JOURNAL OF

COMPOSITE
Article M AT E R I A L S
Journal of Composite Materials
0(0) 1–15
! The Author(s) 2017
Micromechanical modeling of anisotropic Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
water diffusion in glass fiber epoxy DOI: 10.1177/0021998317744649
journals.sagepub.com/home/jcm
reinforced composites

Abedin Gagani1, Yiming Fan2, Anastasia H Muliana2 and

Andreas T Echtermeyer1

Abstract
Fluid diffusion in fiber reinforced composites is typically anisotropic. Diffusivity in the fiber direction is faster than in the
transverse direction. The reason for this behavior is not yet fully understood. In this work, dealing with glass fiber epoxy
composite immersed in distilled water, an experimental procedure for determination of anisotropic diffusion constants
from a laminate is presented. The method has the advantage that it does not require sealing of the samples edges because
3-D anisotropic diffusion theory is implemented for obtaining the diffusion constants. A microscale model is presented,
where matrix and fiber bundles are modeled separately. The matrix properties have been obtained experimentally and
the fiber bundle properties have been deduced by the composite homogenized diffusivity model. The analysis indicates
that the anisotropic diffusion of the composite is due to inherent anisotropic properties of the fiber bundles.

Keywords
Fluid diffusion composites, micromechanical diffusion modeling composites, finite element diffusion composites

pipelines, the water may never penetrate the entire

Introduction thickness of the laminate, causing a gradient of water
Knowing long-term properties of composites is import- concentration in the composite. In order to obtain
ant when the materials shall be used in critical applica- better predictions of the long-term properties, a more
tions that are exposed to various environmental accurate knowledge of the concentration profiles of the
conditions. This topic is important for marine applica- water in the laminate needs to be known. It should also
tions,1,2 oil and gas industry,3 aerospace industry4,5 and be noted that a gradient of water in the laminate can
others. The long-term properties are typically dependent cause internal stresses altering the strength of the
on mechanical loading and environmental conditions material, when compared to the ones measured in
(temperature, moisture, oxidation etc.). Particularly, totally dry or wet conditions.10
the effect of fluid concentration on the mechanical prop- The water uptake can be calculated if the diffusivity
erties of polymeric composites has been discussed by is known. Measurement methods for diffusivity of iso-
Weitsman and Elahi.6 This paper concentrates on the tropic polymers, following the Fickian diffusion, are
diffusion of water, as the environment, through the well established. The typical method is given by
fiber reinforced polymer (FRP) composites, focusing ASTM D5229.11 When a very thin specimen is
on experimental, analytical and numerical studies.
The traditional approach to obtain long-term mech- 1
Department of Mechanical and Industrial Engineering, Norwegian
anical properties of material in water is to saturate the University of Science and Technology, Norway
2
material in water and to measure the mechanical prop- Department of Mechanical Engineering, Texas A&M, USA
erties subsequently.6–9 Using this approach gives dry
and fully saturated properties, obtaining the extreme Corresponding author:
Abedin Gagani, Norwegian University of Science and Technology
conditions. However, it takes very long time for the (NTNU), Department of Mechanical and Industrial Engineering, Richard
water to diffuse into the composite. For thick-walled Birkelandsvei 2B, 7010 Trondheim. Norway.
structures, such as hulls of ships or some high-pressure Email: abedin.gagani@ntnu.no
2 Journal of Composite Materials 0(0)

