traction free boundary condition at the x = 0 end, equivalent to
the equilibrium or compatibility conditions as referred to
in [10], in addition to the xed boundary condition at the
other end, the solution is shown in the row labelled Wrong
solution in Table 1. The satisfaction of natural boundary
condition at free end is at the price of worst approximation
of the stress at the xed end. This solution is wrong because
the total potential energy fails to achieve the minimum a
quadratic displacement function allows.
The confusion associated with natural boundary conditions arises, perhaps, from the way they are described in
textbooks. They are often said to be the kind of boundary
conditions which do not have to be satised when using a
variational principle. This can therefore be understood by
some users in such a way that if these boundary conditions
had been satised a priori, one might expect a better
approximation. This is wrong. A more precise description
of natural boundary conditions should state that they
should not be imposed, as they will be satised automatically by the variational principle. The imposition of natural
boundary conditions prior to application of the variational
principle will result in no better approximation, in general.
The only exception is that when the assumed displacement
allows the exact solution to be obtained, e.g. if a cubic dis-

Natural boundary condition

=0

Displacement boundary condition u=0


p=kx

E, A
O

x, u

Fig. 1. A straight bar under a linearly distributed axial force.

1964

S. Li / Composites Science and Technology 68 (2008) 19621974

Table 1
Comparison of dierent solutions (f = x/L)
Displacement eld
u  6EA
kL3

Stress at free end


2A
r  kL
2

Stress at xed end


2A
r  kL
2

Total potential energy


P  kEA
2 5
L

Exact solution

1  f3

1

144
 5760

Approximate solution
Wrong solution

1
2 2
3
8 1

1
6

 56
 34

140
 5760
135
 5760

Assumed displacement
function
(a + bf)(1  f)
a(1 + f)(1  f)

3f1  f
f1  f

placement function had been assumed for the same problem instead of a quadratic function. Rigorously speaking,
imposition of natural boundary conditions prior to variation is not a matter of the level of approximation. Rather,
it violates the integrity of the minimum total potential
energy principle and is hence wrong.
Another source of the confusion associated with natural
boundary conditions lies between prescription of boundary
conditions and application of loads. If the free end of the
bar as in the above example is subjected to a traction,
say r0, its contribution to the problem should be found
in the potential of external forces as a part of the total
potential energy. It should not alter the form of the
assumed displacement function in any way. The value of
constants a and b will, of course, vary correspondingly
out of the variational process. Following the approach of
the wrong solution as in Table 1, one would have to ddle
the displacement function in a cumbersome manner only
leading to a worse result. In FE applications, prescription
of boundary conditions and application of loads are two
dierent concepts and hence two dierence processes inside
the code. The former must be satised precisely before
seeking equilibrium and the latter will be satised in the
same way and at the same time as the equilibrium is satised, which is usually approximate.
A further potential source of confusion is associated
with bending problems, where elements, such as beams,
plates and shells, involve displacements which are in a generalised sense, i.e. nodal rotations are considered as displacements. Boundary conditions described in terms of
bending moments and shear forces are generalised traction
boundary conditions and, hence, they are natural boundary conditions in the terminology of variational principles.
3. Sucient and necessary number of boundary conditions
for unit cells
In the literature, it is often found, e.g. in [12], that the
number of boundary conditions prescribed at the same part
of the boundary varies from case to case. Without appropriate justication, it is rather confusing. As a result, incorrect usages are often found, e.g. in [4,10]. According to the
theory of continuum mechanics for the deformation problem of materials in 3-D space, at any given point on the
boundary, three prescribed boundary conditions are
required in any logical combination of displacements and
tractions. For instance, for a boundary perpendicular to

the x-axis, the three boundary conditions can be a prescription of the following
8
>
< u or rx in the x-direction
v or sxy in the y-direction
1
>
:
w or sxz in the z-direction
where u, v and w are displacements in x, y and z directions
respectively and r and s are direct and shear stress components with subscripts in their conventional sense. In the terminology of partial dierential equations, displacement
boundary conditions are boundary conditions of the rst
kind and traction ones are the second kind. There could
be a third kind which is mixture of the rst and second
kinds corresponding to elastic support physically. However, it is irrelevant to the present topic and hence will
not be considered.
When the boundary conditions are imposed in the form
of equations relating the displacements or tractions on one
part of the boundary to those of another part of the boundary, the equation boundary conditions should be imposed
to the following
8
>
< u and rx
2
v and sxy
>
:
w and sxz
Instead of three boundary conditions on one part of the
boundary, there are six boundary conditions for two parts
of the boundary.
Bearing in mind that traction boundary conditions are
natural boundary conditions in conventional FE analyses,
they will be left out of the prescription list. For example, if
part of the boundary is subjected to prescribed
8
>
<u
sxy
>
:
sxz
it is sucient and necessary to prescribe u only on this part
of boundary for the FE analysis. Any non-zero shear tractions should be included as externally applied load rather
than boundary conditions in an FE analysis.
Applying the same argument to equation boundary conditions, it is obvious that equation constraints have to be
imposed to all three displacements to be both sucient
and necessary, whereas equations for tractions can and
should be left out, as in [2,3].

