Professional Documents
Culture Documents
www.elsevier.com/locate/compositesa
Abstract
A methodology is established in this paper for the micromechanical finite element analysis of unidirectionally fibre-reinforced composites
having fibres distributed at random over the transverse cross-section, a characteristic that most UD composites in engineering applications
bear. Regular packing is often assumed in micromechanical analyses. While this simplifies the analysis greatly, it will never provide
understanding of aspects of material’s behaviour directly related to the distributions of fibres, such as clustering of fibres over the transverse
cross-section. Within the limited results available in the literature, it appears to be difficult to assess the errors due to unrealistic boundary
conditions prescribed to the representative volume element (RVE) under consideration, as a uniform displacement or traction along the
border of a regularly shaped RVE would not be realistically achievable in general. According to the St Venant’s principle, one would expect
the effects of such incorrectly prescribed boundary conditions affect only a limited zone next to the boundary. A quantitative analysis on the
effects of such incorrectly prescribed boundary conditions is provided in this paper using a numerical means. It has been found that the error
due to incorrect boundary conditions diminishes over a couple of times of the distance of fibre spacing (centre-to-centre). It is generally true
that uniformly prescribed displacement gives a short decay length than that associated with uniformly prescribed traction. Applying this to
the analysis of UD composites with fibres distributed at random over the transverse cross-section, an RVE can be introduced and analysed,
within which a sub-domain is defined appropriately distant to the boundary. The deformation and stresses obtained in this sub-domain can be
considered to be free from the effects of incorrectly prescribed boundary conditions for the RVE analysed and the results extracted only from
this sub-domain can, therefore, be considered correct representation of the behaviour of materials of this type. Examples of application of
such a procedure are shown in the paper, through which some interesting perspectives of UD composites with fibres distributed at random
over the transverse cross-section have been revealed. A significant improvement in predicted transverse Young’s modulus has been obtained
as compared with those by assuming regular packing in relation to experimental data. Rather different hardening characteristics between the
UD composites with regular and random fibre packing have been found in the plastic regime.
q 2005 Elsevier Ltd. All rights reserved.
unified. However, there are also aspects of the composites’ micromechanical analysis of UD composites with fibres
behaviour which cannot be reasonably represented with a distributed at random over the transverse cross-section
regular system, such as effects of fibre clustering and using an RVE assisted with appropriate treatment and
contiguity as well as discrepancies between experimental results processing technique.
data obtained from composites having random fibre The proposed approach is based on the St Venant’s
distribution and predictions based on regular fibre packing. principle, according to which the results obtained in regions
A stress analysis of UD composites of randomly distributed within the RVE sufficiently distant from the boundary where
fibres over the transverse cross-section is of great incorrect boundary conditions are prescribed should be
significance in understanding these. Through a recent close enough to the exact solution as if the boundary
study [10], it has been found that the stress state at the conditions had been prescribed correctly, provided that the
contacting point of fibres remains regular provided that forces associated with such approximate boundary con-
both the fibre and matrix are of finite elastic moduli, ditions are statically equivalent to those in the exact system.
although a high stress concentration is present. The However, the St Venant’s principle itself does not tell how
implication is that the behaviour of the composite, such distant is sufficiently distant. In fact, it varies from problem
as the onset of any inelastic deformation, is still determined to problem in general and the decay length will have to be
by the local stresses rather than any non-local measure determined in an ad hoc manner for a given type of
such as stress intensity factor even in the present of fibre problems. To proceed from what has been achieved in the
contiguity. There is a lack of understanding of the effects literature, the following strategy has been adopted in this
of random fibre distribution over the transverse cross- paper in order to establish a systematic procedure.
section of UD composites, given the efforts made in this A composite with regularly packed fibres is considered
respect in Refs. [11–15]. The influence of fibre packing first. However, some incorrect ‘unit cells’ (not complying
geometry on the transverse creep behaviour of a uni- with the periodicity) are introduced deliberately in order to
directional composite with regularly spaced missing fibres appreciate the cause for error and magnitude of possible
from hexagonal packing was studied in [16]. Given the errors resulting from an incorrectly prescribed boundary
regularity of the irregularities introduced, unit cells can condition by comparing with exact (to the accuracy of the
still be introduced precisely according to the symmetries numerical solver, i.e. FE) solutions which can be obtained
present, although the unit cell may involve more than one after analyses of appropriately defined unit cells. Then,
fibre. A unit cell in the context of particulate-reinforced some more ‘unit cells’ will be introduced consisting of an
composites was proposed [17], which allowed a large array of periodic cells but with an irregularly truncated,
degree of randomness in the distribution of the reinforce- incomplete part of the periodic cell placed on the top of the
ments within the unit cell but periodicity was imposed in array to form an ‘unit cell’. Uniformly prescribed displace-
order to obtain boundary conditions precisely for the unit ment or traction along these borders in absence of symmetry
cell from translational symmetry conditions. The same can will not be correct any more while the forces involved
be done in a 2D case pertinent to UD composites. remain statically equivalent to the correct ones. By
However, the periodicity introduced in both accounts comparing with correct results, conclusions can be reached
[16,17] undermines the randomness to different extents. A with regard to the possible errors due to the incorrect
representative volume element (RVE) containing randomly boundary conditions and the decay length of the effects of
distributed fibres represented by square cylinders was incorrectly prescribed boundary conditions.