considered for the immersion test, e.g. are forming the large surface has solved the prob-
150  150  1.5 mm, following the ASTM D5229– lem.15–19 This is described in detail in the experimental
92(04) without sealing the edges, the diffusion process methods. In this paper, the testing for the composites
can be approximated by 1-D diffusion through the was done with open edges and diffusivity was calculated
thickness. Equation (1) links the mass uptake to diffu- with a 3-D approach. This method was shown to be the
sivity from solving the 1-D Fickian diffusion equation, most rigorous by Arnold et al.18
as described in Crank12 The glass fibers themselves are often seen as non-
permeable to water, assuming to have zero diffusivity.
"  X 2  2
#
8 1 eð2iþ1Þ ðhÞ D? t Several experimental investigations of moisture diffu-
MðtÞ ¼ M1 1 2 ð1Þ sion in fiber-reinforced polymer composites have sug-
 i¼0 ð2i þ 1Þ2
gested different diffusions along the fiber direction and
perpendicular to the fiber direction.20,21 The diffusivity
By fitting the exact solution of the diffusion equation along the fiber direction can be 2–4 times larger than
to an exponential function, the ASTM standard simpli- the transverse diffusivity for glass fiber laminates20,22
fies equation (1) to and about 14 times higher for carbon fiber rods.23
" Taking into consideration the anisotropy of the fiber-
  !# reinforced composites, Shen and Springer14 formulated
D? t 0:75
MðtÞ ¼ M1 1  exp 7:3 ð2Þ a relationship between the diffusivity of matrix, diffu-
h2
sivity of fibers, fiber volume fraction and fiber orienta-
tion in order to determine the diffusivity of composites
where M(t) is the water content, M1 is the water sat- in the lateral (across) fiber direction. The idea is based
uration content, t is time, h is the thickness and D? the on an analogy between moisture diffusion and thermal
diffusivity in the thickness direction of the plate. conduction.24 The model has been widely applied in
When dealing with isotropic materials having com- capturing the moisture diffusion in FRP compos-
plex geometries it is possible to predict water diffusion ites.25,26 Several theories were developed to predict
by numerically solving the governing differential equa- anisotropic diffusivity based on the diffusivity of the
tion either using finite difference or finite element (FE) matrix and the geometrical arrangement of the
method. For immersed epoxy and vinylester, Fan fibers.14,22 Shen and Springer14 proposed an expression
et al.13 determined water concentration in each element to predict transverse and parallel diffusivity from
with the finite difference method by a weighed average matrix diffusion. A comparison of micromechanical
of the nodal concentrations surrounding the element. analytical models has been reported by Bao and
The water concentration in the specimen was deter- Yee.16 It is noted that several experimental studies
mined by the accumulated concentration from all have shown that typical diffusivity of the FRP compos-
elements in the body. ites in the transverse direction is around 1.5–3 times
The diffusivity of an orthotropic composite ply is lower than the diffusivity of the corresponding resin,
anisotropic.14 The diffusivity along the fibers is different e.g. Bond.21 Experimental investigations mentioned
from the diffusivity across the fibers. The standard above have shown that diffusivity along the fiber direc-
method for obtaining diffusivity for a composite ply is tion can be 2–14 times higher than the one in the trans-
the same as for polymers. A flat sample is made, the verse direction. In this situation, it is clear that the
edges are sealed, the sample is fully immersed in fluid diffusivity in FRP composite cannot be adequately cap-
and the diffusivity from the Fickian model is calculated tured by considering only matrix and fiber constituents,
using equation (1) or equation (2). with zero diffusivity for the fiber. Rocha et al.22 pro-
The approach works well for obtaining the through posed an FE model with high diffusivity interphase, in
thickness diffusivity, because regular thin plies or order to explain anisotropic diffusivity in composites.
laminates can be tested. Sealing the edges remains to They concluded that the fiber barrier effect is partly
be a challenge though. It is usually assumed that the responsible for the anisotropic diffusion in composites,
through thickness diffusivity and in-plane diffusivity but this alone does not explain the effect observed. For
across the fibers are the same. This is a good assump- the system investigated here the existing theories could
tion for transversely isotropic materials. The diffusivity not predict the anisotropic behavior found experimen-
in fiber direction is more difficult to obtain. Using a tally and a new approach had to be developed for
standard thin laminate is not well suited for measuring describing the diffusion along and across fiber
diffusivity in the fiber direction. The large surface area directions.
would have to be sealed and it would take a long time Humeau et al.27 performed computed tomography
for water to get into the specimen. Making very thick scans of GF/epoxy samples and observed that voids
specimens and cutting them in a way that the fiber ends are elongated in the fiber direction. However, they
Gagani et al. 3