S. Li / Composites Science and Technology 68 (2008) 19621974

Because of the existence of natural boundary conditions


which should not be imposed, the same part of the boundary under dierent loading cases may be subjected to dierent numbers of boundary conditions. This causes
confusions, often in connection to dierence in the nature
of symmetry. The loading and deformation can be symmetric as well as anti-symmetric. Distinguishing one from
another is essential when deriving appropriate boundary
conditions associated with symmetry. For example, when
a deformable body symmetric about the x-plane (x = 0)
is subjected to symmetric loading, e.g. stretching in the xdirection, there is only one boundary condition on the symmetry plane, i.e.
u0
while the two remaining traction boundary conditions,
sxy = sxz = 0, should not be prescribed. However, when
the same body is subjected to anti-symmetric loading, e.g.
shear in the xy plane, there will be two boundary conditions on the symmetry plane, i.e.
v0

and

w0

and there is only one traction boundary condition, rx = 0,


in this case, which should not be imposed.
To conclude this section, it is clear that at a boundary of
a unit cell, the number of boundary conditions to be
imposed before a nite element analysis can be conducted
is not denite. It depends on the nature of the symmetries
adopted in the denition of the unit cell. However, one
thing remains denite when introducing boundary conditions. It is that only displacement boundary conditions
should be imposed, not the natural boundary conditions.
When bending is involved, for instance in classical laminate
theory, displacements should include all generalised displacements, i.e. translational displacements and rotational
displacements. The natural boundary conditions in this
case are associated with membrane forces, bending
moments and transverse shear forces.
4. Selection of unit cells and their implications
For arguments sake, a microstructure in 2-D space of a
square layout as shown in Fig. 2a is considered rst, which
can be perceived as, but not restricted to, a transverse
cross-section of a UD composite or an in-plane pattern
of a textile composite. The square grid employed to tessellate the pattern in Fig. 2a can be shifted in the x and y
directions by any arbitrary distances without violating
the translational symmetries present in this pattern. The
tessellations as shown in Fig. 2b are perfectly legitimate
arrangements. However, from the meshing point of view
when nite elements are employed to analyse the unit cell,
it would be preferable to choose a tessellation without truncating the inclusion if possible. The particular arrangement
chosen in Fig. 2a is for the purpose of comparison with the
subsequent case to be presented. As the only symmetries
available are translations, in the x and y directions, respec-

1965

tively, the unit cell as shown in Fig. 2c will have to be subjected to equation boundary conditions as given in [2]
relating displacements on opposite sides of the unit cell
while ignoring the traction conditions.
However, if the material under consideration allows
one to idealise it into a microstructure as shown in
Fig. 3a, the repetitive cell as shown in Fig. 3b would be
the unit cell of smallest size if only translational symmetries are employed. Obviously, the size of the unit cell
can be reduced to that as shown in Fig. 3c after the available reectional symmetries about x and y axes have been
utilised. As a result, boundary conditions can be obtained
without equations relating the displacements on the opposite sides of the cell. Having used the reectional symmetries, one needs to bear in mind two issues associated with
a unit cell, as shown in Fig. 3c, which can be easily overlooked. Firstly, the microstructure of the material the unit
cell in Fig. 3c is given in Fig. 3a not as shown in Fig. 2a
although the appearance of the unit cells in Figs. 2c and
3c look identical. Secondly, some macroscopic strain
states, in particular, associated with shear, are anti-symmetric under the reectional symmetry transformations.
Appropriate considerations should be given to the antisymmetric nature when boundary conditions are derived
from the symmetry transformations. As a result, the number of boundary conditions for some loading cases would
be dierent from those for the other cases.
Sometimes, for patterns like the one shown in Fig. 4a,
unit cells of only a quarter of the size, as shown in
Fig. 4c, are seen in the literature. It, in fact, results from
exactly the same considerations as in Fig. 3. A quarter is
sucient only because the presence of the reectional symmetries about the x and y axes within the repetitive cell as
shown in Fig. 4b. Many regular shapes of the inclusion
possess these symmetries, such as diamond, rectangle and
circle, but there are also shapes which do not possess such
symmetries, e.g. that in Fig. 2a. The conditions implied by
a quarter size unit cell are the existence of reectional symmetries, as in Fig. 3. The boundary conditions for a quarter
size unit cell should be derived in exactly the same way as
those for the unit cell in Fig. 3c.
The ultimate unit cells obtained from Figs. 2 and 3 share
the same appearance. However, they are subject to dierent
boundary conditions and are associated with dierent
microstructures. An obvious consequence of the dierence
in the microstructure is that the one in Fig. 3a is macroscopically orthotropic while that in Fig. 2a is not necessarily the case, as will be seen later through an example. Users
of unit cells should be aware of the dierence and decide if
the dierence has any signicance for their particular applications while choosing the unit cell to be employed.
Another regular pattern often encountered in the literature is hexagonal. Arguments, similar to those above for
the square pattern, apply to a large extent. The only dierence is that there are more ways to express the translational
symmetries as discussed fully in [1,2]. Whether the repetitive cell, e.g. the rectangle or any of the hexagons shown

1966

S. Li / Composites Science and Technology 68 (2008) 19621974

Fig. 2. Square packing.

Fig. 3. Square packing with further reectional symmetries.

