Materials Science and Engineering A 452–453 (2007) 483–498

A thermo-viscoelastic analysis of process-induced residual stress in

fibre-reinforced polymer–matrix composites
L.G. Zhao ∗ , N.A. Warrior, A.C. Long
School of Mechanical, Materials and Manufacturing Engineering, The University of Nottingham, University Park, Nottingham NG7 2RD, UK
Received 30 May 2006; received in revised form 17 October 2006; accepted 18 October 2006

Abstract
Process-induced residual stress in fibre-reinforced thermoset polymer–matrix composites was analysed using a thermo-viscoelastic microme-
chanical model and the finite element method. A three-dimensional unit cell with glass fibre and epoxy polymer–matrix, representing the periodic
microstructure of unidirectional fibre-reinforced composites, was considered to compute cure residual stress of fibre composites induced by chem-
ical shrinkage of the epoxy resin and thermal cooling contraction of the whole fibre and resin system. The constitutive behaviour of the epoxy
matrix was described by a cure and temperature-dependent viscoelastic material model. Compared to an elasticity solution, a reduction in residual
stress was predicted due to the stress-relaxation caused by the viscoelastic behaviour of the epoxy matrix. Calculated residual stress shows strong
dependency on the fibre volume fraction and fibre packing. After the cure process is complete, residual stress tends to relax to a constant value.
The effect of residual stress on damage and failure of the model was also studied using the maximum stress failure criterion combined with a
post-failure stiffness reduction technique. Damage onset, in terms of the location and the load level, was shown to be clearly influenced by the
residual stress for both normal and shear loading. Initial and final failure envelopes, predicted for biaxial normal (longitudinal and transverse)
loading and combined shear (longitudinal) and normal (transverse) loading, were shown to be shifted and contracted by the inclusion of residual
stress. For final failure, residual stress was seen to have little effect on the load levels for longitudinal failure but greatly affected the load levels
for transverse and shear failure. Residual stress could be detrimental or beneficial depending on the state of existing residual stress and the loading
conditions.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Residual stress; Thermo-viscoelastic; Cure; Relaxation; Finite element analysis; Damage and failure

1. Introduction cally. Experimental methods include the techniques of section-

Process-induced residual stress in polymer–matrix compos-
ites is a direct consequence of the chemical shrinkage of the
matrix during polymerisation and the mismatch of thermal con-
traction between the fibre and the matrix during cooling. The
formation of residual stresses can have significant effects on
the mechanical performance of composite structures by caus-
ing warpage [1] or initiating pre-load damage such as interface
debonding and matrix microcracking [2,3]. Both warpage and
initial damage can reduce the stiffness and the strength of the
material, as well as acting as sites for nucleation of macrocracks
and environmental degradation.
Process-induced residual stresses in polymer–matrix com-
posites can be determined both experimentally and analyti-
∗ Corresponding author. Tel.: +44 23 92842375; fax: +44 23 92842351.
E-mail address: liguo.zhao@port.ac.uk (L.G. Zhao).
ing/cutting [4], warpage/curvature measurements [1] and the
X-ray or neutron diffraction methods [5,6]. For analytical meth-
ods, residual stresses in composites are generally studied at
either macro or meso/micro-levels. At the macro-level, classi-
cal laminate theory is normally used and gives predictions of
residual stress at the ply level [7–9]. At the meso/micro-level, a
Representative Volume Element (RVE) or unit cell which repre-
sents the meso/microscopic periodic structure of the laminate is
considered and the analysis is often carried out using a numeri-
cal procedure such as the finite element method [10–13]. From
the meso/microscopic unit cell model, residual stress field can
be predicted at the fibre–matrix or tow (yarn) level.
For thermoset polymer composites, a typical curing process
consists of two steps: curing at a constant elevated temperature
and thermal cooling from the curing temperature to room
temperature. During curing, the polymer shrinks as a result of
the purely chemical reaction (polymerisation) and its material
characteristics change dramatically through the transition from

0921-5093/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.msea.2006.10.060
484 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
a liquid state to a solid state while the reinforcement remains
unchanged [14,15]. For thermal cooling, both polymer and
reinforcement contract but by different amounts and in addition
the polymer may change its stiffness significantly. Therefore,
the build-up of residual stress over time is governed by the
change of volume and material properties during the complete
cure process.
For fibre reinforcement, it is generally assumed that mate-
rial behaviour is elastic and remains unchanged during cure,
i.e., independent of the degree of cure and the temperature.
However, for polymer–matrix, material constitutive behaviour
is dependent on time, degree of cure and temperature history
[16–18]. The time-dependent viscoelastic behaviour is illus-
trated by the stress-relaxation phenomenon of the material
[19,20]. Viscoelastic analysis requires the measurement of vis-
coelastic material properties during the cure procedure. Several
studies have focused on the measurements of material modulus
as a function of time, temperature and cure [16–18], which can be
used to predict the residual stress through an integral-formulated
constitutive law [21–24]. On the other hand, constitutive rela-
tionship for viscoelastic behaviour can also be modelled by a
differential formulation, which is often used because of its sim-
pler form and easy adaptation to finite element analysis [11,25].
Effects of residual stress on mechanical behaviour of the com-
posite materials have been studied primarily by analytical meth-
ods. Nimmer [26] and Wisnom [27] showed that the presence of
compressive residual stress at the interface of fibre and matrix
is beneficial for the transverse behaviour of a composite with
low interfacial strength. It was also shown that residual stress
can result in changing and movement of the yielding surfaces for
metal–matrix composites [28,29]. For polymer–matrix compos-
ites, process-induced residual stress may introduce contraction
and shifting of biaxial failure envelopes for transverse loading
as shown in Zhao et al. [13]. It was concluded in Zhao et al. [13]
that the maximum principal residual stress in a polymer–matrix
is mainly tensile and thus detrimental to composite structures
by facilitating the damage initiation and evolution. However,
residual stress could also bring beneficial effects in certain cir-
cumstances, such as for a relatively high strength polymer resin
and a transverse tensile loading condition, due to the presence of
compressive residual stress in the transverse loading direction
at preferential damage locations [13].
Although residual stress has been studied extensively for
polymer–matrix composites, the information is still limited,
especially the effects of residual stress on damage and failure
behaviour. In the present work, finite element analysis was used
to study residual stress and its effect on damage and failure of
fibre-reinforced thermoset polymer–matrix composites using a
micromechanical unit cell model. The residual stress introduced
during curing was determined by considering both chemical
shrinkage of resin and thermal cooling contraction of fibre and
resin. A cure and temperature-dependent viscoelastic material
model was adopted to describe the constitutive relationship of
the polymer–matrix. Relaxation behaviour of residual stress and
effects of fibre volume fraction and packing on residual stress
were investigated. Effects of residual stress on damage and fail-
ure of the unit cell subjected to mechanical loading after curing
were predicted using the maximum stress failure criterion and a
post-failure stiffness reduction technique. Normal loading and
longitudinal shear loading situations were all considered. Fur-
thermore, biaxial normal loading (transverse and longitudinal)
and combined shear (longitudinal) and normal (transverse) load-
ing were applied to the model to investigate the effects of residual
stress on initial and final failure envelopes.
2. Material
2.1. Cure kinetics
Thermoset resin undergoes chemical reactions during the cur-
ing process at an elevated temperature. Curing of a resin occurs
over time and the degree of cure depends on the temperature
history. The chemical kinetics of thermoset resin defines the
dependency of the degree of cure on temperature history. In
order to capture the conversion-dependent material properties,
the resin cure kinetics need to be characterised accurately. For
epoxy resins, the autocatalytic model has been used widely to
describe cure kinetics and this is also adopted here. The model
can be expressed as follows [16,30]:
dα
= (K1 + K2 αn1 )(1 − α)n2 (1)
dt
   