analysed in [18] and compared with the results from its With the outcome for the decay length from the above
counterpart with regular packing and, later, a more realistic described analysis as a guideline, micromechanical FE
case [19] was considered. In such an analysis, precise analysis can then be performed on a UD composite with
prescription of boundary conditions is no longer possible, random fibre distribution in its transverse cross-section using
since there is in general no symmetry about the border of a ‘unit cell’ of a domain, which should contain sufficiently
the RVE and therefore neither displacements nor tractions large number of fibres so that the uniformity in a statistical
are distributed uniformly or with any other regularity along sense can be assumed reasonably. The ‘unit cell’ is loaded
the boundary of the RVE. The prescription of a uniform under uniformly prescribed displacement or traction along its
displacement or traction as in [18,19] represents only an boundary, despite that the deformation and stress field
approximation of some kind. The approximate nature, in obtained in the neighbourhood of boundary are known to be
particular, in the stress distribution, has not been appro- incorrect. Use the decay length obtained in the manner
priately acknowledged in the literature, let alone any described above, a sub-domain can be defined in the middle of
assessment on accuracy of the approximation. Most the ‘unit cell’ by a decay length away from the boundary.
attention has been paid to the bounds of effective Based on the St Venant’s principle, it is expected that the
properties, e.g. [20,21]. It is obvious that there is a lack deformation and stress field in this sub-domain give a close
of consistent and comprehensive approach for addressing approximation to the exact solution, while those from outside
the problem. The objective of this paper is to establish a of the sub-domain will be disregarded completely. The
methodology that provides a reliable means of behaviour of the composite can be obtained from that of
1248 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
the sub-domain correctly. The present approach will guaran- elements in ABAQUS are sufficient to deal with longitudi-
tee a correct stress analysis to be made to UD composites even nal/transverse tension and transverse shear deformation
the fibres are distributed at random over the transverse cross- described by v, w and u0. The active extra degrees of freedom
section and therefore highly relevant to the move towards a in this problem are 30x ; 30y ; 30z and g0yz . The longitudinal shear
better understanding of the behaviour of such materials. deformation described by out-of-plane displacement, u, is
In order to illustrate the implementation of such a uncoupled from v, w and u0 and, hence, can be treated
procedure, analyses will have to be made on UD composites separately. However, there is not an appropriate type of
with fibres distributed at random over the transverse cross- element for this anticlastic problem directly. An analogy to
section. A scheme which generates randomly distributed the steady state heat transfer problem must be used as both
circles representing fibres at a given fibre volume fraction problems are governed by the same Laplace equation.
was devised, so that different cases could be generated for Consequently, the heat transfer elements in ABAQUS can
analysis. ABAQUS CAE [22] was employed for generating be used with the same meshes generated for the analysis in
finite element meshes, performing the analyses and post- the ABAQUS generalized plane-strain problem. The shear
processing the results. modulus is inputted in lieu of heat conductivity and
concentrated force in lieu of concentrated heat flux. In the
output, the heat flux components should be interpreted as
2. A generalized plane-strain problem negative longitudinal shear stresses and temperature as
displacement. The active extra degrees of freedom for this
For the presentation of the problem, the coordinate analysis are g0xz and g0xy . Full details can be found in [9].
system, x, y, and z, for a unidirectionally fibre-reinforced
composite is introduced so that the x-axis lies in the fibre
direction and the y and z-axes lie in the transverse cross- 3. Effects of incorrectly prescribed boundary conditions
section. A UD composite is homogeneous in the x-direction
and, therefore, it is rational to assume that the material Unit cells from regular fibre packing systems, viz.,
undergoes a uniform deformation in the x-direction square and hexagonal, as shown in Fig. 1, can be analysed
microscopically when subjected to macroscopic loads. precisely. The prescription of boundary conditions and
Owing to the homogeneity and the uniformity of the results obtained from such analyses can be found in Li [8,9].
material and its deformation in the x-direction, a two- In order to demonstrate the effects of incorrectly prescribed
dimensional presentation, i.e. generalized plane-strain boundary conditions on the analysis made, an incomplete
problem in the y–z plane [4], can be employed to describe cell truncated at each of the three different locations is
such an otherwise three-dimensional problem. In this attached to a complete cell to form a ‘unit cell’ to be
generalised plane-strain problem, all strain components analysed, as shown in Fig. 2. These truncations bear some
may be present but they are all functions of y and z only, i.e. similarity to those as will be encountered inevitably when an
RVE from a UD composite with fibres distributed at random
3 Z 3ðy; zÞ (1) over the cross-section is dealt with.
where 3Z{3x, 3y, 3z, gyz, gxz, gxy}. In the ‘unit cells’ involving an incomplete part as shown
The displacement field for the interest of present study in Fig. 2, material symmetries are maintained about three
can be given as out of four borders, viz., the bottom, left and right, but there
is no symmetry about the top border. Prescriptions of a
Uðx; y; zÞ Z uðy; zÞ C x30x ; uniformly distributed displacement or traction along the top
border are doomed to be erroneous. However, it is the
Vðx; y; zÞ Z vðy; zÞ; (2) interest of the present study as a first step to show how
Wðx; y; zÞ Z wðy; zÞ wrong the solutions obtained could be. This will then lead to
a systematic approach to be presented later.