Fig. 4. An equivalent layout to that in Fig. 3.

in Fig. 5, used to express the translation symmetries, can be


further reduced in size depends on the existence of other
symmetries, including reectional and rotational. Without
such additional symmetries, the smallest size would be a

complete hexagon as shown in Fig. 5, if one is prepared


to employ translations in directions that are not perpendicular to each other. Otherwise, to involve for orthogonal
translations in x and y directions only, one will have to deal

S. Li / Composites Science and Technology 68 (2008) 19621974

1967

state. The latter is certainly more desirable in most cases


as a macroscopically uniaxial stress state is what one
would need in order to measure an eective property of
the composite.
The translational symmetries require:
ujxbx  ujxbx 2bx e0x
vjxbx  vjxbx 0

and

sxz jxbx  sxz jxbx 0

wjxbx  wjxbx 0
A repetitive cell if only translations
in x and y directions are allowed

syx jyby  syx jyby 0


and

ry jyby  ry jyby 0
syz jyby  syz jyby 0

wjyby  wjyby 0

under translation in y-direction


ujzbz  ujzbz 2bz c0xz
vjzbz  vjzbz

2bz c0yz

wjzbz  wjzbz

5. Boundary conditions for a unit cell from 3-D


microstructure with reectional symmetries

and

szx jzbz  szx jzbz 0


szy jzbz  szy jzbz 0
rz jzbz  rz jzbz 0

2bz e0z

under translation in z-direction

Consider the case illustrated in Fig. 3 in a 3-D scenario.


The boundary conditions can be derived in general as follows, assuming the periods of translational symmetries in
the x, y and z directions are 2bx, 2by and 2bz, respectively.
This is general enough to encapsulate regular packing layouts such as simple cubic with bx =pby = bz = b, body centred cubic with bx by p
b z 4b= 3, face centred cubic
with bx by pbz 4b= 2 and close packed
hexagonal
p
with bx 4b= 2, by 2b and bz 2 3b, respectively,
where b is the characteristic radius, i.e. the radius of largest
sphere which can be accommodated, as dened in [3] for
each packing. Obviously, the size of the unit cells for body
centred cubic, face centre cubic and close packed hexagonal
packing are no longer as compact as in [3] where each
packing was considered individually in order to achieve
the most compact unit cell.
Assume that there exists an intermediate repetitive cell
equivalent to that in Fig. 3b which can represent the material fully using translational symmetries only, and that this
cell is dened in the domain
 bx 6 x 6 bx
 by 6 x 6 by

2by e0y

vjyby  vjyby

with a unit cell having a bigger size, as shown in the rectangle in Fig. 5, which is obviously not a unique choice. When
analysing these unit cells, in general, equation boundary
conditions will have to be employed.

under translation in x-direction


ujyby  ujyby 2by c0xy

Fig. 5. Hexagonal packing.

rx jxbx  rx jxbx 0
sxy jxbx  sxy jxbx 0

 bz 6 x 6 bz
The material is subjected to a set of macroscopic
strains fe0x ; e0y ; e0z ; c0yz ; c0zx ; c0xy g which can be introduced as
six extra degrees of freedom (d.o.f.) in an FE analysis,
e.g. as six individual nodes, each having a single d.o.f.,
or six degrees of freedom of a special node. Each of them
can be prescribed to achieve a macroscopically uniaxial
strain state. Alternatively, upon any of these extra
d.o.f.s, a concentrated force can be applied in order to
impose a macroscopic stress while leaving the others free
in order to produce a macroscopically uniaxial stress

The form of many of the above equations is not unique,


especially, those associated with shear, depending on the
way rigid body rotations are constrained. For instance,
the rst two equations in (5) can be replaced by
ujyby  ujyby 0

vjyby  vjyby 2by e0y 2bx c0xy

ujyby  ujyby by c0xy

vjyby  vjyby 2by e0y bx c0xy

without aecting the results. The lack of uniqueness here


also contributes to the likelihood of confusion when introducing boundary conditions for unit cells.
The use of boundary conditions derived from translational symmetries alone as shown in (4)(6), has been illustrated fully in [3]. It is of interest in this paper to derive
appropriate boundary conditions, when further reectional
symmetries are present in the intermediate repetitive cell as
dened above.
To apply further reectional symmetries about x, y and
z planes, the problem has to be considered separately for
individual loading cases expressed in terms of macroscopic
stresses fr0x ; r0y ; r0z ; s0yz ; s0zx ; s0xy g, as presented in the following
subsections.
In deriving the boundary conditions in the following
subsections, the principle of symmetries will be employed.
It states that symmetric stimuli, i.e. loads, result in symmetric responses, including displacements, strains and stresses,
while anti-symmetric stimuli produce anti-symmetric
responses.
5.1. Under r0x
Consider rst the x-faces of the unit cell, i.e. those perpendicular to the x-axis. r0x as a stimulus is symmetric
under reection about x-plane (perpendicular to x-axis).
Responses v, w and rx are symmetric while u, sxy and sxz