−E01 −E02
K1 = K01 exp , K2 = K02 exp (2)
RT RT
where α is the degree of cure, t the time, K1 and K2 the rate
constants, n1 and n2 the exponents, K01 and K02 the temperature-
independent factors, E01 and E02 the activation energies, and R
and T are the universe gas constant and the temperature, respec-
tively. The values of the parameters are taken from Eom et al.
[16] and are listed in Table 1. As shown by Eom et al. [16],
the predicted degree of cure from Eqs. (1) and (2) agrees with
the experimental measurements very well for cure temperatures
between 150 ◦ C and 170 ◦ C. For illustration, Fig. 1 gives the
degree of cure as a function of time at 150 ◦ C, 160 ◦ C and 170 ◦ C
cure temperatures. It can be seen that the reaction advances very
rapidly at the beginning and then approaches a steady state. The
final degree of cure increases with the cure temperature.
2.2. Viscoelastic model
Constitutive formulations for viscoelastic polymeric materi-
als generally fall into two categories: differential and integral
forms. Integral formulations are convenient for stress and strain
Table 1
Values of kinetic parameters for the autocatalytic model [16].
Parameters Values
K01 (s−1 ) 2.7321 × 105
K02 (s−1 ) 3.8231 × 105
E01 (J/mol) 7.2776 × 104
E02 (J/mol) 6.6934 × 104
n1 1.07
n2 2.43
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
Fig. 1. Calculated degree of cure vs. time at different curing temperatures for
epoxy resin.
analyses and the experimental data required for the determina-
tion of material constants is not onerous [31,32]. Differential
formulations are often used because of their simple forms and
easy adaptations to finite element analysis [11,25]. In the present
work, a non-linear viscoelastic constitutive model in differen-
tial form developed in Xia et al. [25] was used to describe the
mechanical behaviour of the epoxy resin. In this model, the total
strain rate is assumed to be the sum of the elastic and the creep
strain rate,
ε̇ij = ėij + ċij , (3)
where ε̇ij , ėij and ċij are the total strain rate, the elastic strain
rate and the creep strain rate, respectively.
The stress rate σ̇ij is then calculated through Hooke’s law,
σ̇ij = Cijkl ėkl = Cijkl (ε̇kl − ċkl ), (4)
where Cijkl are the stiffness components and are related to the
Young’s modulus E and the Poisson’s ratio ν of the material.
As shown in Xia et al. [25], viscoelastic material behaviour
for epoxy resin can be well described by a combination of Kelvin
elements connected in series in a uniaxial representation. In this
case, creep strain rate is expressed as the sum of the strain rate
of each Kelvin element, i.e.
 n n 
 
m Sijkl 1 m
ċij = ċij = σkl − c , (5)
E m τm τm ij
m=1 m=1
where Sijkl is the compliance components (an inverse of the stiff-
ness components Cijkl ), τ m = ηm /Em (m = 1, 2, . . ., n) denotes the
retardation time with Em being the spring stiffness and ηm being
the dashpot viscosity for the mth Kelvin element, respectively.
The retardation time τ m in Eq. (5) determines a time duration
after which the contribution from the individual Kelvin element
becomes negligible. Therefore, the number of Kelvin elements
adopted in the constitutive equation depends on the required
time range. For simplicity, a time scale factor β was introduced
in Xia et al. [25] and it was assumed that
m−1
τm = (β) τ1 . (6)
485
In this way, all τ m are related through the scale factor β. A time
span of order of n would be covered if n Kelvin elements were
adopted and the value of β was taken to be 10 [25].
The experimental data in Xia et al. [33] also show clearly the
non-linear viscoelastic behaviour of the studied epoxy polymer.
The description of the non-linear behaviour in the above model
was achieved in Xia et al. [25] by letting Em be a function of the
current equivalent stress, thus,
 
σeq − 22.764
Em = E1 (σeq ) = 1.055 × 105 exp − . (7)
18.0
Also for epoxy resin, it was shown in Xia et al. [33] that the
material shows different behaviour in uniaxial tension and com-
pression. To account for this, the equivalent stress was defined
as:

