Thermoelastic Analysis of Residual Stresses in

Unidirectional, High-Performance Composites

JOHN A. NAIRN
Central Research and Development Department
E . I . DuPont de Nemours 6.Company
Experimental Station
Wilmington, Delaware 19898

We present a thermoelastic analysis of the composite cylin-

der model for a unidirectional composite including anisotropic
fibers and an interphase region. We have found the magnitude
of the residual thermal stresses on the micromechanics level
induced by differential shrinkage between the anisotropic fi-
bers and the matrix. For typical composites the largest residual
stress is tension along the fiber direction, and a simple lower
bound expression for this stress is given. Prediction of the
magnitude of the thermal stresses requires knowledge of the
thermal and physical properties of the matrix. The relevant
properties for epoxy and thermoplastic matrices are discussed.
The magnitude of the residual stresses can be reduced by
tailoring the interphase region, but only if the interphase region
serves to reduce the temperature for the onset of stress buildup.
The volume fraction dependence of the longitudinal and trans-
verse thermal expansion coefficients of the composite is com-
pared to analogous expressions in the literature which do not
include anisotropy of the fibers.

INTRODUCTION residual thermal stresses have not been a variable

H igh performance composite structures are
made with high modulus fibers, such as graph-
ite or Kevlar@aramid fibers, and some polymeric
matrix material. Current methods of combining the
fibers and the matrix all involve some processing
step at an elevated temperature. Because the ther-
mal expansion properties of the fibers and the ma-
trix are very different, residual thermal stresses will
build up as the part cools to room temperature. As
seen in Table 1, the total volume change on cooling
from the processing temperature is larger for the
matrices than for the fibers. These volume change
differences can be large, and large residual thermal
stresses are to be expected.
The magnitude of the residual thermal stresses in
high performance composites has been measured
by photoelasticity [l]and by deflection of an un-
balanced laminate [2,3].Bailey and coworkers have
suggested that the residual thermal strains affect
microcracking and stress-corrosion in cross-ply lam-
inates [3 to 61. Apart from Refs. 3 to 6, however,
little work has been done to characterize the effect
of residual thermal stresses on properties of high
performance composites. This gap is due, in part,
to the widespread use of epoxies as composite ma-
trices. Because the thermal properties of epoxy
matrices are not widely divergent, the inevitable
in research on high performance composites. Re-
cent interest in thermoplastic matrices may change
things. The differential shrinkage between ther-
moplastic matrices and graphite or Kevlar@fibers is
generally larger than between typical epoxy matri-
ces and the same fibers. Before thermoplastic com-
posites can be understood, it is important to inves-
tigate the magnitude and the effect of residual
thermal stresses on high performance composites.
In this paper, we have analyzed thermal stresses
in unidirectional composites on the micromechanic
level; that is the residual stresses that are expected
in the matrix of a single, unidirectional lamina. The
residual stresses on the micromechanic level are
distinct from the residual stresses that arise in mul-
tidirectional laminates due to the anisotropy of the
thermal expansion coefficients of individual lamina.
We have estimated the magnitude of the thermal
stresses from the component properties by solving
for the thermal stresses in the composite cylinder
model. The composite cylinder model is a model
for a unidirectional composite consisting of concen-
tric cylinders. Surrounding the fiber are two hollow
cylinders representing an interphase region and the
bulk matrix. The problem of thermal stresses in
concentric cylinders has been studied by several
people [7 to 121. The work presented here is an
extension of that work to include anisotropy of the

POLYMER COMPOSITES, APRIL, 1985, Yo/. 6, No. 2 123

 John A. Nairn

Table 1. Mechanical Properties of Some Matrices and Fibers.