1968

S. Li / Composites Science and Technology 68 (2008) 19621974

are anti-symmetric. On the symmetry plane (x = 0), the


symmetry conditions require
ujx0 ujx0
vjx0 vjx0
wjx0 wjx0

rx jx0 rx jx0
and

sxy jx0 sxy jx0


sxz jx0 sxz jx0

and

wjx0 wjx0 ! free

while traction boundary conditions sxyjx=0 = sxzjx=0 = 0


are natural boundary conditions and should not be imposed. The same argument will be adopted hereafter without referring to their nature of being natural boundary
conditions.
Considering the opposite faces at x = bx, and applying
the symmetry condition, one has

wjxbx wjxbx

rx jxbx rx jxbx
sxy jxbx sxy jxbx
sxz jxbx sxz jxbx

wjxbx wjxbx ! free

rx jxbx rx jxbx ! free


and

sxy jxbx 0
sxz jxbx 0
11

The only remaining boundary condition on side x = bx is


ujxbx bx e0x

syz jy0 0
14

Leaving the natural boundary conditions aside, a single


boundary condition on side y = 0 is obtained
vjy0 0

15

The symmetry conditions on the two opposite faces at


y = by require
ujyby ujyby
vjyby vjyby

syx jyby syx jyby


and

wjyby wjyby

ry jyby ry jyby

16

syz jyby syz jyby

The above in conjunction with (5) lead to


ujyby ujyby ! free
vjyby

syx jyby 0

by e0y

and

ry jyby ry jyby ! free


syz jyby 0

wjyby wjyby ! free

17
10

In conjunction with the translational symmetry conditions


as given in (4), one obtains
ujxbx bx e0x
vjxbx vjxbx ! free

and

syx jy0 0
ry jyby ry jyby ! free

13

which can be rewritten as

wjy0 wjy0 ! free

sxz jx0 0

ujx0 0

and

syx jy0 syx jy0


ry jy0 ry jy0
syz jy0 syz jy0

and

vjy0 0
rx jx0 rx jx0 ! free
sxy jx0 0

In the above, the conditions vjx=0, wjx=0 and rxjx=0 do not


yield any constraints and they should hence be left free, as
indicated by ! free. The same notation will be adopted
for the rest of the paper. In (8), there are three boundary
conditions in eect. However, only the displacement
boundary condition should be imposed on side x = 0, i.e.

ujxbx ujxbx
vjxbx vjxbx

vjy0 vjy0
wjy0 wjy0

ujy0 ujy0 ! free

which can be re-expressed as


ujx0 0
vjx0 vjx0 ! free

ujy0 ujy0

12

The boundary condition above introduces an extra


d.o.f. e0x into the system. In an FE analysis, it can be prescribed in order to prescribe a macroscopically uniaxial
strain state e0x . However, if one wishes to impose a macroscopically uniaxial stress state r0x , an appropriate concentrated force can be applied to this d.o.f. The
macroscopically eective stress r0x can be worked out from
the concentrated force easily as discussed in [3], while the
nodal displacement at this extra d.o.f. gives the eective
macroscopic strain e0x directly.
Consider now the y-faces. The stimulus r0x is also symmetric under reection about y-plane (perpendicular to yaxis). Responses u, w and ry are symmetric while v, syx
and syz are anti-symmetric about y-plane. Hence, on the
symmetry plane (y = 0) the symmetry conditions require

which result in a single boundary condition for side y = by


vjyby by e0y

18

The boundary condition above introduces another extra


d.o.f. e0y into the system. To impose a uniaxial macroscopic
stress r0x , the d.o.f. e0y should be left free, so that r0y 0. The
nodal displacement at the extra d.o.f. e0y gives this macroscopic strain directly produced by r0x as a result of Poisson
eect, from which the eective Poisson ratio can be easily
evaluated. On the other hand, this extra d.o.f. can be prescribed to impose a macroscopic strain e0y , or a concentrated force can be applied to it to impose a macroscopic
stress r0y .
Applying the same arguments to the z-faces, the boundary conditions on sides z = 0 and z = bz can be obtained as
wjz0 0

and

wjzbz bz e0z ; respectively


e0z

19

was introduced through translawhere the extra d.o.f.


tional symmetry conditions (6). To impose a macroscopic
stress r0x alone, e0z should be left free, so that r0z 0. The
nodal displacement at the extra d.o.f. e0z gives this macroscopic strain directly. Alternatively, it can be prescribed
accordingly in order to impose a macroscopic stress or
strain in this direction.
To summarise, under a macroscopic stress r0x , the
boundary conditions on the three pairs of the sides of
the unit cell are given by (9), (12), (15), (18) and (19),
namely

S. Li / Composites Science and Technology 68 (2008) 19621974

ujx0 0 and ujxbx bx e0x


vjy0 0 and vjyby by e0y
wjz0 0

and

wjzbz

ujxbx 0
20

bz e0z

where the extra d.o.f. e0x is subjected to a concentrated force


associated with r0x , while e0y and e0z should be left free to produce a macroscopically uniaxial stress state r0x .
5.2. Under

1969

r0y

24

Consider now the pair of sides parallel to the y-plane.


The stimulus s0yz is anti-symmetric about y-plane (y = 0).
The responses u, w and ry are symmetric while v, syx and
syz are anti-symmetric. Hence, on the symmetry plane,
the symmetry conditions require
ujy0 ujy0
vjy0 vjy0
wjy0 wjy0

syx jy0 syx jy0


ry jy0 ry jy0
syz jy0 syz jy0

and

25

With similar considerations as given above, the boundary conditions for the unit cell under r0y are identical to
those in (20). The only dierence is that it should be the
extra d.o.f. e0y that is subjected to a concentrated force associated with r0y , while e0x and e0z are left free to produce a
macroscopically uniaxial stress state r0y . The nodal displacements at those extra d.o.f.s give the corresponding
macroscopic strains directly.