(R − 1)I1 + (R − 1)2 I12 + 12RJ2
σeq = , (8)
2R
where I1 = σ 1 + σ 2 + σ 3 is the first invariant of the stress tensor,
J2 = Sij Sij /2 the second invariant of the deviatoric stress Sij and
R is the ratio of the tensile to compressive “yield stress”. When
√= 1, Eq. (8) reduces to the von Mises equivalent stress σeq =
R
3J2 . As illustrated in Xia et al. [25], the stress–strain behaviour
predicted by the above model agreed very well with experimental
data for varying loading rates and loading histories.
2.3. Cure and temperature-dependent material properties
The composite constituents considered here are glass fibre
and epoxy resin. The properties of glass fibre are assumed to
remain constant and independent of temperature with Young’s
modulus E = 72.5 GPa, Poisson’s ratio ν = 0.22 and the coeffi-
cient of thermal expansion χ = 5.0 × 10−6 ◦ C−1 [12].
For epoxy resin, the Young’s modulus is a function of the
temperature T and the degree of cure α, which is expressed as
[24]:
E0
E(T, α) = α (9)
cosh(aT )b
where E0 is the Young’s modulus for fully cured resin at room
temperature, and a and b are constants. The function (9) is also
applied for the spring stiffness Em of each Kelvin element to
simulate the dependency of spring stiffness on the temperature
and the degree of cure, i.e.,
E1 (σeq )
Em = α . (10)
cosh(aT )b
The thermal expansion coefficient χ of the epoxy resin is
expressed as a linear function of the temperature with a slope of
χl − χr
K= (11)
Tl − T r
where χr = 63 × 10−6 ◦ C−1 for room temperature Tr = 23 ◦ C and
χl = 139 × 10−6 ◦ C−1 for Tl = 110 ◦ C [12]. Therefore, the ther-
486 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
Table 2
Values of parameters of the viscoelastic model for the epoxy resin [25]
Parameters Values
E0 (MPa) 3400
ν 0.42
β 10
τ1 6.116
R 1.15
a 0.012
b 1.44
mal expansion coefficient χ of the epoxy resin is given by:
χl − χr
χ= (T − Tr ) + χr . (12)
Tl − T r
In addition, the Poisson’s ratio for epoxy resin is taken to be
constant and assumed to be independent of degree of cure and
temperature. Values of the material properties and the model
parameters required by Eqs. (3)–(12) are all given in Table 2
[25].
With the cure and temperature-dependent material properties
proposed above, the constitutive model presented in Section 2.2
is able to describe the viscoelastic behaviour of the epoxy resin
throughout the curing process. For fully cured resin at room tem-
perature, the predicted tensile stress–strain behaviour is shown
in Fig. 2 for three different strain rates which agree with both
the predictions in Xia et al. [25] and the experimental data in
Xia et al. [33].
3. Analysis of residual stress
For constrained shrinkage and thermal cooling contraction of
the resin, the total induced strain can be expressed as:
dεij = deij + dcij + δij ds + δij χ(T ) dT, (13)
where dεij are the total strain increments, deij the elastic strain
increments, dcij the creep strain increments, ds the free shrinkage
strain increment due to the chemical reaction (cross-linking) in
the absence of constraint, χ(T) the thermal expansion coefficient
Fig. 2. Predicted tensile stress–strain behaviour of the fully cured epoxy resin
at room temperature.
which is dependent on the temperature, δij the Kronecker delta
and dT is the temperature change.
From Eq. (13), the stress–strain relationship can be derived
as:
dσij = Cijkl dekl = Cijkl {dεij − dcij − δij ds − δij χ(T ) dT },
(14)
where dσ ij are the stress increments and Cijkl is the stiffness
components related to the Young’s modulus E and the Poisson’s
ratio ν of the material.
In Eqs. (13) and (14), the creep strain can be obtained from
the viscoelstic model in Section 2.2 and the thermal strain can
be obtained from the assigned thermal expansion coefficients
and temperature history. The chemical shrinkage strain needs
to be determined from the volumetric shrinkage of the resin
associated with the reaction process. For a given incremental
change in the degree of cure during reaction dα, the associated
change in specific volume of the resin dV can be expressed as
[21,34]:
dV = dα Vsh
T
(15)
where VshT is the total volume change in the resin at the end
of cure. The isotropic resin shrinkage strain of a unit volume
element of resin, resulting from an incremental volume rein
shrinkage is then given by [21,34]:
√
ds = 1 + dV − 1.
3
(16)
The cure shrinkage in the resin during the whole cure process is
the cumulative sum of all the incremental contributions as deter-
mined through Eqs. (15) and (16). The fibre itself is assumed
not to undergo any chemical contraction and creep deformation
during cure. Thus, Eq. (14), which was derived for the residual
stress analysis in the resin, is still applicable for fibre reinforce-
ment by simply ignoring the creep and the chemical shrinkage
strains.
4. Finite element model
In composites, the actual fibre distribution is quite random
over a cross section. For simplicity, most micromechanical mod-
els assume a periodic arrangement of fibres for which a unit
cell can be isolated. The unit cell has the same elastic con-
stants and fibre volume fraction as the composite. The periodic
fibre sequences commonly used are the square, square-diagonal
and hexagonal arrays. In this work, the square array as shown
in Fig. 3 is primarily considered in the finite element analysis
and the matrix is assumed to be perfectly bonded to the fibres
throughout the analysis. In addition, the square-diagonal and
hexagonal arrays were also considered in order to study the effect
of fibre packing on the generation of residual stress.
Two micromechanical unit cell models are considered
depending on the type of load applied after the curing process.
For normal loading, due to the symmetry only a quarter of the
unit cell (Fig. 4a) is considered. Boundary conditions used for
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498 487

Fig. 3. Illustration of the Representation Volume Element or unit cell (enclosed

by dashed lines) for a square fibre array.

the quarter unit-cell model are:

u = 0 at face ABFE, v = 0 at face BCGF
and w = 0 at face ABCD, (17)

Equal u for face DCGH, equal v for face ADHE

and equal w for face EFGH, (18)
where u, v and w stand for the displacements in the x, y and
z-directions, respectively. Also, equal u, v and w means that the
entire face has the same displacement in the x, y and z-directions,
respectively.
For longitudinal shear loading or combined transverse normal
and longitudinal shear loading, a whole unit cell (Fig. 4b) was
considered in order to facilitate the satisfaction of deformation
requirements for shear. Boundary conditions used for residual
stress analysis of the whole unit cell are:
u, v, w = 0 at node B, v, w = 0 at node C
and v = 0 at node F. (19)
In addition, the following kinematic constraints are needed to
meet the periodic boundary conditions of the whole unit cell:
Fig. 4. (a) A quarter of unit cell, (b) the whole unit cell and (c) predicted longi-
u(EFGH) − u(ABCD) = u(F) − u(B) (20) tudinal shear deformation for square packing.
v(EFGH) − v(ABCD) = v(F) − v(B) (21)
To satisfy the deformation requirements for longitudinal
w(EFGH) − w(ABCD) = w(F) − w(B) (22) shear loading or combined transverse normal and longitudinal
shear loading applied after curing process, boundary conditions
u(DCGH) − u(ABFE) = u(C) − u(B) (23)
(19) used for the residual stress analysis need to be modified
v(ADHE) − v(BCGF) = v(A) − v(B) (24) [35]. (Note that the rest of boundary conditions (20)–(24) remain
unchanged.) The modified boundary conditions (19) become:
where u(EFGH), u(ABCD), u(DCGH) and u(ABFE) are the
x-direction displacements for faces EFGH, ABCD, DCGH u = 0 for face ABFE, v = 0 for face BCGF
and ABFE; v(EFGH), v(ABCD), v(ADHE) and v(BCGF) the y- and w = 0 at the model centre. (25)
direction displacements for faces EFGH, ABCD, ADHE and
BCGF; w(EFGH) and w(ABCD) the z-direction displacements Following the above boundary and constraint conditions
for faces EFGH and ABCD; u(F), u(B) and u(C) the x-direction (20)–(25), the predicted deformed shape for pure longitudinal
displacements for nodes F, B and C; v(F), v(B) and v(A) the y- shear loading is illustrated in Fig. 4c where shear traction
direction displacements of nodes F, B and A; and w(F) and w(B) was applied at face ABFE and face DCGH in the z-direction.
are the z-direction displacements for nodes F and B. Note that the deformed shape for longitudinal shear shown in
488 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
Sun and Vaidya [35] was obtained for displacement-controlled
loading where the equal displacement in the z-direction for
face ABFE and face DCGH was imposed. For traction loading
applied to the model in this work, the equal displacement in the
z-direction for face ABFE and face DCGH is ignored and the
obtained deformed shape shows non-equal displacement in the
z-direction for these faces (see Fig. 4c). If equal displacement
in the z-direction for face ABFE and face DCGH was imposed,
the model would be over constrained when subjected to traction
shear loading.
The mesh generated for the quarter model is shown in Fig. 5
where the fibre volume fraction is taken to be 60% and 15-noded
three-dimensional wedge elements are used. The number of
elements is approximately 1300 for the quarter unit cell model
(5200 for the whole unit cell model). The average element size
within the fine mesh region is around 4% of the fibre radius. Note
that the mesh shown in Fig. 5 only has one layer of elements in
the fibre direction. Analyses have been carried out for meshes
with multi-layers (up to 20) elements in the fibre direction,
and the same strain and stress fields (independent of the z-
coordinate) were obtained. Hence, for computational efficiency,
the mesh with only one-layer elements in the fibre direction
(see Fig. 5) was considered throughout this work. Furthermore,
mesh sensitivity analyses have shown that the mesh shown in
Fig. 5 is fine enough to produce accurate results compared to a
doubly refined mesh, with a difference within 0.3% in terms of
Fig. 5. Finite element mesh for a quarter of unit cell for square packing (dashed
line indicates the fibre bounding).
residual stress and failure load level. In addition, analysis was
also carried out using 20-noded three-dimensional quadrilateral
elements and the results are almost the same as those for
15-noded wedge elements used throughout in the present work.
5. Damage and failure prediction
As shown in Fig. 3, at the microscale the model consists of
fibre reinforcement and resin matrix. Depending on the failure
mechanism, the following criteria for damage onset prediction
were used.
For transverse failure, damage tends to occur within the resin
matrix and is related to the stress state within the (x, y) plane. In
this case, failure is predicted from the maximum principal stress
criterion in the (x, y) plane, i.e.,
σmax
(x,y)
≥ σut , (26)
(x,y)
σmin ≤ σuc , (27)
(x,y) (x,y)
where σmax and σmin are the maximum and minimum principal
stresses in the (x, y) plane, and σut and σuc are the tensile and
compressive strengths of the resin, respectively.
For longitudinal failure, damage might occur in both the fibre
and the resin. Obviously, longitudinal failure is due to the nor-
mal stress in the fibre direction and the failure criterion can be
expressed as:
σzz ≥ σut , for tensile loading (28)
σzz ≤ σuc , for compressive loading (29)
where σ zz is the normal stress in the z-direction (fibre direction)
and σut and σuc are the tensile and compressive strengths of the
fibre or the resin, respectively.
Under longitudinal shear loading, failure is expected to occur
within the matrix resin due to its relatively low strength. In this
case, the onset of damage is based on the overall maximum
principal stress within the matrix, i.e.,
σmax ≥ σut , (30)
σmin ≤ σuc , (31)
where σ max and σ min are the overall maximum and minimum
principal stresses, and σut and σuc are the tensile and compressive
strengths of the resin, respectively.
Material strengths used for failure prediction are taken from
Soden et al. [42] with σut = 2150 MPa and σuc = 1450 MPa for
the glass fibre, and σut = 80 MPa and σuc = 120 MPa for the
epoxy resin.
In simulating material damage, it is common practice to
reduce the stiffness (or stiffness in a certain direction) to a near
zero value following the onset of damage. Selective and non-
selective stiffness reduction schemes are often used. Selective
schemes are typically applied for composites where the load-
carrying nature is dependent on the damage orientation [36]. In
the present study, under normal loading, damage is distinguished
by the transverse and longitudinal failure for matrix and thus a
selective scheme is used. Specifically, for transverse failure only
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498 489