Material Property Value Ref.
Graphite EL 220 GPa
ET 14 GPa
UA 0.20
UT 0.25
ffL -0.36 ppm/"C [17-181
ffT 18 pprn/"C
AV 0.54% (177°C to 25OC)

Kevlar' 49 4 124 GPa

ffL -2 ppm/OC
ffT 59 pprn/OC [321
AV 1.8% (177" to 25OC)

3501-6 Epoxy E 4.3 GPa

U 0.34 [17-181
ff 40 pprn/OC
AV 1.8% (177OC to 25°C)

Polysulfone AV 3.2% (T, (185°C) to 25°C) 1311

Polyethylene Terephthalate AV 13.6% (T, (200°C) to 25°C)' [321
Polyethylene AV 22% (Tc(120°C) to 25OC) [341
* T. = Temperature for onset of crystallizationwhen cooled at l.!PC/min.

fiber. The discussion of the solution is directed

specifically towards composites containing graphite
or Kevlar@fibers.
COMPOSITE CYLINDER MODEL
In this section, the method of solving for thermal
stresses in the composite cylinder model is outlined.
The composite cylinder model, illustrated in Fig. I,
includes a matrix, a fiber, and an interphase region.
Fiber
We assume that the matrix and the interphase are
linearly elastic and isotropic and that the fiber is
linearly elastic and transversely isotropic. The Motrix
bonds between the components are assumed to be
perfect. lnterphose
We begin by writing down the form of the
Fig. 1 . Cross section of composite cylinder model for a unidirec-
stresses in each component. The stresses in the tional composite.
matrix and in the interphase are taken from the
general solution for an isotropic, hollow cylinder
under internal pressure, external pressure, and uni- Crf = A7
form axial stress [13]. The results are:
u+f = A7 (3)
uzf = As
where As and As are constants.
The shear stresses in all three components are
zero. This fact follows from shear-lag analysis [14,
1.51. According to Rosen [lS] the shear stress at the
interface is a maximum at the fiber end and decays
and according to
T of e-r/a (4)
where x is distance from fiber end and rl in fiber
diameters is

gZi = A6
Here r, 9, and z indicate radial, hoop, and longitu-
dinal stresses, subscript rn and i are for matrix and
interphase, and A1 to A6 are constants. The stresses
in the transversely isotropic fiber have the form
(see Appendix)
For typical graphite and epoxy properties (see Ta-
ble I ) with V'= 50 percent, 9 is less than Yj fiber
diameters. That is, the shear stresses are end effects
which rapidly decay to zero and can be ignored in

124 POLYMER COMPOSITES, APRIL, 1985, MI. 6, NO. 2

 Themnoelastic Analysis of Residual Stresses

a composite cylinder model for a continuous fiber Table 2. Elements of the Matrix A in €9 9.
composite. Matrix
The solution of the composite cylinder problem Element Expression
is now reduced to determining the constants Al to
As by using the boundary conditions. The method
is an extension of the solution presented by Poritsky
to solve for thermal stresses in concentric, linear- --V m
elastic, isotropic cylinders [7]. The boundary con- ELVI
ditions for the stresses are
um=O at r = b
-2 :i;- + - 2)
u r n = ud
uri = urf at
at r= c
r=a
(6) -(&+a)
2 (% + v,4)
u,,,,(b2- c') + uzi(c2- a') + uzka2)= 0 Em 1- v, €;

The last condition is force balance in the z direction. --1

From these conditions the problem reduces to Em

A2 = -b2A1 2u;
-

A4 = A ] ( 1 - 5) - As
Ei

1
-
Ei

-
1 -
vm v, - ur
€; i -
E T uT)

As = A3("'1') a2 + A6( 1 - $)
There are now four unknowns-Al, A3, AS, and
A6. They are found by using the following four
displacement restrictions:

eZm = cZi
czi
at r = c
at r = a
= czf
(8)
- -
1-u; -
(
vm
€; 1 - v,
+----
1-urn
Em
l+u,
Em
1
1 - v, 1
u m = u d at r = c
urf= u~ at r =a -2
-
E;
where tZ is longitudinal strain and ur is radial dis-
placement. With the radial symmetry in the cylin- -u,
-
der model ur = re@.Using the conditions in E9 8 and Ei
the stress-strain relations for transversely isotropic
and isotropic materials, the problem reduces to four
equations in four unknowns: fTF

where Ts is the temperature for onset of stress build

up, TF is the final temperature, and A3(T) is the
constant A3 found by using the mechanical proper-
ties of the components at temperature T. In the
and the elements of the matrix A are in Table 2.
discussion that follows, we set T = -1 and calculate
Here, T is the temperature change and it is negative
stress build up per degree of cooling.
for cooling. Because T appears as only a linear term
and does not appear in the matrix A , we can set T
%
THERMAL STRESSES-NO INTERFACE
= -1 and consider stress build up per degree of
cooling. The total stress build up can be found by When the volume fraction of interphase is zero,
multiplying the stress per degree of cooling by the E 9 9 reduces to a 2 by 2 linear system. Setting c =
total temperature change on cooling. The situation a, adding the first and second rows and the third
in which the materials have temperature dependent and fourth rows of E 9 9 results in
mechanical properties can easily be handled by
setting T = 1 and integrating the stresses from E9 9
over the cooling cycle. For example