From (26), the boundary conditions on y = 0 are obtained


as

5.3. Under r0z

ujy0 wjy0 0

The boundary conditions are again identical to those in


(20). However, the extra d.o.f. e0z should be subjected to a
concentrated force associated with r0z , while e0x and e0y are
left free to produce a macroscopically uniaxial stress state
r0z .
5.4. Under s0yz
The nature of shear stresses is slightly more complicated
than their direct counterparts. With respect to a reectional
symmetry, one of the three shear components is symmetric
while other two are anti-symmetric. Under the reection
about the x-plane, the stimulus s0yz is symmetric. The
responses v, w and rx are symmetric while u, sxy and sxz
are anti-symmetric. Hence, on the symmetry plane, x = 0,
the symmetry conditions require
ujx0 ujx0
vjx0 vjx0
wjx0 wjx0

sxy jx0 sxy jx0


sxz jx0 sxz jx0

21

wjx0 wjx0 ! free

and

rx jx0 rx jx0 ! free


sxy jx0 0

vjy0 vjy0 ! free

ry jy0 0

and

wjy0 0

26

syz jy0 syz jy0 ! free

27

Notice that there are two displacement boundary conditions in this case as opposed to the x = 0 plane on which
there is only one boundary condition as given in (24). They
have to be imposed in order to dene the unit cell properly
under this loading condition. There is one traction boundary condition ryjy=0 = 0 which has been ignored as a natural boundary condition.
Applying the reectional symmetry to the two opposite
faces at y = by, one obtains
ujyby ujyby
vjyby vjyby

syx jyby syx jyby


and

wjyby wjyby

ry jyby ry jyby

28

syz jyby syz jyby

Combining the above with (5), one obtains


ujyby 0

syx jyby syx jyby ! free


and

wjyby 0

ry jyby 0
syz jyby syz jyby ! free
29

ujyby wjyby 0

sxz jx0 0
22

Ignoring the traction boundary conditions, the only


boundary condition to be imposed is
ujx0 0

syx jy0 syx jy0 ! free

which lead to the following boundary conditions for side


y = by

which can be rewritten as


ujx0 0
vjx0 vjx0 ! free

ujy0 0

vjyby vjyby ! free

rx jx0 rx jx0
and

Rewrite

23

On the opposite faces at x = bx, the reectional symmetry conditions are similar to (21) but on x = bx instead
of x = 0. In conjunction with the translational symmetry
conditions (4), they lead to the boundary condition

30

Similarly, the boundary conditions on z = 0 and z = bz


can be obtained, bearing in mind that s0yz is anti-symmetric
about z-plane
ujz0 vjz0 0
ujzbz 0

and

vjzbz bz c0yz

31

where the extra d.o.f. c0yz is introduced through the translational symmetry conditions (6), which can be associated
with ujzbz instead of vjzbz if the rigid body rotation of
the unit cell is constrained dierently. There will be no difference whatsoever as far as the deformation is concerned,

1970

S. Li / Composites Science and Technology 68 (2008) 19621974

provided that it has been dealt with correctly. The same applies to the consideration of the two subsequent loading
cases without further explanation. To impose a macroscopic stress s0yz , a concentrated force can be applied to
the d.o.f. c0yz . The nodal displacement at c0yz , obtained after
the analysis, gives the corresponding macroscopic strain directly. Since w is not constrained on z = 0 and z = bz, these
faces do not have to remain at after deformation.
As a summary, all boundary conditions for the unit cell
under macroscopic stress s0yz are as follows
ujxbx 0
ujx0 0
ujy0 wjy0 0 ujyby wjyby 0
32
ujz0 vjz0 0 ujzbz 0 & vjzbz bz c0yz
Notice that there are dierent numbers of conditions on
dierent sides. In general, symmetry produces one condition while anti-symmetry results in two. The same applies
to the subsequent shear loading cases where details of the
derivation are omitted.
5.5. Under s0xz
After considering all symmetry conditions, the boundary conditions for the unit cell subjected to this loading
condition can be obtained as
vjx0 wjx0 0 and vjxbx wjxbx 0
vjy0 0 and vjyby 0
ujz0 vjz0 0

and

ujzbz

bz c0xz

33

& vjzbz 0

5.6. Under s0xy

wjz0 0

and

and

6. 2-D problems
The 3-D presentation of boundary conditions readily
degenerates into 2-D problems, such as plane stress, plane
strain, generalised plane strain problems and anticlastic
problems. They apply to the rectangular unit cell obtained
from the hexagonal layout shown in Fig. 6, as well as to the
square one shown in Fig. 3. They are given as follows without detailed derivations, in the yz plane, for example.
6.1. Under r0y and r0z
When a 2-D unit cell, in the yz plane, is subjected to
macroscopic stresses r0y or r0z , the boundary conditions
are the same as follows
vjy0 0

and

vjyby by e0y

wjz0 0

and

wjzbz bz e0z

35

The dierence between the implementations to achieve


these two macroscopically uniaxial stress states is a concentrated force that needs to be imposed to the extra d.o.f. e0y
or e0z , respectively.
6.2. Under s0yz

The corresponding boundary conditions are


vjx0 wjx0 0 and vjxbx wjxbx 0
ujy0 wjy0 0

boundary conditions have to be imposed, as stated in the


Kirchho theorem [11]. However, as far as nite element
applications are concerned, which are based on some variational principles, the uniqueness of a nite element solution can be guaranteed without the traction boundary
conditions which are implied and hence satised automatically by the variational principle concerned [11].