where 30x is the macroscopic longitudinal direct strain and u, For the purpose of illustration, only the loading
v and w are the displacements in x, y and z-directions, conditions of uniaxial tension in the x and z directions will
respectively, in this generalised plane-strain problem. be made, respectively. The boundary conditions for the ‘unit
Finite element analyses of this generalized plane-strain cells’ to be analysed are listed in Table 1, where l, b and h
problem can be carried out using a commercial code, are the dimensions of the cell in the x, y and z directions,
ABAQUS [22] with unit cells established as two-dimen- respectively. 30x ; 30y ; 30z are macroscopic strains, referred to as
sional entities lying in the y–z plane, i.e. perpendicular to the extra degrees of freedom as introduced in [9] for the
fibres. The loading cases include longitudinal and transverse convenience of applying boundary conditions in terms of
tension and/or shear in this study. Loads are applied to extra macroscopic strains. Alternatively, concentrated forces Fx,
degrees of freedom f30x ; 30y ; 30z ; g0yz ; g0xz ; g0xy g, corresponding to Fy and Fz can be prescribed to these extra degrees of
the macroscopic strains in the composite, introduced through freedom corresponding to macroscopic stresses. Sz is the
the boundary conditions. The generalized plane-strain traction component in the z-direction on the top border.
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1249
Fig. 1. Appropriate unit cells (a) square and (b) hexagonal fibre packing.
Assume the fibre and the matrix are both isotropic, Among the cases analysed, Case-0 represents a correct
homogeneous and linearly elastic with mechanical proper- prediction for the given packing systems when it is under
ties as follows. For the matrix: EmZ1 GPa, vmZ0.3 and for uniformly prescribed displacement w, while Cases-1–3
the fibre: EfZ10 GPa, vfZ0.2. The fibre volume fraction, contain an error due to the incorrectly prescribed boundary
Vf, of both systems is 60%. The radius of the fibre can be conditions on the top borders of the ‘unit cells’. As a
expressed in terms of the fibre volume fraction as qualitative illustration of the error, the von Mises stress
rffiffiffiffiffiffiffiffiffiffiffiffiffi contour plots for the part corresponding to complete
rffiffiffiffiffi pffiffiffiffiffiffiffiffi
Vf 2 3Vf periodic cell of each ‘unit cell’ for square and hexagonal
r Z 2b and b packing, respectively, were generated and are shown in
p p
Figs. 3 and 4. The two columns of contour plots correspond
for square and hexagonal packing, respectively. At the given to two types of prescribed boundary conditions on the top
fibre volume fraction, rZ0.87404b for the square unit cell border, viz., uniform normal displacement and traction.
and rZ0.81338b for the hexagonal unit cell. In subsequent Compared with the correct solution as shown for Case-0
analysis, both b and l will be assumed to be unity. under prescribed displacement, i.e. Column (b) of Case-0,
Fig. 2. ‘Unit cells’ involving a truncated part (a) square and (b) hexagonal fibre packing.
1250 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
Table 1
Displacement boundary conditions and loading
Boundary Conditions
Displacement yZ0 vZ0
zZ0 wZ0
yZb vZ b30y
Loading Point O u0 Z l30x or concentrated Fx
Top border wZ h30z or traction SzZFz/bl
Fig. 4. von Mises stress contours in the regular part of the ‘unit cells’ from
hexagonal packing (a) the domain analysed, (b) under prescribed uniform
displacement and (c) under prescribed uniform traction.
Table 2
Longitudinal and transverse Young’s moduli
effective elastic properties are not extremely sensitive to the the error amounts to as high as around 20% in some
exact stress distribution in the unit cell, since they represent cases. Interestingly, the improving trend from Case-1 to 3 as
some overall measures, often in a kind of average sense. compared with Case-0 is observed once again to a certain
Nevertheless, the differences are still appreciable and extent in the computed effective Young’s moduli as
Fig. 5. Arrays of periodical cell (a) square packing, (b) hexagonal packing.
1252 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
shown in Table 2, with the only exception of Ex0 from the 4. Decay of the effects of incorrectly prescribed
hexagonal packing case, which is due to the significantly boundary conditions
reduction of fibre volume fraction in the ‘unit cells’
analysed as Ex0 is dominated primarily by fibre volume Either the prescribed uniform displacement or traction
fraction. The results presented in Table 2 have all been results in a force system imposed to the ‘unit cells’, which is
produced under the boundary condition of prescribed expected to be statically equivalent to that in exact solution
uniform displacement. By applying the St Venant’s principle, one would expect the
The analysis made and results obtained in this section effects due to incorrectly prescribed boundary conditions
suggest that incorrectly prescribed boundary conditions decay away from the boundary where incorrect boundary
could have a significant effect on the accuracy of the conditions are prescribed. To demonstrate this decaying
outcome. Using the results without appropriate treatment trend, periodical cells have been stacked together. On the
could be misleading. This applies equally to the analyses top, an incomplete chunk is added, as in the cases in Section
made on RVEs for UD composites with fibres distributed at 3, to form ‘unit cells’ to be analysed in this section. They
random over the transverse cross-section, such as that in have been schematically shown in Fig. 5 for square and
[18]. However, the magnitude of the error can be minimised, hexagonal fibre packing systems, respectively. Through
if not effectively eliminated, after an appropriate treatment such analyses, a clear trend with regard to the decaying
is introduced, which will be developed in the sections to characteristics can be obtained, which would serve as a
follow. guideline for subsequent development of the approach to
Fig. 6. von Mises stress contours for each complete periodic cell from the square packed arrays (a) under prescribed uniform displacement, (b) under prescribed
uniform traction.