the modulus in the transverse direction was reduced to a near

zero value (0.01 times the original value) after damage onset,
while for longitudinal failure, only the modulus in the longitu-
dinal direction was reduced (also 0.01 times the original value).
Under shear loading, damage is independent of the material ori-
entation and non-selective stiffness reduction is applied. In this
case, once the failure criterion is satisfied, the modulus for the
resin was reduced to a near zero value (0.01 times the original
value), including the shear modulus, irrespective of the material
direction. On the other hand, failure of the fibre corresponds to
final failure of the composites. Therefore, the modulus of the
fibre was non-selectively reduced to a near-zero value at the
location of damage onset.
The stiffness degradation scheme, together with the resid-
ual stress analysis from equation (14), was programmed into
a user-defined material subroutine (UMAT) interfaced with the
commercial finite element code ABAQUS standard [37]. During
the analysis, the stress level was calculated at the Gauss integra-
tion points for each time increment and examined for damage
detection using the above failure criteria. Once the failure crite- Fig. 6. Contour plot of the maximum principal residual stress (MPa) for
rion was satisfied, the stiffness reduction was applied for further Vf = 60%.
analysis until final failure of the model. A similar damage model
has been used in Zhao et al. [38] for damage progression anal- The distribution of the maximum principal residual stress
ysis in E-glass/vinylester non-crimp fabric composite laminates in the resin along the interface is presented in Fig. 7, where
and in Zhao et al. [13] for the study of effects of residual stress the elastic solution and the chemical shrinkage contribution are
on transverse failure of polymer–matrix composites. also included. Compared to the purely elastic solution, a reduc-
tion in residual stress was predicted due to the stress-relaxation
6. Results and discussion caused by the viscoelastic behaviour of the epoxy matrix. Also
the results show that the chemical shrinkage of resin makes only
6.1. Residual stress a relatively small contribution to the overall residual stress due to
the relatively low modulus at the high cure temperature (150 ◦ C).
The cure process considered here has two stages: curing at At the central position in the interface of the quarter unit cell
150 ◦ C for 3 h and thermal cooling from 150 ◦ C to 23 ◦ C (room (θ = 45◦ ), the maximum principal residual stress due to chemi-
temperature) at a cooling rate of 2 ◦ C/min. Therefore, finite ele- cal shrinkage has a value of 5.9 MPa, about 12% of the overall
ment analysis was performed in two discrete steps, where step maximum principal residual stress 48.8 MPa. This indicates that
one is the chemical shrinkage stress analysis at 150 ◦ C and step the curing shrinkage still makes a reasonable contribution to the
two is the thermal cooling stress analysis from 150 ◦ C to 23 ◦ C. overall residual stress, and should be included for stress analysis
The shrinkage residual stress was calculated by applying a given in polymer composites.
amount of resin volume shrinkage. For the epoxy resin con-
sidered here, the total volume shrinkage Vsh T was chosen to be

3% [39–41], which corresponds to a shrinkage strain of 0.99%

(about 1%).

6.1.1. Distribution of residual stress

A contour plot of the maximum principal residual stress is
shown in Fig. 6 for a square fibre packing with a fibre volume
fraction of 60%. It can be seen that, as expected, the resin experi-
ences a tensile maximum principal residual stress while the fibre
has a compressive maximum principal residual stress. The great-
est value (48.8 MPa) occurs in the central region (at θ = 45◦ ) of
the fibre–matrix interface of the quarter unit-cell model, within
the resin. According to the maximum stress failure criterion,
the residual stress can introduce resin failure along the interface
of the fibre and the resin, which agrees with the experimental
observation of interface microcracking shown in Gentz et al. Fig. 7. Distribution of the maximum principal residual stress in the rein along
[2,3]. the fibre/matrix interface immediately after cooling to 23 ◦ C (Vf = 60%).
490 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498

Fig. 8. Distribution of the maximum principal residual stress in the resin along
the fibre/matrix interface for three different fibre volume fractions.