The boundary conditions for a unit cell under macroscopically uniaxial shear stress s0yz in the yz plane are as
follows

ujyby by c0xy & wjyby 0

wjzbz 0
34

It has been shown in this section that, using reectional


symmetries additional to translational ones, unit cells can
be formulated with rather conventional boundary conditions (20) for loading in terms of microscopic stress r0x , r0y
or r0z , (32) for s0yz , (33) for s0zx and (34) for s0xy which do not
involve equations associating the displacements on opposite
sides of the unit cell. The price to pay is the fact that under
dierent loading conditions, dierent boundary conditions
may have to be employed. However, equation boundary
conditions associated with extra d.o.f.s will have to remain.
As these extra do.f.s are not really a physical part of the
mesh in terms of geometry, they can be placed any where
and hence should not cause any problem. Most commercial
FE codes have provisions to incorporate such equation
boundary conditions.
To conclude this section, it is worthwhile pointing out
that in a general sense, in order to guarantee the uniqueness
of an elasticity solution, both displacement and traction

wjy0 0

and

vjz0 0 and

wjyby 0

36

vjzbz bz c0yz

A concentrated force at the extra d.o.f. c0yz delivers the macroscopically uniaxial shear stress states.

Smallest unit cell possible but with


equation boundary conditions

Simplest unit cell without


equation boundary conditions

Fig. 6. Hexagonal packing with reectional symmetries.

S. Li / Composites Science and Technology 68 (2008) 19621974

As on boundary y = 0, displacement v is not constrained


in any way and there is no restriction whether the side
should remain straight after deformation. The same applies
to all other sides.
6.3. Generalised plane strain problem and macroscopically
uniaxial stress state r0x
For generalise plane strain problems, an extra d.o.f. e0x ,
in addition to e0y , e0y and e0yz , has to be introduced, which
can be dealt with in the same manner as other extra
d.o.f.s corresponding to macroscopic strains. This extra
d.o.f. should be left free when applying macroscopically
uniaxial stress states r0y and r0z but constrained for s0yz , as
s0yz is anti-symmetric under the reectional symmetry while
e0x is symmetric. It should be pointed out that neither plane
stress nor plane strain is capable of reproducing the macroscopically eective uniaxial stress states under which eective properties are measured experimentally according to
their denitions. For UD composites, the generalised plane
strain problem is the only 2-D formulation which is capable of achieving macroscopically eective uniaxial stress
state.
When applying a macroscopically uniaxial stress state
r0x , the boundary conditions are the same as in (35). However, the concentrated force should be applied to the extra
d.o.f. e0x while leaving e0y and e0z free. The nodal displacements at these extra d.o.f.s e0x , e0y and e0z give the macroscopic strains directly, which can be used to work out the
eect of Youngs modulus in the x-direction and Poisson
ratios mxy and mxz.
6.4. Under s0xz and s0xy in an anticlastic problem
The anticlastic problem in the yz plane involves only
one displacement u. When a macroscopically uniaxial shear
stress s0xz is applied, from Section 5.5, the boundary conditions for the unit cell can be obtained as
ujz0 0

and

ujzbz bz c0xz

37

while edges y = 0 and y = by are left free. A concentrated


force can be applied to the extra d.o.f. c0xz to deliver a macroscopically uniaxial stress state s0xz and the nodal displacement at c0xz gives this macroscopic strain directly.
Similar arguments apply to the macroscopically uniaxial
stress state s0xy and from Section 5.6, the corresponding
boundary conditions are obtained as
ujy0 0

and

ujyby by c0xy

38

while edges z = 0 and z = bz are left free.


It is a little disappointing that nite elements for the
anticlastic problem are not usually found in many popular
commercial FE codes although they are meant to be simpler than those for other deformation problems. Some
users employed 3-D elements. However, based on the common mathematical governing equation with some other
physical problems, such as heat transfer, an analogy can