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1253
analysis of UD composites with fibres distributed at random displacement on the top while those in Figs. 6(b) and 7(b)
over its transverse cross-section. were produced under uniformly prescribed traction on the
By applying the loads and boundary conditions in a top. The pattern of contour plots in the column of Case-0 in
similar way to that for the cases in Section 3, the ‘unit cells’ Fig. 6(a) is identical for every cell as Case-0 represents a
constructed here can be analysed. Same as before, Case-0 in correct solution to the problem. Comparing the patterns in
both packing systems produce correct results under Case-1–3 with that in Case-0, it can be observed that as the
prescribed uniform displacement in consistence with those cells locate further away from the top border where
in Section 3. incorrect boundary conditions are prescribed, i.e. from No.
The von Mises stress contour plots for every periodical 1 to 5, the patterns of contours get closer to that in Case-0. In
cell involved in Case-1–3 in those arrays are shown in fact, no noticeable difference can be spotted from cell No. 2
Figs. 6 and 7 for the two packing systems, respectively. The onwards. Relatively, it takes longer distance for the
extra incomplete part added on the top in each case is not boundary effects to decay, if a uniform traction is prescribed
included in these contour plots as it represents a zone on the top as is obvious in Fig. 6(b), by comparing with
severely affected by the boundary effects and, therefore, the Case-0 in Fig. 6(a), although disturbances settle down
deformation and stress distribution in such an asymmetri- eventually by No. 4 or so. The same trends repeated in Fig. 7
cally truncated part are doomed to be incorrect and do not for hexagonal packing where arrangements of the plots have
represent the actual response of material. All complete cells been made in exactly the same order as that in Fig. 6.
as shown in Figs. 6 and 7 are labelled from No. 1 to 5 for The results shown in Section 3 suggest that the boundary
the cells from top to bottom. The contour plots in Figs. 6(a) condition of uniformly prescribed displacement represents a
and 7(a) were generated under uniformly prescribed better approximation than that of uniformly prescribed
Fig. 7. von Mises stress contours for each complete periodic cell from the hexagonally packed arrays (a) under prescribed uniform displacement, (b) under
prescribed uniform traction.
1254 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
Fig. 9. Traction along L1–L5 for different cases of square array unit cell.
1.2 Square packing
1
L1 of Case-3 the applied macroscopic stress. The trend as observed from
L1 of Case-2 the displacements as discussed above applies unanimously.
0.8 L1 of Case-1 The only tiny inconsistence is in the traction along L1 in
Case-0. As L1 was a part of the traction boundary and
0.6
tractions at the boundary can only be evaluated as accurate
0.4 as the energy minimisation, i.e. FE approximation, allows, a
L1-L5 for Case-0
L1 for Case-1-3
tiny non-vanishing distribution for L1 of Case-0 is observes
0.2 L2-L5 for Case 1-3 in Fig. 9(b). However, this is insignificant and will diminish
as the mesh refines.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 The distributions of the same displacements and tractions
y
in the case of hexagonal packing are shown in Figs. 10 and 11.
(b) 0.025
There is a slightly different arrangement in the presentation
Normalised normal displacement, w
Square packing of the data. The results along borders L2 and L4 have been
0.02
reflexed between left and right for the ease of comparison
L1-L5 for Case-0 with those along L1, L3 and L5. The decaying trend is similar
0.015 L1 for Case-1-3 L1 of Case-1
L2-L5 for Case 1-3
to that for the square packing, in general, although the two
L1 of Case-2
packing systems are distinctly different. The sharp bends in
0.01
Fig. 10 and the discontinuities in Fig. 11(a) coincide with the
L1 of Case-3
interface between the fibre and matrix.
0.005
In all the cases as shown in Figs. 8–11, it is consistent
that differences between the results obtained with incorrect
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 boundary conditions (Cases-1–3) and those with correct
L1 - L5 of Case-0 boundary conditions (Case-0) are significant only up to
–0.005
y
border L1. In other words, the effects due to incorrectly
prescribed boundary conditions diminish over the length of
Fig. 8. Displacement along L1–L5 for different cases of square array unit a complete periodic cell, which is directly related to
cell. the centre-to-centre spacing between fibres. By the distance
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1255
Fig. 10. Displacement along L1–L5 for different cases of hexagonal array Fig. 11. Traction along L1–L5 for different cases of hexagonal array unit
unit cell. cell.
ðb
1 and effective properties are predicted based on the
3ave
z Z wL dy (4) deformation within this sub-domain, reliable results can
bh 0
be obtained, despite the fact that the analysis has been made
where superscript L indicates the border under consideration under incorrect boundary conditions. This approach will be
which could be any of L1–L5 in the ‘unit cells’ analysed. b is developed and illustrated in this section. It will also be
the width of array unit cell which is equal to 1 in this study and shown that noticeable errors could find their way to the
h is the distance from the bottom to border L. The effective results without such a sub-domain treatment. Such a sub-
transverse Young’s modulus in the z-direction associated domain approach has been employed in [15]. However,
with border L can be obtained from neither procedure nor justification had been provided there.
save
z
Ez0 Z : (5)
3ave
z 5.1. Generation of random fibre distribution
In Table 3, Case-0 gives correct answer from each
concerned. For other cases, results show their convergence In order to facilitate analyses on UD composites having
from L2 onwards. It can be noticed that the converged values random fibre distribution over its transverse cross-section, a
tend to leave a tiny margin to the correct ones, amounting to means of generating such a random distribution is desirable.