6.1.2. Effects of fibre volume fraction and packing

The effects of fibre volume fraction on process-induced resid-
ual stress were studied for square packing arrangement. Fig. 8
shows the maximum principal residual stress in the resin along
the fibre–matrix interface for three different volume fractions,
i.e., Vf = 50%, 60% and 70%, respectively. For all three volume
fractions, the distribution of maximum principal residual stress
follows a similar pattern, with the greatest value at θ = 45◦ and
the lowest values at θ = 0◦ and 90◦ . The magnitude of residual
stress was seen to increase with increase in fibre content, since
higher fibre content tends to prevent free shrinkage of the resin
more significantly and causes increased residual stress. With
the increase of fibre volume fraction, the maximum principal
residual stress always remains tensile at θ = 45◦ . However, at
θ = 0◦ and 90◦ , the maximum principal residual stress tends to
become compressive with the increase of fibre volume fraction.
For Vf = 70%, it is noticed that a compressive maximum prin-
cipal residual stress, up to −50.2 MPa, was developed at θ = 0◦ Fig. 9. Contour plot of the maximum principal residual stress (MPa) within the
and 90◦ . Increased tensile residual stress at θ = 45◦ might facili- matrix for: (a) hexagonal packing and (b) square-diagonal packing (Vf = 60%).
tate crack initiation or interface debonding in these areas, while
developed compressive residual stress at θ = 0◦ and 90◦ might
be beneficial in preventing interface debonding in those areas
[26,27].
In addition to the square packing, hexagonal and square-
diagonal packing (see the inserted schematic drawings in Fig. 9
for hexagonal and square-diagonal packing) were also consid-
ered to record the effects of fibre packing on residual stress
(fibre volume fraction was 60% for both packing arrangements).
Contour plots of the maximum principal residual stresses for
hexagonal and square-diagonal packing are shown in Fig. 9
(see Fig. 6 for the results of square packing). Distributions of
the maximum principal residual stress of the resin along the
fibre/matrix interface are presented in Fig. 10 for the three dif-
ferent fibre-packing arrangements. Clearly, the distribution of
residual stress is different for the three different fibre arrange-
ments. The magnitude of residual stress also depends on the
fibre arrays, with higher values for square and square-diagonal
packing and lower values for hexagonal packing. The magnitude Fig. 10. Distribution of the maximum principal residual stress in the resin along
of maximum principal residual stress for hexagonal packing is the fibre/matrix interface for three different fibre packing (Vf = 60%).
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498 491

about 36.8 MPa at θ = 45◦ , about 25% lower than those seen in
square and square-diagonal packing. This may be a result of the
more uniform distribution of resin which is seen in hexagonal
packing.
Furthermore, closer inspection of the contour plots of resid-
ual stress (see Figs. 6 and 9) indicates that, for each kind of fibre
packing, high and tensile maximum principal residual stress
tends to appear in a localised high resin content area (large
gaps between neighbouring fibres) while low maximum prin-
cipal residual stress appears in a localised low resin content
area (small gaps between neighbouring fibres). As expected,
high resin content area experiences severe overall contraction
and high tensile residual stress tends to be introduced in these
areas as a result of the constraint of surrounding fibres. On
the other hand, the severe contraction in the high resin content
area will bring the fibres in the composites much closer to each
other and introduces compression in the low resin content area,
which lowers the residual stress considerably in these areas (see
Figs. 6 and 9).
The fibre-packing study suggests that evenly distributed
fibres, as in hexagonal packing, could reduce the magnitude of
residual stress by weakening the overall fibre constraints. For
real materials, there is usually a fairly random distribution of
fibres across a section, e.g., some regions are close to square
packing and some are close to hexagonal packing. Such random
fibre distributions will also lead to non-evenly distributed resin
contents in composites. As a consequence, the distribution of
local residual stress in real composites will vary across the
sections. High resin content regions will fail predominantly due
to a high and tensile residual stress in these areas. This can be
observed from the sites of microcracking (interface debonding),
which are located along the fibre–matrix interface in the high Fig. 11. (a) Distribution of the maximum principal residual stress in the resin
resin content area as shown in Gentz et al. [2,3]. Further FE along the fibre/matrix interface after 48-h relaxation at room temperature and (b)
relaxation of the maximum principal residual stress at the centre of the interface
analyses for realistic fibre arrangements in real composites (θ = 45◦ ) against time.
are currently being undertaken, which will give additional
interesting information for the residual stress distribution and
the effect of residual stress on damage and failure behaviour At the outset (within 200 min), residual stress shows a rapid
in real composites. The analyses may be conducted on a reduction with time. Following this, the reduction becomes more
Representative Volume Element for random fibre arrange- gradual, and residual stress seems to relax towards a steady-state
ments following periodic boundary conditions, as shown in value (rather than disappear completely). In 48 h, the maximum
[43–46]. principal residual stress at the centre of the interface (θ = 45◦ )
was relaxed from 48.8 MPa to 44.8 MPa for Vf = 60% and from
6.1.3. Relaxation of residual stress 67.2 MPa to 62.0 MPa for Vf = 70%. An overall evolution his-
Due to the viscoelastic behaviour of the resin, residual stress tory of the maximum principal residual stress at the centre of
is expected to relax with time after the curing process. In this the interface (θ = 45◦ ) is shown in Fig. 12, where residual stress
section, the relaxation behaviour of residual stress was stud- development during the isothermal curing, thermal cooling and
ied over the equivalent of 2 days at room temperature after the relaxation are clearly indicated. For both volume fractions, the
cure processing. Relaxation behaviour of residual stress was also relaxation in residual stress (4–5 MPa) in 48 h seems to be com-
investigated by varying the cooling rate during the thermal cool- parable to the amount of residual stress (6–8 MPa) built up during
ing. In the following, analyses for square packing with two fibre isothermal curing (before cooling).
volume fractions (60% and 70%) were considered. Due to the stress-relaxation behaviour, it is expected that
Distributions of the maximum principal residual stress in the cooling rate may affect the residual stress. Distributions of the
resin along the interface are shown in Fig. 11a after 48 h follow- maximum principal residual stress in the resin along the inter-
ing the curing process. Residual stress is shown to be relaxed for face (Vf = 60%) are shown in Fig. 13 for three different cooling
both fibre volume fractions. Fig. 11b shows the relaxation of the rates, i.e., 10 ◦ C/min, 2 ◦ C/min and 0.1 ◦ C/min, respectively. For
maximum principal residual stress with time at the centre of the cooling from 150 ◦ C to 23 ◦ C, the corresponding periods are
fibre–matrix interface (θ = 45◦ ) of the quarter unit cell model. 12.7 min for 10 ◦ C/min, 63.5 min for 2 ◦ C/min, and 1270 min
492 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498

Fig. 12. Evolution of the maximum principal residual stress at the centre of Fig. 13. Distribution of the maximum principal residual stress in the resin along
the interface (θ = 45◦ ) during a complete cure process, including the relaxation the fibre/matrix interface for varying cooling rate (Vf = 60%).
profile after cure.

for 0.1 ◦ C/min, respectively. From Fig. 13, it can be seen that
a lower cooling rate (longer cooling period) will result in more
stress-relaxation, introducing less residual stress. However, the
difference is rather small, indicating a weak dependency of resid-
ual stress on the cooling rate.
6.2. Effects of residual stress on damage and failure
To study the influence of residual stress/strain on the response
of the unit cell model, the damage onset and final failure were
examined under mechanical loading for a square packing model

Fig. 14. Effect of residual stress on damage initiation for transverse (a) tensile loading and (b) compressive loading.
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498 493

with a fibre volume fraction of 60%. After curing and ther-

mal cooling analysis, distributed traction (normal or shear) was
applied to the model surfaces. At each time increment of the
analysis, the initiation and evolution of damage are monitored
using the maximum stress failure criterion and stiffness reduc-
tion technique described in Section 5. Effects of residual stress
were addressed in two respects: effects on damage onset and
effects on failure envelopes. The load considered includes trans-
verse and longitudinal normal loading and longitudinal shear
loading.