1971

be drawn so that elements for these problems can be


adapted for the analysis of the present anticlastic problem.
More details can be found in [2]. This can keep the analysis
in 2-D. Analysing a 2-D problem in 3-D is bad in many
ways as discussed in [2].
7. Deformation of the sides of unit cells
As examples of the applications of the boundary conditions derived above, several examples of unit cells have
been analysed. Particular attention in this section will be
paid to the deformation of the sides of the unit cell, which
do not always remain at/straight after deformation, when
boundary conditions have been derived and imposed
rigorously.
7.1. 3-D unit cell for particle reinforced composites with
simple cubic particle packing
A unit cell for simple cubic packing was presented in [3]
and a mesh was generated with a spherical geometry for
particle and appropriate constituent material properties
in the examples. The same problem will be analysed again
here while taking advantage of additional reectional symmetries in the unit cell as presented in [3]. As a result, only
an octant is required as the unit cell for the present analysis. Without losing generality, only macroscopically uniaxial stress states r0x and s0yx are examined here. The boundary
conditions are as given in (20) and (32), respectively.
Under the macroscopically uniaxial stress state r0x , the
results are identical to the results given in [3] when the corresponding octant is taken out of the unit cell in [3] for
comparison. This should not undermine the unit cells formulated in [3] as the present ones are only applicable if
the particle possesses required reectional symmetries. It
can be noted that all sides remain at after deformation.
This is imposed by the boundary conditions, as in (20).
The same is expected when the unit cell is under other macroscopic stress r0y or r0z or any combination of r0x , r0y and
r0z . However, this should not be taken for granted
universally.
When the unit cell is subjected to a macroscopically
uniaxial shear s0yz , the von Mises stress contour plot is
shown in Fig. 7. The results obtained here also agree identically with those in [3], although the corresponding contour plot was not shown in [3]. According to the
boundary conditions given in (32), only the x-faces, i.e.,
x = 0 and x = bx, have to remain at after deformation,
while the boundary conditions on the remaining faces
impose no restriction in this regard. As a result, the
remaining two pairs of faces, i.e. y-faces and z-faces, do
not have to remain at. Fig. 7 illustrated the curved trend
for these two faces. The curvature of these faces reduces
as the disparity of properties between the particle and
the matrix reduces. In fact, at faces are expected when
the particle and matrix share identical properties. The
same observation applies to the unit cell when it is

1972

S. Li / Composites Science and Technology 68 (2008) 19621974

Fig. 7. Deformation of a unit cell for particle reinforced composite with a


simple cubic packing.

subjected to either of the two remaining macroscopic


stress states, s0zx and s0xy . The only dierence is that the
faces remaining at after deformation become y-faces
and z-faces instead, respectively, while other faces warp
after deformation, in general.
It should be noted that, while (20) applies to any combination of macroscopic direct stress states, (32) is only applicable to s0yz . Boundary conditions (33) and (34) have to be
used for s0zx and s0xy , respectively. One has to reconsider to
use the unit cell as proposed in [3] if any combination of
s0yz , s0xy and s0xy has to be applied.
7.2. 2-D unit cell for UD bre reinforced composites with
square bre packing
Applying boundary conditions as given in Section 6, 2D unit cells can be analysed pertinent to UD bre reinforced composites with circular bre cross-section. The
examples here correspond to the cases published in [2].
However, unit cells of smaller sizes have been used here,
taking advantage of reectional symmetries present in the
problem. As in [2], generalised plane strain conditions
apply to the problem for macroscopic stress states r0x , r0y ,
r0z and s0yz , the x-axis being along the bre while the anticlastic problem for macroscopic stress states s0zx and s0xy

can be analysed using heat transfer as an analogy to avoid


3-D modelling.
Once again, perfect agreement in results can be obtained
between the unit cells presented here and those in [1,2] for
both square packing and hexagonal packing. Similar observations to those in their 3-D counterparts of the previous
subsection can be made on the deformation of the sides
of the unit cells. Under direct macroscopic stress states,
all the sides of a square unit cell remain straight after deformation. However, under other loading conditions or for
hexagonal unit cells, sides may not remain straight after
deformation as shown in Fig. 8 unless the bre has the
same elastic properties as the matrix. For macroscopic
stress states s0zx and s0xy , the sides may look straight from
the perspective along the x-axis (bre direction) but the
yz plane itself warps into a curved surface. The sides are
in fact curved in space.
The objective of the examples in Figs. 7 and 8 is to demonstrate that the sides of the properly established unit cells
do not always remain at/straight after deformation. Intuitively formulated unit cells assuming at/straight sides
after deformation are incorrect in general. The underlying
principle for the formulation of unit cells is the principle
of symmetries, while intuition is often subject to limitations. Once the symmetries present in the microstructure
in the material have been made proper use of, correct unit
cells can be obtained. The boundary conditions derived in
this way will not result in any boundary eect as presented
in [4]. In fact, if one analyses an assembly of cells, e.g. the
one as shown in Fig. 3b, the results obtained will be exactly
the same as those from the analysis of that in Fig. 3c, provided that the boundary conditions have been imposed correctly in both cases.
8. Eects of microstructures implied by dierent unit cells
The unit cells as sketched in Figs. 2c and 3c have the
same geometry but are subjected to dierent boundary conditions. They are therefore dierent unit cells. The dierences do not always lead to dierent results, especially
when the inclusions (bre or particle) possess sucient
reectional symmetries. However, it could be badly wrong

Fig. 8. Curved edges in deformed unit cells: (a) square unit cell under macroscopic transverse shear s0yz (1 MPa), (b) hexagonal unit cell under macroscopic
transverse shear s0yz (1 MPa), and (c) hexagonal unit cell under macroscopic transverse tension r0y (1 MPa).