an error below 0.5%. These are coincident with the The process is equivalent to generating randomly spaced,
discrepancies as observed in the distributions of displace- non-overlapping circular discs in a plane within a domain,
ments and traction along those borders as mentioned earlier given the generalised plane-strain idealisation. There could
in relation to Figs. 8–11. It is expected that the margin would be different approaches to achieve this goal. Digitising
narrow, as the number of complete periodic cells are included microscopic photographs such as that given in Fig. 12 is an
in the array of the ‘unit cells’. obvious one but the procedure could be costly and time
From the analysis made and results obtained above, it can consuming. A theoretical way would be beneficial. One
be concluded that the effects of incorrectly prescribed possibility is to adopt a ‘coin-dropping’ or random insertion
displacement boundary conditions take a distance of a few process, avoiding any overlap by removing overlapping
times of fibre spacing to decay. Beyond the decay length, the ones. A process of this type was performed in [13]. While
deformation and stress and strain fields in the composites the random nature of the obtained distribution is obvious, it
give a good approximation of the correct material responses, may be difficult to achieve a specified disc (fibre) volume
provided that the RVE contains a sufficient number of fibres. fraction, especially higher volume fractions. Fibre volume
fraction is often a dominant parameter in determining the
performance of the composite and a reasonable control of it
is essential. A different process was employed in [14]. An
5. Analysis of UD composites with fibres distributed
alternative scheme is proposed as follows, which is capable
at random over the transverse cross-section
of achieving any possible volume fraction and yet main-
Regularly packed fibre-matrix systems, such as square taining the random nature in distribution.
and hexagonal, are mostly simplistic idealisations. Majority Starting from a set of regularly, e.g. hexagonally, packed
of UD composites have fibres distributed at random over the discs of radius R in the y–z plane as shown in Fig. 13. The
transverse cross-section. When carrying out a micromecha- spacing between discs 2b can pffiffiffi be calculated from volume
nical analysis on any of such materials using a regularly fraction Vf as b2 Z pR2 =2 3Vf . Introduce two frames,
shaped RVE, e.g. a square, containing sufficient number of Frame 1 and 2, the latter being a sub-domain of the former.
fibres to maintain the characteristics of the random The distance from the border of Frame 2 to that of Frame 1
distribution [23], the boundary of the RVE will not coincide is a few times of b, according to the conclusion reached in
with any symmetry. Along the boundary of such an RVE,
accurate prescription of boundary conditions is impossible
in general. Prescribing uniform displacements along the
boundary could be an approximation but the obtained
deformation in the neighbourhood of boundary could be
subject to a significant distortion, as demonstrated in
Sections 3 and 4 of this paper earlier. However, making
use of the outcome obtained in Section 4, it can be expected
that sufficiently accurate prediction of the deformation can
be obtained from an inner sub-domain of the RVE, provided
that inner sub-domain is sufficiently away from the external
boundary, by a couple of times of a typical fibre spacing
according to the outcome of Section 4 on the study of
decaying characteristics. If the behaviour of the composite Fig. 12. Random fibre distribution observed from the real composites.
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1257
Z Z
Y Y
(b)
(a)
Z Z
Y Y
(c) (d)
Fig. 15. Random fibre distribution obtained after. (a) 265 iterations, Vf1 Z 65:69%, and Vf2 Z 64:17% in Frame 1 and 2 with original VfZ65%, (b) 286 iterations,
Vf1 Z 65:84% and Vf2 Z 64:26% in Frame 1 and 2 with original VfZ65%, (c) 382 iterations, Vf1 Z 66:38% and Vf2 Z 64:59% in Frame 1 and 2 with original VfZ
65%, (d) 262 iterations, Vf1 Z 61:00% and Vf2 Z 59:40% in Frame 1 and 2 with original VfZ60%.
remain straight but allowing a pattern of deformation like a EfZ10 GPa and nfZ0.2, with subscripts f and m indicating
four-bar mechanism to fulfil a macroscopic pure shear state. the phase of the constituent, for all the cases with 65% fibre
This was implemented in ABAQUS through a simple user volume fraction. For the case with 60% fibre volume
subroutine MPC (multiple point constraints). Through this fraction, material properties of both fibre and matrix are
mechanism, an extra degree of freedom g0yz can be listed in Table 5 [24].
introduced. In order to apply a microscopic shear stress of The analyses were made on the domain Frame 1. The
s0yz , a concentrate force of As0yz can be prescribed to this results from the sub-domain Frame 2 were extracted and
extra degree of freedom. considered to represent the correct responses of the material
The fibre and matrix are both assumed to be isotropic and free from the effects of incorrectly prescribe boundary
homogeneous with elastic properties, EmZ1 GPa, nmZ0.3, conditions along the border of Frame 1.
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1259
Fig. 16. Meshes for unit cells with fibres distributed at random (a) VfZ65%, (b) VfZ60%.
5.3. Results and discussion stress state in the y-direction macroscopically is applied to
the RVE, Frame 1. The results were extracted from Frame 2,
Analyses have been carried out for all the four cases as where the effects of incorrectly prescribed boundary
shown in Fig. 15. The data extraction process as described conditions are expected to have diminished.
above has been applied to each of them. The results are The von Mises stress contour plot for one of the three
presented and discussed below. cases at 65% fibre volume fraction is shown in Fig. 17 where
only the sub-domain, Frame 2 as defined in Fig. 15(a) is
5.3.1. Deformations and stress distributions displayed while the analysis was performed on the whole
To show the pattern of deformation and stress distri- RVE, i.e. Frame 1. The contour plots were shown in a
bution in the unit cell, a load which results in 1 MPa uniaxial deformed configuration with the magnitude of deformation
Fig. 17. Contours of von Mises stress in Frame 2 as a part of RVE for Case-1 under macroscopically uniaxial transverse tension.