6.2.1. Effects on damage initiation

Under transverse normal loading in the x-direction, the loca-
tion and the load level for damage initiation are illustrated in
Fig. 14 for the cases with and without residual stress (quarter
unit cell model). The location of damage initiation is represented
by the small dark area. For tensile transverse loading, the cause
of failure was due to the tensile maximum principal stress in
the (x, y) plane which reaches the resin tensile strength. When
no residual stress is considered, damage initiates at the (y = 0, Fig. 15. Contour plot of the residual stress (MPa) in the x-direction for the
x = xmax ) corner of the quarter cell, C in Fig. 4(a). If residual matrix.
stress is included, damage initiates at the fibre/matrix interface
at y = 0. For compressive transverse loading in the x-direction,

Fig. 16. Effect of residual stress on damage initiation for longitudinal (a) tensile and (b) compressive loading.
494 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498

analyses show that residual stress does not influence the loca-
tion of damage initiation and damage always initiates at the
(y = 0, x = xmax ). In this case, the cause of failure was due to the
compressive minimum principal stress in the (x, y) plane which
reaches the resin compressive strength. However, the residual
stress state acts to increase the initial failure load level for ten-
sile loading and reduce the level for compressive loading. This
is due to the negative (compressive) residual stress at position S
in the x-direction as shown in Fig. 15. As discussed in Zhao et
al. [13], when no residual stress was considered, the maximum
principal stress in the (x, y) plane has the greatest positive value at
position S with a direction along the x-axis which will introduce
initial failure at position S (see the left-hand plot in Fig. 14a)
for tensile loading in the x-direction. Considering the residual
stress in the x-direction, a negative value (−51.4 MPa) exists at
position S (Fig. 15). This negative residual stress at position S in
the x-direction would delay the onset of failure by negating part
of the tensile maximum principal stress at position S through
superposition, thus bringing a beneficial effect. Accompanying
this beneficial effect, damage initiation at the (y = 0, x = xmax )
corner for no residual stress (the left-hand plot in Fig. 14a) was
switched to the fibre–matrix interface at y = 0 when the residual
stress is included (the right-hand plot in Fig. 14a). For com-
pressive loading in the x-direction, when no residual stress was
considered, the minimum principal stress in the (x, y) plane has
the greatest negative value at position S with a direction along
the x-axis which will introduce initial compressive failure at
position S (see the left-hand plot in Fig. 14b). And the negative
x-direction residual stress at position S would exacerbate the
onset of failure at position S by increasing the compressive min-
imum principal stress under compressive loading, thus having a
detrimental effect.
Under longitudinal normal loading, the location and the load
level for damage initiation are shown in Fig. 16 for the cases
with and without residual stress (quarter unit cell model). It can
be seen that under tensile loading, damage always initiates in Fig. 17. Contour plot of residual stress (MPa) in the z-direction for (a) the matrix
the matrix in the centre of fibre/matrix interface (θ = 45◦ ), irre- and (b) the fibre.
spective of the residual stress, because of concentration of the
normal stress in the z-direction in this region (Fig. 17a). Under
compressive loading, damage initiates in the fibre, but in the and without residual stress. Note that the whole unit cell model
centre of the interface (θ = 45◦ ) for the case without residual was used for the finite element analysis. It can be seen that under
stress and at the ends of the interface (θ = 0◦ and 90◦ ) for the shear loading, damage always initiates in the matrix, but at the
case with residual stress. This is because the z-direction resid- horizontal areas of fibre/matrix interface for the case without
ual stress in the fibre (Fig. 17b) switched the concentration of residual stress as a result of concentration of maximum principal
compressive normal stress in the z-direction from at the ends stress in these areas and at the four corners of the whole unit cell
of the interface (θ = 0◦ and 90◦ ) to at the centre of the interface model for the case with residual stress as a result of high tensile
(θ = 45◦ ). Residual stress decreases the initial failure load level maximum principal residual stress at the four high resin content
for longitudinal tensile loading due to the presented high tensile corners (Fig. 6). Residual stress decreases the initial failure load
z-direction residual stress in the resin, particularly at the cen- level for longitudinal shear loading due to the presented tensile
tre region (θ = 45◦ ) of the interface (Fig. 17a). Residual stress maximum principal residual stress in the high resin content area
decreases the initial failure load level for longitudinal compres- (Fig. 6), as for the transverse shear loading case studied in Zhao
sive loading because of the negative z-direction residual stress et al. [13].
in the fibre (Fig. 17b), but only slightly since the magnitude of
the negative residual stress in the fibre is small compared to the 6.2.2. Effects on biaxial failure envelopes
fibre’s compressive strength. Under biaxial longitudinal (z-direction) and transverse (x-
Under longitudinal shear loading, the location and the load direction) normal loading, failure envelopes were constructed
level for damage onset are shown in Fig. 18 for the cases with by considering different biaxial load ratios and the results are
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
Fig. 18. Effect of residual stress on damage initiation for longitudinal shear loading.
shown in Fig. 19a and b for initial and final failures, respectively.
Failure envelopes for no residual stress, as well as the test data
of final failure for unidirectional E-glass/epoxy composites from
Soden et al. [47], are also included for comparison. Initial failure
level corresponds to the onset of damage. With further increase
of load after damage onset, damage develops from the location
of onset and spreads over the unit cell. Final failure occurs when
Fig. 19. Effect of residual stress on failure envelopes for biaxial longitudinal
and transverse normal loading: (a) initial failure and (b) final failure.
495
the damage spreads across the section normal to the loading
direction which makes the unit cell unable to carry any further
load. From Fig. 19, it can be seen that by considering the resid-
ual stress, both initial and final failure envelopes are shifted and
contracted from those derived by excluding residual stress. A
similar shifting effect of residual stress on failure envelopes was
also shown in Zhao et al. [13] for transverse biaxial normal load-
ing. Also, Aghdam and Khojeh [29] showed that residual stress
caused a shifting of initial yield surfaces for unidirectional fibre-
reinforced metal–matrix composites under transverse loading.
By comparing Fig. 19a and b, it can be seen that initial and final
failure envelopes are very different in the regions with tensile
longitudinal loading, since the initial failure is mainly trans-
verse and occurs within the matrix, which is much earlier than
the final matrix or fibre failure. In the regions with compressive
longitudinal loading, the initial and final failure envelopes are
comparable due to the rapid spread of damage over the unit cell
after the damage onset.
From the failure envelopes in Fig. 19b, residual stress is
shown to have little effect on the load levels for longitudinal
final failure due to the high fibre strengths (see the two straight
edges on the left and the right), but greatly affects the transverse
failure behaviour due to the relatively low resin strengths. In the
transverse failure region, residual stress is mainly detrimental for
transverse compression by causing earlier compressive failure
as a result of the negative x-direction residual stress at position
S. For transverse tension, residual stress has a complex effect
depending on the load ratios. Residual stress is beneficial for
load ratio −14.0 < RLT < 4.0 and detrimental for 4.0 < RLT < 48.0
(RLT is the ratio of longitudinal load to transverse load). This
is attributed to the combined influence of negative x-direction
residual stress at position S (Fig. 15) and the tensile residual
stress at the central area (θ = 45◦ ) of interface (Fig. 6). In partic-
ular, for load ratio −14.0 < RLT < 4.0, transverse damage starts
at position S for no residual stress due to the local tensile max-
imum principal stress in the x-direction. When residual stress
is included, the negative residual stress at position S in the x-
direction (Fig. 15) would postpone the occurrence of failure
and bring beneficial effects by negating part of the local ten-
496 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498