POLYMER COMPOSITES, APRIL, 7985, Vol. 6, No. 2 125

 John A. Nairn

where the elements of A"' are in Table 3. The fiber

stresses A7 and As are found with Eq 7 by setting c
= a and setting As = As = 0. This result is a
"'1
250 I

50
I I

AS/3501-6
I I
I I I I I I

generalization of Poritsky's result I71 to include

anisotropy of the inner cylinder. (Note: A! does ~

not reduce to Poritsky's results for th case of an a

?
isotropic fiber. The reason is due to an error in the
last step of Poritsky's derivation [7]).
In Fig. 2 is plotted the radial dependence of all
the stresses for a graphite/epoxy composite with Vf
= 50 percent. The matrix and fiber properties are -150- i -
for Hercules 3501-6 epoxy resin and Hercules AS
-200
graphite fiber. These properties are listed in Table
1 . The data in Table 1 come from the work of -250 I 1 I I 1 I 1 I

Adams, et al. 116 to 181. The largest stress in the

matrix is the longitudinal stress; it is a tensile stress
and equal to 193 kPa/"C. This result can be used
to calculate total longitudinal stress at room tem-
perature, but first we need to know from what
temperature the stress begins to build up. As shown
in other work, the thermal stresses in epoxy matri- 350 - AS/3501-6
ces begin to build up from the cure temperature - -
300
111. For Hercules 3501-6, the cure temperature is Aa€d(l -v)
177°C. If it is assumed that the matrix properties 250
5 Longitudinal
are independent of temperature below the cure
temperature, we calculate that the longitudinal
stress in the matrix at 25°C will be 29.3 MPa. This
stress level is approximately 25 percent of the ul-
timate tensile strength of the matrix.
The transverse stresses, ,u and ugm, are also
plotted in Fig. 2. The sum of these two stresses is a
constant. This fact must follow from the absence of -50 -
shear stresses which require +,, to be independent -
of r. Note that the radial stresses are compressive. -'ooO.O d.1 d2 Oj3 d4 d5 d6 0:7 018 d9 1.0
Fiber Volume Froction
In other words, the matrix with a larger thermal
Fig. 3. Tiiernal stresses in the matrix as a function offiber volume
expansion coefficient than the fibers has been fraction. Results f o r Hercules AS grapizitejher in Hercules 3501-
shrink fit around the fiber. The required properties 6 epoxy matrix.
for a shrink fit are discussed below.
The matrix stresses as a function of fiber volume
fraction are plotted in Fig. 3. The longitudinal stress
the transverse direction is smaller than A a = a, -
at2or the differential thermal expansion in the lon-
is zero at zero fiber volume fraction and rises rap-
gitudinal direction. The transverse fiber modulus,
idly eventually reaching 230 kPa/"C. Two limiting ET, is not sufficiently large to be considered a rigid
results for longitudinal thermal stresses in concen- rod when compared to the matrix modulus. The
tric cylinders are also plotted in Fig. 3. The line other limiting curve labelled AaE,,F where
labelled AaErn/(l - urn) is the longitudinal stress
expected for a thin matrix shell around an isotropic
rigid rod where A a = a",- aL.The matrix stresses
do not reach this upper limit for two reasons. A ~ T
= awl- CZT or the differential thermal expansion in is the stress expected for one dimensionally con-
strained thermal shrinkage. This curve applies
Table 3. Elements of Matrix A" in Eq 11.
when there is free thermal shrinkage in the trans-
~~~ verse direction and the matrix is constrained lon-
Matrix gitudinally by a rod with modulus E L . Because umr
Element Expression
and ( T ~are
~ ,not
~ zero, the thermal shrinkage in the

transverse direction is not free and we find that

AaE,,F is always below uZm;that is, AaErnFis a
useful lower bound for longitudinal stresses in ma-
trices of unidirectional high performance compos-
ite. The conditions which make it a lower bound
are discussed below. The AaE,F has a rapid rise at
low Vfwhich is analogous to the rapid rise in uzm.
From Eq 1 1 it can be seen that the origin of this
A::
-
A:; rapid rise is the fact that EL >> Em and F rapidly
2