1260 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
amplified by a constant factor of 300. The most highly sub-domain Frame 3 as indicated in Fig. 15(a). To
stressed areas are always found at interfaces, especially in demonstrate this, the von Mises stress contour plots for
areas where fibres are close to each other in the direction of this same sub-domain, Frame 3, but from different analyses,
loading. The distortion of the shape of fibre cross-section is one on Frame 1 and the other on Frame 2, are shown in Fig.
small relative to overall deformation because of the higher 19. Both plots share many common features, although the
stiffness than that of the matrix. The deformation is mostly distance between the borders of Frame 2 and 3 is not quite to
taken by the matrix although stresses in the fibres tend to a decay length yet (at least, not as much as that between
show higher levels than those in the matrix. This explains Frame 1 and 2). The consistence in their trends reinforces
why observed stress concentrations are so located. As a the justification made to the proposed methodology in this
result of deformation, the edges of Frame 2 do not remain paper.
straight. This provides an indication how wrong it would be
if Frame 2 was analysed alone and straight edges had been 5.3.2. Effective properties
assumed in prescribing displacement boundary conditions. As explained previously, effective properties describe the
In fact, a case is shown in Fig. 18 with the same deformation overall behaviour and are not extremely sensitive to the
magnification factor of 300, which was from the analysis precise distributions of stresses and strains resulting from
made directly on Frame 2 with uniform displacements the macroscopic load applied. However, as an important
prescribed along the edges of Frame 2. Compared with the aspect of micromechanical analysis, some results obtained
more accurate analysis made on Frame 1 from which Frame from the analyses made will be presented here.
2 was extracted, as shown in Fig. 17, significant difference The evaluation of the longitudinal Young’s modulus,
can be found. The maximum von Mises stress in the Ex0 , will not performed, since this would require
incorrect analysis is 9.879 MPa in the fibres and 4.501 MPa computation of average of stress sx over the domain of
in the matrix while in the more accurate analysis they are Frame 2, which is time consuming. A relative simpler
14.371 and 10.997 MPa, respectively. Using the von Mises way to work out Ex0 is from the strain energy, which is
criterion with a given yield stress for the matrix, the more straightforward. However, this only produce an up
incorrect analysis would overestimate the load for the onset bound which is not necessarily better than other bounds.
of plastic deformation in the matrix, for instance, by about Given the fact that this particular effective property can be
2.5 times in this particular example. evaluated with excellent accuracy using even a crude
Given the erroneous analysis as shown in Fig. 18, means, e.g. the rule of mixtures, it is not worthwhile to
reasonable accuracy can still be conceived in the pursue it further here.
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1261
Fig. 19. Frame 3 extracted from (a) Fig. 17 and (b) Fig. 18, both with a deformation magnification factor of 300.
In order to obtain effective transverse Young’s moduli, where y1 and yr are the y-coordinates of the left and right
for instance, that in the z-direction, the average tractions on edges and zb and zt are the z-coordinates of the bottom
the top and bottom borders of Frame 2 are found to produce and top edges of Frame 2, respectively. The effective
an average stress save ave
z and the average strain 3z is obtained Young’s modulus can be obtained from (5). The similar
from displacement w along these borders as follows procedure applies to other effective elastic properties
ð y ð yr when the RVE is under appropriate macroscopic uniaxial
ave 1 r
top bot
sz Z s dy C sz dy (6) stress state.
2ðyr K yl Þ yl z yl The results from all three cases at 65% fibre volume
ð y ð yr fraction (Case-1–3 as shown in Fig. 15) have been obtained
1 r
3ave
z Z top
w dy C w bot
dy : (7) and listed in Table 4 for comparison with their counterparts
ðyr K yl Þðzt K zb Þ yl yl from regular hexagonal packing. The differences between
1262 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
Table 4
Effective properties of UD composites with regular and random fibre packing (65% fibre volume fraction)
Table 6
Comparison of effective properties of UD composites obtained from various methods with those from the case of randomly packed fibres (60% fibre volume
fraction)
Effective properties Experimental [24] Hashin and Rosen [1] Li and Zou [25,26] McCartney [27] Present
Ex (GPa) 45.6 45.85 45.76 45.76 N/A
Ey (GPa) 16.2 9.740 (lower bound); 11.80 12.034 14.544
13.106 (upper bound)
Gxy (GPa) 5.83 4.318 4.339 4.3179 4.7186
nxy 0.278 0.2517 0.2515 0.2517 0.2447
nyz 0.4 0.3346 (lower bound); 0.4020 0.3889 0.3732
0.5055 (upper bound)
Gyz (GPa) 5.786a 3.235 (lower bound); 4.208 4.332a 4.863
4.910 (upper bound)
a
Values were worked out assuming transverse isotropy.
significantly higher than any other theoretically predicted and almost all theoretical predictions based on regular fibre
ones, closer to the experimental results. It should be noted packing.
that the upper or lower bounds of Hashin and Rosen [1] are
the bound only for the fibre-matrix system which can be 5.3.3. Behaviour in plastic regime
represented by concentric cylinders (though called The analysis of the case of 60% fibre volume fraction
‘random’ there) as assumed in their work and they are was run beyond the limit of linear elasticity by assuming a
not necessary to be the bounds for different systems, e.g. perfectly plastic behaviour for the matrix beyond elastic
that with fibres distributed at random as presented in this limit, while the fibres were assumed to remain elastic. The
paper. Results in [27] were based on a concentric cylinder yield stress of the matrix was assumed to be syZ80 MPa
model, which is similar to that of [1] and not a precise and the von Mises yield criterion was employed. The von
representation of the real physical problem. In the results Mises stress contour plots is shown in Fig. 20 corresponding
obtained here as shown in Table 6, improvements are to the linear elastic limit or the onset of plastic deformation,
found for all moduli, the transverse Young’s modulus in which usually takes place at a very low load level due to the
particular, except some slight drift away from the high stress concentrations in regions where fibres become
experimental data in Poisson’s ratios. This is believed to almost contiguous. Two equivalent plastic strain contour
bring forward some explanations to the discrepancies in the plots are presented in Fig. 21 at macroscopic strain levels of
transverse Young’s modulus between experimental data 0.7 and 1.5%, respectively. Fully plastic deformation
Fig. 20. Distribution of von Mises stress in Frame 2 at the onset of plastic deformation.