sile maximum principal stress through superposition. While,

for load ratio 4.0 < RLT < 48.0, transverse damage always starts
at the central area (θ = 45◦ ) of the interface irrespective of the
residual stress. Thus residual stress turns out to be detrimental
due to the presented tensile residual stress at the central region
(θ = 45◦ ) of the interface (Fig. 6). It should be pointed out that,
for transverse tension loading in the x-direction, whether the
residual stress is detrimental or beneficial depends on both the
magnitude of the residual stress and the resin strength [13]. For
example, with relatively low resin strength, tensile residual stress
at the central area (θ = 45◦ ) of the interface would cause pre-
loading failure or earlier post-loading failure by reaching the
material strength easily, even in the presence of the negative x-
direction residual stress at position S. For relatively high resin
strength, the negative residual stress in the x-direction at posi-
tion S would become beneficial by delaying failure occurrence
[13].
Experimental data in Fig. 19b follow the predictions in
the tension–tension region, but show a big discrepancy in the
tension–compression region. There are four aspects which might
contribute to the discrepancy. Firstly, test data in Soden et al. [47]
were taken from the biaxial tests performed by Al-Khalil et al.
[48] on nearly unidirectional ±85◦ E-glass/MY750-epoxy tubes
(with a fibre volume fraction of approximately 60%), which are,
strictly speaking, bi-directional composites. Therefore, the tests
could overestimate the transverse compressive strength due to
the additional contributions from the fibres present in the trans-
verse direction. Secondly, the curing procedure for the tested
tubes was different, with 2 h at 90 ◦ C followed by 1.5 h at 130 ◦ C
Fig. 20. Effect of residual stress on failure envelopes for combined longitudinal
and 2 h at 150 ◦ C [47], which might introduce different residual shear and transverse normal loading: (a) initial failure and (b) final failure.
stress states in the tested tubes and affect their failure behaviour,
particularly for transverse loading situations. Thirdly, the stress-
based criterion for compressive damage prediction in the matrix ing, bi-directional composites. Therefore, the fibres present in
resin used in this work might underestimate the failure level, the transverse direction could make additional contributions to
since the test curve for compressive stress and strain shows a the transverse compressive failure strength. On the other hand,
softening behaviour [33]. This means that compressive strength the stress-based criterion, used in the present work for compres-
does not reflect the final failure correctly under compressive sive damage prediction in the matrix resin, might underestimate
loading. Other failure criteria, such as a strain-based failure cri- the failure level due to the softening stress–strain behaviour for
terion, may be more appropriate for detection of compressive compressive loading [33]. Also the curing procedure for the
failure in the resin matrix. Finally, non-ideal fibre distributions tested tubes was different, with 2 h at 100 ◦ C followed by 2 h at
and concentrations in real composites could also make contribu- 150 ◦ C [47], which might affect their failure behaviour, partic-
tions to this difference [43]. Further work is currently ongoing ularly for transverse loading situations, as a result of different
in order to clarify this interesting and complex issue. residual stress states introduced in the tested tubes. Finally, non-
Failure envelopes for combined longitudinal shear and trans- ideal fibre distributions and concentrations in real composites
verse normal loading are given in Fig. 20a and b where the could also make contributions to this difference [43]. In order to
shear loading is applied in the z-direction and normal loading is clarify this interesting and complex issue, further work is cur-
applied in the x-direction. With the inclusion of residual stress, rently ongoing.
failure envelopes for combined shear and normal loading are also As seen in Fig. 20a and b, in most cases residual stress is
shown to be shifted and contracted from those derived by ignor- shown to be detrimental for combined longitudinal shear and
ing the residual stress. The load levels for initial and final failure transverse normal loading. This is due to the dual influence of the
levels are very close to each other. Compared to experimental tensile maximum principal stress along the fibre–matrix inter-
data [47], predicted failure envelopes again show good agree- face (Fig. 6) and the negative residual stress in the x-direction
ment within the transverse tension region but large differences at the right-bottom corner of the quarter unit cell (position S in
within the transverse compression region. Test data in Soden et Fig. 15). However, there is an area with a low ratio of longitu-
al. [47] were taken from the combined tension–torsion tests per- dinal shear to transverse normal loading for which the residual
formed by Hutter et al. [49] on circumferentially filament wound stress appears to be beneficial as shown in Fig. 20b. In this area,
E-glass/epoxy tubes (Vf = 62%), which are again, strictly speak- the ratio of shear to normal loading is less than 1 and the load-
 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498
ing can be regarded as equivalent to the transverse x-direction
tensile loading for which the residual stress is beneficial for the
material studied (see Figs. 14a and 19b).
7. Conclusions
A thermo-viscoelastic micromechanical model and the finite
element method have been used to study process-induced resid-
ual stress in unidirectional fibre-reinforced polymer–matrix
composites. Viscoelastic behaviour was assigned to the epoxy
matrix with cure and temperature-dependent material properties.
From a three-dimensional unit cell, residual stress was computed
by considering the chemical shrinkage of the epoxy resin and the
thermal cooling contraction of the whole fibre and resin system.
Due to the stress-relaxation caused by the viscoelastic
behaviour of the epoxy matrix, a reduction in residual stress