S. Li / Composites Science and Technology 68 (2008) 19621974

if one is used blindly in place of the other, in particular,


when the inclusions do not show the required symmetries.
The purpose of this section is to illustrate such dierences,
as nothing in the literature seems to suggest that the implications have been fully recognised.
Assume a 2-D microstructure involving inclusions of an
elliptical cross-section inclined at 30. The ellipse is of 2:1
aspect ratio and occupies a volume fraction of 40%. The
elastic properties of the inclusion and the matrix are
assumed as listed in Table 2. The same mesh as shown in
Fig. 9 will be used for both unit cells corresponding to
microstructures shown in Figs. 2 and 3, respectively.
The von Mises stress contour plots at deformed congurations under macroscopic stress states r0z and s0yz
(=1 MPa) are presented and compared in Fig. 10. It is
obvious that dierently assumed microstructures, as
implied by the two dierent unit cells result in dierent
stress distributions microscopically. The dierences are
even more pronounced when eective properties are
Table 2
Properties of the constituents for the unit cells
Properties

Inclusion

Matrix

E (GPa)
v

10
0.2

1
0.3

1973

Fig. 9. Mesh for the unit cell with an reinforcement of an elliptical crosssection.

extracted from these unit cells and compared as listed in


Table 3 where properties gij are dened as the ratio of shear
strain cj to the direct strain ei when the unit cell is subjected
to a macroscopically uniaxial direct stress state ri, and lij is
the ratio of shear strain cj to shear strain ci when the unit
cell is subjected to a macroscopically pure shear stress state
si [13]. These properties in the materials principal axis vanish for orthotropic material, as is the case for the unit cell
corresponding to Fig. 3c. Material represented by the unit
cell corresponding to Fig. 2c is not orthotropic but mono-

Fig. 10. Deformation and von Mises stress contour plots for the unit cell corresponding to (a) Fig. 2c under r0z , (b) Fig. 3c under r0z , (c) Fig. 2c under s0yz ,
and (d) Fig. 3c under s0yz .

1974

S. Li / Composites Science and Technology 68 (2008) 19621974

Table 3
Eective properties corresponding to the unit cells
Eective
properties

Unit cell in the sense of


Fig. 2c

Unit cell in the sense of


Fig. 3c

E1 (GPa)
E2 (GPa)
E3 (GPa)
G23 (GPa)
G13 (GPa)
G12 (GPa)
m23
m13
m12
g14
g24
g34
l56

4.603
2.183
1.864
0.6803
0.7016
0.9180
0.3586
0.2606
0.2436
0.01377
0.08768
0.04289
0.1143

4.605
2.372
1.890
0.6710
0.7144
1.0288
0.3663
0.2612
0.2378
0
0
0
0

clinic in general relative to the coordinate system as shown


in Fig. 9. The dierences, as illustrated here, will disappear
when the ellipse is replaced by a circle but this is not a sufcient reason for ignoring the dierences. When a unit cell
is used, the user ought to be clear about the implications of
the unit cell adopted on the microstructure of the materials,
e.g. the one in Fig. 2 or the one in Fig. 3, which are apparently dierent enough from each other.
9. Conclusions
Unit cells for micromechanical analyses have to be
introduced with due consideration of the microstructures
implied by the unit cell. Boundary conditions for unit cells
representing microstructures of periodic patterns should
follow entirely from the symmetries present in the microstructure represented by the unit cell, rather than from
ones intuition. The symmetries include translations, reections and rotations. Using translations alone leads to
boundary conditions in the form of equations relating displacements on opposite sides of the boundary of the unit
cell. Further use of reection symmetries, if they exist,
can avoid such equation boundary conditions, making

the application of boundary conditions easier. However,


users must be aware of the dierences in the microstructures implied by the boundary conditions for the unit cell.
Although unit cells may look identical geometrically, dierent boundary conditions imposed would associate the unit
cell with rather dierent microstructures. It has been illustrated in this paper that such dierences in the microstructures may result in rather dierent eective properties of
the composites represented by the unit cells.
References
[1] Li S. On the unit cell for micromechanical analysis of bre-reinforced
composites. Proc Roy Soc Lond A 1999;455:81538.
[2] Li S. General unit cells for micromechanical analyses of unidirectional composites. Composites A 2001;32:81526.
[3] Li S, Wongsto A. Unit cells for micromechanical analyses of particle
reinforced composites. Mech Mater 2004;36:54372.
[4] Fang Z, Yan C, Sun WA, Shokoufandeh, Regli W. Homogenization
of heterogeneous tissue scaold: a comparison of mechanics, asymptotic homogenization, and nite element approach. Appl Bionics
Biomech 2005;2:1729.
[5] Peng X, Cao J. A dual homogenisation and nite element approach
for material characterisation of textile composites. Composites, Part
B 2002;33:4556.
[6] Llorca J, Needleman A, Suresh S. An analysis of the eects of matrix
void growth on deformation and ductility in metalceramic composites. Acta Metall Mater 1991;39:231735.
[7] Adams DF, Crane DA. Finite element micromechanical analysis of a
unidirectional composite including longitudinal shear loading. Comput Struct 1984;18:115365.
[8] Nedele MR, Wisnom MR. Finite element micromechanical modelling
of a unidirectional composite subject to axial shear loading.
Composites 1994;25:26372.
[9] Zahl DB, Schmauder S, McMeeking RM. Transverse strength of
metal matrix composites reinforced with strongly bonded continuous
bres in regular arrangement. Acta Metall Mater 1994;42:298397.
[10] Yeh JR. Eect of interface on the transverse properties of composites.
Int J Solids Struct 1992;29:2493502.
[11] Fung YC, Tong P. Classical and computational solid mechanics.
London: World Scientic; 2001.
[12] Aitharaju VR, Averill RC. Three-dimensional properties of woven
fabric composites. Compos Sci Technol 1999;59:190111.
[13] Lekhniskii SG. Theory of elasticity of an anisotropic elastic body. San
Francisco: Holden Day; 1963.

S-ar putea să vă placă și