1264 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
Fig. 21. (a) Equivalent plastic strain (von Mises) for Frame 2 at 0.7% overall strain applied. (b) Equivalent plastic strain (von Mises) for Frame 2 at 1.5% overall
strain applied.
establishes as the slip-lines through the matrix become more is irregular as shown in Fig. 21(a), the established slip-lines
or less interconnected. Plastic deformation localises along possesses a fairly regular pattern in Fig. 21(b).
the slip-line. Although the fibres are distributed entirely at The macroscopic stress–strain curve can also be obtained
random and indeed the plastic deformation at its early stage through such a micromechanical analysis, as shown in solid
A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266 1265
80 Hexagonal unit cell loaded in z-direction The analysis made here was based on a small
Hexagonal unit cell loaded in y-direction deformation assumption for the sake of simplicity, which
60 is probably just sufficient for the levels of deformation
40
involved. At higher levels of the deformation, appropriate
consideration of the effects of geometric change of the mesh
20 Correct results: Frame 2 extracted from Frame 1 needs to be made since strain levels in local zones, where
slip-lines are developed can be significantly high and the
0 mesh can be greatly distorted there. To perform the analysis
0 1 2 3
Strain (%) satisfactorily, one might have to use some kind of re-
meshing facility in these local zones. This is obviously
Fig. 22. Macroscopic stress–strain curve. beyond the scope of the present paper.
The analyses in this section are only presented as
line in Fig. 22. The data for this curve were extracted from examples of application. Through them the implementation
Frame 1 for simplicity, i.e. they contain errors due to of the methodology as proposed in this paper for analyses of
incorrectly prescribed boundary conditions. A stress–strain UD composites with fibres distributed at random over the
curve could be obtained after similar treatment as performed transverse cross-section has been demonstrated, in which
previously by extracting information from Frame 2. the effects of incorrectly prescribed boundary conditions
However, it was found very time consuming, given the have been minimised if not eliminated. The results also
great number of increments required for this nonlinear show some interesting perspectives of UD composites
analysis. Only two stages corresponding to the states of purely due to the random distribution of fibres over the
deformation at the onset and full establishment of transverse cross-section.
plasticity have been processed in this way. The stresses
and strains after the treatment are presented in Fig. 22 in
blobs. In this particular aspect, the difference between data 6. Conclusions
obtained from Frame 1 and those extracted from Frame 2
does not seem to be significant. In micromechanical finite element analyses of UD
Also included in Fig. 22 are two stress–strain curves for composites of fibres distributed at random over the
the same but based on regular, hexagonal packing, shown in transverse cross-section using ‘unit cells’, boundary con-
dashed lines. They correspond to loading in the y and z ditions for such ‘unit cells’ can no longer be prescribed
directions, respectively. The two curves are identical mostly precisely due to the lack of symmetry about the boundary.
owing to the transverse isotropy the geometry possesses. Through the analyses as presented in this paper, the effects
The bifurcation at early stages of plastic regime is due to the of incorrectly prescribed boundary conditions for such
loss of similarity between the two cases and hence loss of RVEs on the predicted behaviour of the composite
the isotropy at the onset plastic deformation because of the represented by the ‘unit cells’ have been studied. The
emergence of isolated plastic zones [28]. This is soon decaying characteristics of such effects away from the
superseded by consistent responses as the isotropy resumes boundary have also been examined. A decay length has been
when the plasticity is fully developed. found to amount to a couple of times of the centre-to-centre
From the analysis, it can be seen that, while the onset of fibre spacing. Using the decay length obtained, analyses
plastic deformation takes place as early as around 7 MPa of have been made to UD composites in which fibres are really
the applied macroscopic stress, the state of fully plastic yield distributed at random over the transverse cross-section.
of the composite is not established until around 110 MPa of Although, the prescribed uniform normal displacement
the applied macroscopic stress as can be seen from the along the boundary represents an incorrect boundary
stress–strain curve as shown in Fig. 22. The latter condition, correct deformation and stress distribution can
corresponds to a situation of full establishment of slip- be obtained from a sub-domain of the unit cell away from
lines in the matrix. the boundary by a number of times of average fibre spacing.