was predicted when compared to a purely elastic solution. After
cure, residual stress tends to relax to a certain value and resid-
ual stress appears not to be affected significantly by varying the
thermal cooling rate.
Computed residual stress shows strong dependency on the
fibre volume fraction and fibre packing. A higher fibre vol-
ume fraction will result in much greater residual stress lev-
els due to stronger fibre constraints on the matrix contraction.
Fibre-packing studies suggest that evenly distributed fibres, as
in hexagonal packing, could reduce the magnitude of residual
stress by weakening the overall fibre constraints. For real mate-
rials, there is usually a fairly random distribution of fibres in a
section—some regions are squarely packed and some are hexag-
onally packed, which also leads to unevenly distributed resin
pockets. As a consequence, the distribution of local residual
stress in real composites will vary across the sections. Regions
with high resin contents will fail easily due to high, tensile resid-
ual stress in these areas.
Using the maximum stress failure criterion, effects of residual
stress on damage and failure of the model were addressed in two
aspects; effects on damage onset and effects on failure envelopes.
Residual stress shows clear influences on damage onset, in terms
of the location and the load level, for both normal and shear load-
ing. The inclusion of residual stress results in a contraction and
movement of the failure envelopes, for both initial and final fail-
ures, predicted for biaxial normal (longitudinal and transverse)
loading and combined shear (longitudinal) and normal (trans-
verse) loading. For final failure, residual stress shows little effect
on the load levels for fibre-dominated longitudinal failure, but
greatly affects the load levels for matrix-dominated transverse
and shear failure. It can be concluded that process-induced resid-
ual stress in fibre-reinforced polymer–matrix composites could
be detrimental or beneficial depending on the loading conditions
in service.
Acknowledgements
This work is funded by the UK Engineering and Physical
Sciences Research Council (EPSRC) through the Nottingham
Innovative Manufacturing Research Centre (NIMRC).
497
References
[1] H. Dannenberg, SPE J. (July) (1965) 669–675.
[2] M. Gentz, D. Armentrout, P. Rupnowski, L. Kumosa, E. Shin, J.K. Sutter,
M. Kumosa, Compos. Sci. Technol. 64 (2004) 203–220.
[3] M. Gentz, B. Benedikt, J.K. Sutter, M. Kumosa, Compos. Sci. Technol. 64
(2004) 1671–1677.
[4] H.E. Gascoigne, Exp. Mech. 34 (1994) 27–36.
[5] O. Kesler, J. Matejicek, S. Sampath, S. Suresh, T. Gnaeupel-Herold, P.C.
Brand, H.J. Prask, Mater. Sci. Eng. A257 (1998) 215–224.
[6] B. Benedikt, M. Kumosa, P.K. Predecki, L. Kumosa, M.G. Castelli, J.K.
Sutter, Compos. Sci. Technol. 61 (2001) 1977–1994.
[7] M.A. Stone, I.F. Schwartz, H.D. Chandler, Compos. Sci. Technol. 57 (1997)
47–54.
[8] P. Olivier, J.P. Cottu, Compos. Sci. Technol. 58 (1998) 645–651.
[9] A. Gopal, S. Adali, V.E. Verijenko, Compos. Struct. 48 (2000) 99–
106.
[10] B. Andersson, A. Sjögren, L. Berglund, Compos. Sci. Technol. 60 (2000)
2011–2028.
[11] Y. Chen, Z. Xia, F. Ellyin, J. Compos. Mater. 35 (2001) 522–542.
[12] Y. Zhang, Z. Xia, F. Ellyin, Compos. Sci. Technol. 64 (2004) 1613–
1621.
[13] L.G. Zhao, N.A. Warrior, A.C. Long, Int. J. Solids Struct. 43 (2006)
5449–5467.
[14] J. Lange, S. Toll, S.J.-A.E. Månson, Polymer 36 (1995) 3135–3141.
[15] F. Flores, J.W. Gillespie Jr., T.A. Bogetti, Polym. Eng. Sci. 42 (2002)
582–590.
[16] Y. Eom, L. Boogh, V. Michaud, Polym. Eng. Sci. 40 (2000) 1281–
1292.
[17] S.L. Simon, G.B. McKenna, O. Sindt, J. Appl. Polym. Sci. 76 (2000)
495–508.
[18] D. O’Brien, P.T. Mather, S.R. White, J. Compos. Mater. 35 (2001) 883–904.
[19] Y.K. Kim, S.R. White, Polym. Eng. Sci. 36 (1996) 2852–2862.
[20] H. Choo, M.A.M. Bourke, M. Daymond, Compos. Sci. Technol. 61 (2001)
1757–1772.
[21] T. Bogetti, J.W. Gillespie Jr., J. Compos. Mater. 26 (1992) 626–660.
[22] Y.K. Kim, S.R. White, Mech. Compos. Mater. Struct. 5 (1998) 327–
354.
[23] P. Prasatya, G.B. McKemma, S.L. Simon, J. Compos. Mater. 35 (2001)
826–848.
[24] E. Ruiz, F. Trochu, Composites: Part A 36 (2005) 806–826.
[25] Z. Xia, Y. Hu, F. Ellyin, Polym. Eng. Sci. 43 (2003) 734–748.
[26] R.P. Nimmer, J. Compos. Technol. Res. 12 (1990) 65–75.
[27] M.R. Wisnom, J. Compos. Mater. 24 (1990) 707–726.
[28] M.M. Aghdam, D.J. Smith, M.J. Pavier, J. Mech. Phys. Solids 48 (2000)
499–528.
[29] M.M. Aghdam, A. Khojeh, Compos. Struct. 62 (2003) 285–290.
[30] M.R. Kamal, Polym. Eng. Sci. 14 (1974) 231–239.
[31] Y.C. Lou, R.A. Schapery, J. Compos. Mater. 5 (1971) 208–234.
[32] W.G. Knauss, I. Emri, Polym. Eng. Sci. 27 (1987) 86–100.
[33] Z. Xia, Y. Hu, F. Ellyin, Polym. Eng. Sci. 43 (2003) 721–733.
[34] X. Huang, J.W. Gillespie Jr., T. Bogetti, Compos. Struct. 49 (2000)
303–312.
[35] C.T. Sun, R.S. Vaidya, Compos. Sci. Technol. 56 (1996) 171–179.
[36] D.M. Blackketter, D.E. Walrath, A.C. Hansen, J. Compos. Technol. Res.
15 (1993) 136–142.
[37] ABAQUS 6.5, Hibbitt, Karlsson and Sorensen Inc., Providence, RI, USA,
2005.
[38] L.G. Zhao, N.A. Warrior, A.C. Long, Compos. Sci. Technol. 66 (2006)
36–50.
[39] J.D. Russell, SAMPE Q. (January) (1993) 28–33.
[40] K. Oota, M. Saka, Polym. Eng. Sci. 41 (2001) 1373–1379.
[41] C. Li, K. Potter, M.R. Wisnom, G. Stringer, Compos. Sci. Technol. 64
(2004) 55–64.
[42] P.D. Soden, M.J. Hinton, A.S. Kaddour, Compos. Sci. Technol. 58 (1998)
1011–1022.
[43] V.N. Bulsara, R. Talreja, J. Qu, Compos. Sci. Technol. 59 (1999) 673–
682.
498 L.G. Zhao et al. / Materials Science and Engineering A 452–453 (2007) 483–498

[44] A. Wongsto, S. Li, Compos. Part A: Appl. Sci. Manuf. A36 (2005)
1246–1266.
[45] J.H. Oh, K.K. Jin, S.K. Ha, J. Compos. Mater. 40 (2006) 759–778.
[46] K.K. Jin, J.H. Oh, S.K. Ha, J. Compos. Mater., in press, 2006.
[47] P.D. Soden, M.J. Hinton, A.S. Kaddour, Compos. Sci. Technol. 62 (2002)
1489–1514.
[48] M.F.S. Al-Khalil, P.D. Soden, R. Kitching, M.J. Hinton, Int. J. Mech. Sci.
38 (1996) 97–120.
[49] U. Hutter, H. Schelling, H. Krauss, In Failure Modes of Composite Mate-
rials with Organic Matrices and Other Consequences on Design, NATO,
AGRAD, Conference Proceedings No. 163, Munich, Germany, October
13–19, 1974.