approaches 1 as Vf increases.

126 POLYMER COMPOSITES, APRIL, 1985, Vol. 6, No. 2

 Thermoelastic Analysis of Residual Stresses
At all volume fractions the radial stress is com-
pressive, indicating a shrink fit of the matrix around
the fiber. In real composites, close packing of fiber
can result in "tricorn" regions [19] which can
change the compressive radial stresses to tensile
stresses. Asamoah and Wood [ 191 have shown that
the transition from compressive to tensile stresses
in hexagonal arrays does not occur until the volume
fraction exceeds 78 percent. Although local high
volume fraction regions may occur and lead to
tensile stresses at the fiber matrix interphases, we
expect that at commercial fiber loadings (50 to 60
percent) that the compressive radial stresses are
dominant.
In concentric isotropic cylinders, the thermal
stessses are zero when the thermal expansion coef-
ficients are equal. Because the fiber has two expan-
sion coefficients, the thermal stresses can never be
eliminated by changing the thermal expansion coef-
ficient of the matrix. It is possible, however, to
eliminate either the longitudinal matrix stress or
the transverse matrix stresses (radial and hoop).
This point is illustrated by plotting the matrix
stresses as a function of matrix expansion coefficient
for a composite with Vj-= 50 percent (see Fig. 4).
Either the longitudinal or the transverse matrix
stresses can be equal to zero, but they are zero at
different values of a,.
Both the radial and the hoop stresses are zero
when a,,,= 13.4 ppm/"C. This value of a, (denoted
a!,) is easily found from E 9 11; it's found by setting
A1 = 0 and solving for A3 and a:,. The result is
(1 + R )
a,",= aT
(1 + VA) + (1 + hll)R
VA vlltR
+ a L (1 + VA) + (1 + vm)R
where
and
6oorr-------
1 1
500
' --T7
A AV, = AV,
I
AS/3501-6
I the thermal stresses in a KevlarO composite than it
-200I I I I I I I
0 10 20 30 40 50 60 70 80
Matrix Expansion Coef. (pprn/'C)
Fig. 4 . Thermal stresses in the matrix as a function of the matrix
thermal expansion coefficient. Results for Hercules AS graphite
fiber in Hercules 3501-6 epoxy matrix. Vertical line labeled AVf
= AV, is matrix thermal expansion coefficient which results in
thermal volume expansion match between fibers and matrix.
POLYMER COMPOSITES,APRIL, 1985, Vol. 6, No. 2
A3 = czm = (am - ~L)E,,~FT
( 15 )
In high performance composites made with high
modulus fibers such as Kevlar" or graphite, R is
much larger than 1 and a:, is given approximately
by
When a, is equal to a:, the longitudinal matrix
stress is given exactly by E 9 15 and the transverse
stresses are zero. For typical composite matrices,
however, amis greater than a:. In this case E q 15
is a useful lower bound to the longitudinal matrix
stress and the radial stresses are compressive; i.e.,
the matrix will shrink fit around the fiber. If a matrix
is found with a, less than a:, then E 9 15 will be an
upper bound to the longitudinal matrix stress and
the fiber matrix interface will be in tension. If v, =
V 2 , E q 16 reduces to
3a: = a: = 2aT + a L = ff; (17)
and the point for elimination of the transverse
stresses is when the volume expansion coefficient
of the matrix is equal to the volume expansion
coefficient of the fibers. For typical composite ma-
trices, u,,, is about 1/3 and a:, is slightly larger than
1/3 ay. The value of a?,, which leads to volume
expansion match with graphite fibers is plotted in
Fig. 4-it is very near the point of zero transverse
stresses.
The value of a,,,which eliminates the longitudinal
matrix stress is below a:. Even if a matrix with such
a low expansion coefficient could be found, it would
(13) probably not make a desirable composite matrix.
The reason is that below a:,, the radial stresses at
the fiber/matrix interphase are tensile and a very
weak interphase could be expected.
Based on Fig. 4 and the restriction that the radial
stresses at the interface are not tensile, the optimal
matrix for minimization of thermal stresses is when
a, is e ual to (given by E q 13) which occurs
3