It is interesting to notice that the composite with This provides a practical methodology for micromechanical
regularly packed fibres tends to reproduce the elastic, finite element analysis of such composites. The method-
perfectly plastic behaviour as assumed for the matrix, while ology can certainly be generalised to 3D problems, pertinent
noticeable strain hardening can be observed for the to the micromechanical analysis of particulate-reinforced
composite with fibres distributed at random over composites. Results from the examples included show some
the cross-section. The random distribution of the fibres interesting aspects of UD composites of fibres distributed
1266 A. Wongsto, S. Li / Composites: Part A 36 (2005) 1246–1266
at random over the cross-section. In particular, the predicted [12] Zohdi T, Feucht M, Gross D, Wriggers P. A description of
transverse Young’s and shear moduli from UD composites macroscopic damage through microstructural relaxation. Int
J Numer Methods Eng 1998;43:493–506.
with fibres distributed at random over the transverse cross-
[13] Bulsara VN, Talreja R, Qu J. Damage initiation under transverse
section tend to give noticeably higher values than those loading of unidirectional composites with arbitrarily distributed fibers.
obtained with regular packed fibres and, therefore, they Compos Sci Technol 1999;59:673–82.
would help to narrow the gap between experimental data [14] Gusev AA, Hine PJ, Ward IM. Fiber packing and elastic properties of
and existing theoretical predictions. The random distri- a transversely random unidirectional glass/epoxy composite. Compos
Sci Technol 2000;60:535–41.
bution of fibres also tends to enhance the strain hardening of
[15] Sautter M, Dietrich Ch, Poech MH, Schmauder S, Fischmeister HF.
the material significantly in the plastic regime when it is Finite element modelling of a transverse-loaded fibre composite.
subjected loads transverse to the fibres, which cannot be Effects of section size and net density. Comput Mater Sci 1993;1:
predicted with a regularly packed fibre arrangement. 225–33.
[16] Zhang L, Ernst LJ, Brouwer HR. Transverse behaviour of a
unidirectional composite (glass fibre reinforced unsaturated polye-
ster). Part I. Influence of fibre packing geometry. Mech Mater 1998;
Acknowledgements
27:13–36.
[17] Segura do J, Llorca J. A numerical approximation to the elastic
The authors would like to express their gratitude to an properties of sphere-reinforced composites. J Mech Phys Solids 2002;
anonymous referee of Journal of Computational Materials 50:2107–21.
for his/her comments on the manuscript submitted to that [18] Brockenbrough JR, Suresh S. Plastic deformation of continuous fiber-
reinforced metal-matrix composites: effects of fiber shape and
journal as an earlier attempt of publication of the work as
distribution. Scripta Metall Mater 1990;24:325–30.
presented in this paper. [19] Brockenbrough JR, Suresh S, Wienecke HA. Deformation of metal-
matrix composites with continuous fibres: geometrical effects of fibre
distribution and shape. Acta Metall Mater 1991;39:735–52.
References [20] Hazanov S, Huet C. Order relationships for boundary conditions effect
in heterogeneous bodies smaller than the representative volume.
[1] Hashin Z, Rosen BW. The elastic moduli of fiber-reinforced materials. J Mech Phys Sol 1994;41:1995–2011.
ASME J Appl Mech 1964;31:223–32. [21] Jiang M, Ostoja-Starzewski M, Jasiuk I. Scale-dependent bounds on
[2] Adams DF, Doner DR. Longitudinal shear loading of a unidirectional effective elastoplastic response of random composites. J Mech Phys
composite. J Compos Mater 1967;1:4–17. Sol 2001;49:655–73.
[3] Adams DF, Doner DR. Transverse normal loading of a unidirectional [22] HKS. ABAQUS user’s manual 2002; Version 6.2.
composite. J Compos Mater 1967;1:152–64. [23] Drugant WJ, Willis JR. Micromechanics-based nonlocal constitutive
[4] Adams DF, Crane DA. Finite element micromechanical analysis of a equation and estimates of representative volume element size for
unidirectional composite including longitudinal shear loading. elastic composites. J Mech Phys Sol 1996;44:497–524.
Comput Struct 1984;18:1153–65. [24] Soden PD, Hinton MJ, Kaddour AS. Lamina properties, lay-up
[5] Chen CH, Cheng S. Mechanical properties of fiber reinforced configurations and loading conditions for a range of fibre-
composites. J Compos Mater 1967;1:30–41. reinforced composites laminates. Compos Sci Technol 1998;58:
[6] Nedele MR, Wisnom MR. Finite element micromechanical modeling 1011–22.
of a unidirectional composite subjected to axial shear loading. [25] Li S, Zou Z. Unit cells and micromechanical finite element analysis of
Composites 1994;25:263–72. unidirectionally fibre-reinforced composites. ECCM9, Composites:
[7] Li DS, Wisnom MR. Finite element micromechanical modeling of from fundamentals to exploitation. In: Eur Conf Compos Mater. 9, 4–
unidirectional fibre-reinforced metal–matrix composites. Compos Sci 7 June 2000, Brighton, UK.
Technol 1994;51:545–63. [26] Zou Z, Li S. Backing out fibre properties from effective properties of
[8] Li S. On the unit cell for micromechanical analysis of fibre-reinforced composites using unit cells. FRC 2000, Composites for the
composites. Proc R Soc London, A 1999;455:815–38. Millennium. In: Proceedings of the eighth conference fibre reinforced
[9] Li S. General unit cells for micromechanical analyses of uni- composites.
directional composites. Composites Part A 2000;32:815–26. [27] McCartney LN. Analytical method of calculating the thermo-elastic
[10] Zou Z, Li S. Stresses in an infinite medium with two similar circular constants of a multi-phase unidirectional composite. NPL Report
cylindrical inclusions. Acta Mech 2002;156:93–108. DMM(A); 1992.
[11] Fang D, Liu T. Transverse plastic deformation of metal-matrix with [28] Boehm HJ. Institute of Lightweight Design and Structural Biome-
randomly arranged continuous fibers. Comput Mater Sci 1997;7: chanics, The Vienna University of Technology. Private communi-
343–50. cation; 1999