when amis approximately equal to a;. Because a;
for Kevlar@fibers is closer to a: for typical matrices
than is a)' for graphite fibers, it is easier to minimize
is in a graphite composite.
COMPOSITE THERMAL EXPANSION
COEFFICIENT
Much work has been done towards predicting the
thermal expansion coefficients of composites from
the properties of the components [12, 20 to 231.
Schapery [ZO]has shown that for the particular case
of unidirectional isotropic fibers is an isotropic ma-
trix, the composite longitudinal and transverse ex-
I pansion coefficients are
90 100
and
127

where
v = U," V", + Uf Vf (20)
In composites with isotropic fibers, Eqs 18 to 20
are exact when the Poisson's ratios of the fiber and
the matrix are equal. In composites with anisotropic
fibers, Yates, et al. [24] have suggested that Eys 18
to 20 are valid if af and Ef in Eq 18 are replaced by
aL and E L of the fiber, and af and uf in Eq 19 are
replaced by aT and U A of the fiber. In this section, I
examine the applicability of Eqs 78 to 20 to the
composite cylinder model with an anisotropic fiber.
The complete solution to Ey 1 1 defines all the
displacements in the composite cylinder model. By
calculating the longitudinal and radial displace-
ments, per degree the composite thermal expansion
coefficients can be found. The results as a function
of fiber volume fraction for a graphite/3501-6 com-
posite along with results from Eys 18 to 20 are
plotted i n Fig. 5. The results show that Eq 18 is an
excellent approximation to aLc found from the com-
posite cylinder model; the two results for aLcin Fig.
5 are superposed. The condition for exact applica-
bility ofEq 18 with anisotropic fibers can be derived
from the composite cylinder model. The longitudi-
nal composite expansion coefficient from the com-
posite cylinder model is
where A1 and A3 are found with E9 11. By substi-
tuting the expressions for A1 and A:, it is possible to
show that Eq 18 with v f = U A and E f = E L is exact
when um = uA. That is, the analog of Schapery's
condition that Poisson's ratios of the fiber and the
matrix are equal is the Poisson's ratio of the matrix
equals the axial Poisson's ratio of the fibers.
Equation 18 is also accurate when the transverse
stresses (a, and ulm) are zero, which, as shown
+ol 0
-10 I I I I I I I I I
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.8 09
Fiber Volume Fraction
Fig. 5 . Coitiparison of the predictions f . r composite thermal ex-
punwon coefficients from the composite cylinder model (labeled
CC,%f)aiitl from the moclifierl Schapery 1.201approach as suggested
l x j Yutm, et ul 1241 (laheled S-Y) Note c x l r (ccm) rind ( ~ (S-Y)
1 ~
ure in close enorigh agreement to he sriperposable Only the lube1
alCi s given $)r the two results
128
John A. Nairn
above, occurs when thermal expansion coefficient
of the matrix is equal to all,. In this case, cZm =
AaE,,F and Eq 21 reduces to Eq 18. This second
condition for applicability of Eq 18 applies regard-
less of differences between the matrix Poisson's
ratio and the fiber axial Poisson's ratio.
Equation 19 for a T C has a shape similar to the
composite cylinder model. At Vf = 1, however, it
does not approach the correct limit which is a T for
the fiber. This result is a direct consequence of the
assumptions used by Yates, et al. [24] to extend
Schapery's result [2O] to composites with aniso-
tropic fibers. No such simple extension will com-
pletely take the fiber anisotropy into account. For
many purposes, however, the general agreement
for aLc is adequate.
THERMAL STRESSES-EFFECT OF
INTERPHASE
There is experimental evidence that tailoring the
interphase region between the fiber and the matrix
can affect the composite properties [25, 261. Using
the composite cylinder model with an interphase
region, we now look for effects of the interphase
region on the matrix thermal stresses in a composite
with Vf= SO percent. In what follows, the inter-
phase region is 10 percent of the composite by
volume and is identical to the matrix in all proper-
ties except one. The residual thermal stresses in the
matrix will be calculated as a function of the unique
property.
The effect of the thermal expansion coefficient
of the interphase, a,, of the matrix thermal stresses
is plotted in Fig. 6. As a, increases above am,all
matrix thermal stresses are reduced in magnitude.
At high a,, it is possible to eliminate the transverse
thermal stresses in the matrix. In Fig. 4 , the point
of zero transverse thermal stresses is when ai = 182
ppm/"C. When a, = 182 ppm/"C, the transverse
matrix stresses are eliminated, but the interphase
itself will have very large thermal stresses. This
effect is due to the large thermal property mismatch
that now exists between the interphase and both
the matrix and the fibers. As the interphase region
AS/3501-6/10% Inl.
Longitudinal
e
g 100
7
I Rodial
1.0
-100: , I L 1 I
50 100 150 200 250
lnterphose Expomion Coef. (ppm/*C)
Fig. 6. Thermal stresses in the matrix as a function of interphase
thermal expansion coefficient. Results f o r Hercules AS graphite
fiber in Hercules 3501-6 epoxy matrix. Interphase properties same
as matrix except for its thermal expansion coefficient.
POLYMER COMPOSITES, APRIL, 1985, Vol. 6, No. 2
 Thermoelastic Analysis of Residual Stresses
gets smaller, the airequired to eliminate the trans-
verse thermal stresses gets larger. When the volume
fraction of the interphase region is lowered to 5
percent, the required ai is 323 ppm/"C. In short,
by increasing the thermal expansion coefficient of
the interphase region, the matrix thermal stresses
can be reduced, but this approach is limited by the
increase of residual thermal stresses in the inter-
phase itself.
The effect of the interphase modulus and Pois-
son's ratio on the thermal stresses in the matrix can
easily be checked. When varied over a reasonable
range, neither the interphase modulus nor the in-
terphase Poisson's ratio has any effect on the matrix
thermal stresses (plots not included).
In short, none of the mechanical properties of a
linearly elastic, perfectly bonded, interphase region
can be used to significantly lower the overall ther-
mal stress buildup per degree of cooling in a high
performance composite. Therefore, the thermal
stresses in the composite after cooling to room
temperature can be affected by the interphase only
if the interphase lowers the temperature for the
onset of thermal stresses. Although the stress per
degree of cooling would be unchanged, the tem-
perature range of the cooling cycle for which ther-
mal stresses are increasing would be reduced. An
example is provided by our results on the effect of
matrix properties on the buildup of thermal stresses
in the composite [l].In that paper, we showed that
with an amorphous thermoplastic matrix the ther-
mal stresses build up from the glass-transition tem-
perature, Tg.In contrast, with an epoxy matrix, the
thermal stresses build up from the cure tempera-
ture even when the glass-transition temperature for
the epoxy matrix is below the cure temperature.
These results suggest that an amorphous thermo-
plastic matrix with Tg below the temperature for
onset of thermal stress build up in the bulk matrix
would reduce the thermal stresses in the matrix.
Obviously, if pushed too far, this strategy would
lead to a composite with poor high temperature
properties.
The most likely situation where a low Tg inter-
phase might be helpful is for composites with high
melting-point, semicrystalline, thermoplastic matri-
ces [27]. It is well known that semicrystalline ther-
moplastics can have very large volume changes on
cooling from the melt to room temperature [l].If
all this volume change leads to thermal stress build
up, the matrix will either have very large thermal
stresses or become cracked. Typically, much of the
volume change on cooling takes place at high tem-
peratures during crystallization. An optimal inter-
phase, therefore, would be an amorphous thermo-
plastic with a Tgjust below the crystallization tem-
perature of the semicrystalline matrix. This com-
posite would avoid thermal stress buildup during
the largest volume changes of the matrix and would
have good high temperature properties.
DISCUSSION AND CONCLUSIONS
We have solved for the residual thermal stresses
in the composite cylinder model. The model applies
POLYMER COMPOSITES, APRIL, 1985, Vol. 6, No. 2
to unidirectional composites and includes aniso-
tropy of the reinforcing fibers. With no interphase
region, the largest residual thermal stress in the
matrix is the longitudinal stress. We have found a
simple lower bound expression for this stress; the
lower bound expression holds for typical compos-
ites where the volume expansion coefficient of the
matrix is larger than the volume expansion coeffi-
cient of the fibers. The radial stress is compressive
indicating a shrink fit of the matrix around the
fibers.
Depending on the matrix properties, the magni-
tude of the residual stresses can be quite large. It
can be expected that the residual stress might affect
matrix dominated properties of the composite such
as solvent resistance, fatigue, fracture, or compres-
sive strength. It is important to know the magnitude
of the residual stresses and the results presented
here indicate the required matrix properties for
their prediction. Those matrix properties are the
temperature dependence of the modulus, the ther-
mal expansion properties, and the temperature and
mechanism for the onset of stress build up.
A few comments can be made about the mecha-
nisms of stress buildup in three classes of matrix
resins-epoxies, amorphous thermoplastics, and
semicrystalline thermoplastics. For epoxies and
amorphous thermoplastics, the stresses build up
from the cure temperature and the glass-transition
temperature, respectively [ 11. The temperature for
the onset of stress buildup with semicrystalline
thermoplastic may be complex and dependent on
processing conditions. If the semicrystalline matrix
assumes enough solid-like character early in the
crystallization region, the constrained shrinkage
will be large, resulting in large residual stresses or
cracking. If the crystallization shrinkage is not con-
strained by the fibers, the stresses will be smaller
but the dimensional changes will be significant.
Either way, the fact that semicrystalline thermo-
plastics in general shrink much more than epoxies
or amorphous thermoplastics could be a potential
problem.
When evaluating the thermal expansion proper-
ties of a matrix resin, it is important to use the right
experiments. The relevant experiments for compos-
ite matrices are the low pressure volume versus
temperature experiments. These experiments re-
flect the low pressures typically used in composites
fabrication. The low pressure thermal expansion
properties for epoxies and many thermoplastics are
known (see, for example, I28 to 321). For thermo-
plastics, low pressure shrinkage is not equivalent to
mold shrinkage: in general mold shrinkage is much
lower than the low pressure shrinkage. Experimen-
tally, when the correct matrix properties are used,
the predicted thermal stresses correlate very well
with the observed thermal stresses [l].
Using the composite cylinder model, we have
investigated the applicability of Schapery's I201
expressions for composite thermal expansion coef-
ficients to composites with anisotropic fibers. His
expression for longitudinal thermal expansion coef-
ficient works very well. It is found to be exact when
129
 John A. Nairn
the axial Poisson’s ratio of the fiber is equal to the
Poisson’s ratio of the matrix. When his expression
for the transverse thermal expansion coefficient is
extended as suggested by Yates, et a2 [24] it is found
in fair agreement with the composite cylinder
model. As expected, a simple extension of Schap-
ery’s results [ZO] will not completely account for
fiber anisotropy.
The composite cylinder model solved here in-
cludes an interphase region. For linearly elastic,
perfectly bonded, interphase regions, the only
method of tailoring its properties to reduce com-
posite thermal stresses is to pick an interphase
which will lower the temperature for the buildup
of thermal stresses. We suggest that an amorphous
thermoplastic interphase may be beneficial in com-
posites with high melting point, semicrystalline,
thermoplastic matrices.
NOMENCLATURE
Em = Young’s Modulus.
GI, = Shear Modulus.
um = Poisson’s ratio.
VA = Axial Poisson’s ratio (ULT).
VT = Transverse Poisson’s ratio (urn).
am = Stress.
unl, = Displacement.
ffm = Linear thermal expansion coefficient.
a7 = Volume thermal expansion coefficient.
m, i , f = Subscripts for matrix, interphase, and fi-
ber.
L, T = Subscripts for longitudinal and trans-
verse.
r, 4, z = Subscripts for radial, hoop, and longitu-
dinal directions.
REFERENCES
1. J. A. Nairn and P. Zoller, J. Mater. Sci., 20,355 (1985).
2. F. R. lones, M. Mulheron, and J. E. Bailey, . J.
_Mater. Sci.,
18, 1533 (1983).
3. 1. E. Bailev. P. T. Curtis. and A. Parvizi, Proc. R. SOC. Lond.
k, 366,599 (1979).
4. F. R. Jones, A. R. Wheatley, and J. E. Bailey, “Composite
Structures,” p. 415, edited by I. H. Marshall, Applied Sci-
ence Publishers, Barking, (1981).
5. J. E. Bailey, and A. Parvizi,]. Mater. Sci., 16, 649 (1981).
6 . J. E. Bailey, T. M. W. Fryer, and F. R. Jones, “Advances in
Composite Materials,” Vol. 1, pg. 514, edited by A. R.
Bunsell, C. Bathias, A. Martenchar, D. Menkes, and G.
Verchery, Pergamon, Paris, (1980).
7. H. Poritsky, Physics, 5, 406 (1934).
8. H. Poritsky, Phil. Mag., 24, 209 (1937).
9. B. E. Gatewood, Phil. Mag., 32,282 (1941).
10. B. E. Gatewood, Quart. Appl. Math., 6, 84 ( 1 984).
11. R. A. Strub, Tans ASME, 7 5 , 73 (1953).
12. W. Schneider, Kunststoffe, 61, 273 (1971).
13. S. P. Timoshenko and J. N. Goodier, “Theory of Elasticity,”
p. 69, McGraw Hill, New York ( 1 970).
14. H. L. Cox, Brit. J . Appl. Phys., 3, 72 (1952).
15. B. W. Rosen, AIAAJournal, 2, 1985 (1964).
16. J, M. Mahishi and D. F. Adams, J. Mater. Sci., 18, 447
(1983).
17. G. C. Grimes, D. F. Adarns. andE. G. Dusablon, Northrup
Corp./Univ. of Wyoming, USA, Report NOR-80-158
(1980).
130
18. D. A. Crane and D. F. Adams, Univ. of Wyoming, USA,
Report UWME-DR-101-101-1( 1981).
19. N. K. Asamoah and W. G. Wood, J. Strain. Anal., 5, 88
(1 970).
20. R. A. Schapery,]. Compos. Mater., 2,380 (1968).
21. B. W. Rosen and Z. Hashin, Int. J . Eng. Sd.,8, 157 (1970).
22. G. A. Van Fo Fy, Soviet Phys. Doklady, 11, 176 (1966).
23. S. DasGupta, Int. J . Solids Structures, LO, 1221 (1974).
24. B. Yates, M. J. Overy, J. P. Sargent, B. A. McCalla, D. M.
Kingston-Lee, L. N. Phillips, and K. F. Rogers, J . Mater.
Sci., 13,433 (1978).
25. L. T. Drzal, M. J. Rich, and P. F. Lloyd, 5th Annual Meeting
of the Adhesion Society (1982).
26. L. Ying, SAMPE Quarterly, April (1983).
27. J. T. Hartness, 14th National SAMPE Conf., 26 (1982).
28. P. Zoller, J. Polym. Sci., Polym. Phys. Ed., 20, 1453 (1982).
29. P. Zoller,]. Polym. Sci., Polym. Phys. Ed., 18, 157 (1980).
30. P. Zoller,]. Polym. Sd., Polym. Phys. Ed., 18, 897 (1980).
31. P. Zoller,]. Polym. Sci., Polym. Phys. Ed., 16, 1261 (1978).
32. P. Zoller and P. Bolli, ]. Macromol. Sci. Phys., B18, 555
(1980).
33. G. Lubin, ed., “Handbook of Composites,” p. 281, Van
Nostrand Reinhold Company, New York (1982).
34. 0. Olabishi and R. Simha, Macromolecules, 8, 206 (1975).
APPENDIX
Consider the form of the stresses in hollow, ani-
sotropic cylinders under internal and external pres-
sure and uniform axial stress. By symmetry there
are no shear stresses, cZ. is a constant,
du
tr =-
dr
and
U
€6 =-
r
For an orthotropic material, the stress strain rela-
tions reduce to
and the equilibrium equations reduce to
da, 1
dr
~ +; (6, - a@) = 0
Differentiating the first equation in Eq A3 and sub-
stituting into E q A4 yields
We now restrict our anisotropy to transversely iso-
tropic materials. Under this restriction C,, =
C++,Crz= C+,, and E q A 5 reduces to the result for
isotropic hollow cylinders [13]. The form of the
stresses in the transversely isotropic fibers are,
therefore, expected to be the same as the form of
the stresses in isotropic solid cylinders.
POLYMER COMPOSITES, APRIL, 1985, Vol. 6, No. 2