reported that the anisotropy in diffusivity of samples 800, in order to avoid erroneous weight gain
without voids was as great as for the samples with measurements.22
voids, and concluded that voids elongation could not
explain the anisotropic diffusion behavior.
Experimental methods
In this work, 3-D diffusion has been studied
experimentally using samples with fibers oriented in dif- Fiber volume fractions were obtained by burn-off tests
ferent directions. The effect of voids on the moisture and density measurements. The burn-off test was per-
saturation content has been studied theoretically and formed following ASTM Standard D3171.29 The
experimentally. The orthotropic diffusion constants volume fraction obtained experimentally was 59.5%.
have been obtained from the experiments using 3-D The difference between volume fraction obtained by
anisotropic diffusion theory. Finally, a micromechani- burn-off test and by density measurement was only 0.2%.
cal approach has been developed where fiber bundles Matrix diffusion constants were measured through a
are modeled as anisotropic. set of four neat resin samples, manufactured by resin
molding and degassing, in order to reduce potential
voids caused by molding. Similar to the glass-fiber
Materials and methods epoxy samples, four neat resin samples having dimen-
sions 50  50  1.5 mm were obtained by cutting with a
Materials water-cooled diamond saw, machining and grinding
A typical glass fiber epoxy used for wind turbine and using 800-grit sandpaper.
marine applications was selected for this study. The All samples were cured at room temperature for 24 h
glass fibers were HiPer-TexTM unidirectional (UD) fab- and post-cured in an oven at 80 C for 12 h, according
rics from 3B having average radius 9 mm and density to the manufacturer’s recommendations. Before condi-
2.5 g/cm3. The epoxy used was HexionTM RIMR135TM tioning the samples were dried at 60 C for 72 h in an
resin and RIMH137TM hardener mixed with a mass oven, monitoring their weight until no substantial
ratio of 100:30. The glass transition temperature of weight change occurred. The curing process and the
the resin is 84.7 C, the curing temperature 80 C and long-time drying at 60 C ensure that complete curing
its density is 1.2 g/cm3.28 A thick laminate was made occurs for both neat resin and composite samples. The
by vacuum-assisted resin infusion of 80 plies resulting samples are shown in Figure 2.
in an average ply thickness of 0.85 mm. Samples were The void volume fraction in the laminate was mea-
cut from this laminate using a water-cooled diamond sured by image analysis of optical microscope images,
saw in order to obtain samples with fibers parallel and opportunely converted in black and white, Figure 3.
transverse to the surface, as shown in Figure 1. The void volume fraction was obtained by analysis
For each of the configurations in Figure 1, four sam- of the image color histogram, and resulted in
ples having dimensions 50  50 mm were cut with a 0.44%  0.32%. Figure 4 shows the micrographs of
water-cooled diamond saw, machined to a thickness the FRP composite at different normal directions: lon-
of 1.5 mm and grinded with sandpaper having grit gitudinal fiber and transverse fiber directions.

Figure 1. Composite samples configurations. Dimensions are 50 mm  50 mm  1.5 mm.

4 Journal of Composite Materials 0(0)

A conditioning chamber was made by filling a Samples mass was measured regularly with a Mettler
thermally insulated bucket with distilled water. An elec- Toledo AG204 DeltaRange scale, having a sensitivity
trical resistance was placed in the bucket together with of 0.1 mg and a capacity of 61 g. They were removed
a thermocouple, in order to have a constant tempera- from the conditioning chamber four at a time, their
ture of 60 C  1 C. An aluminum sample holder was surfaces dried with a cloth and weighed immediately,
used for keeping the samples in a vertical position and in order to prevent water from diffusing out of the
to expose all flat faces to water. samples.
All specimens had open edges. In order to enforce
1-D flow conditions in a diffusion experiment it is good
practice to maximize the width-to-thickness ratio of the
samples or to seal the edges with stainless steel. The
sealant is often glued to the small surface of the sam-
ples’ edges. This glue may absorb water, influencing the
measurement, or, even worst, fall off from the sample
during the experiment. It was therefore decided to keep
open the edges of all samples, as the width-to-thickness
ratio was quite big: 33. Saturation was measured up to
506 h, when the weight gain curve had flattened out.

Modeling diffusion in FRP composites

Diffusivities of water in a polymer or composite can be
obtained from measuring the water uptake with time.
One approach is to simply measure the macroscopic
diffusivities in the principal orthotropic directions of
the material. This approach will be described first for
the simple 1-D case and the better 3-D approach. It is a
relatively simple method sufficient for characterizing
the diffusion behavior. An engineer who wants to deter-
mine the concentration profile of water within a com-
ponent made of FRP composites would not need any
additional information, provided the diffusion in FRP
composites can be modeled as Fickian diffusion.
Figure 2. Samples used for the diffusion experiment. The second approach explains diffusivity from the

Figure 3. (a) Optical microscope image showing voids; (b) converted black and white threshold image used for void volume fraction
evaluation.
Gagani et al. 5

Figure 4. (a) Laminate coordinate system; (b) intra-bundle void in the 2–3 plane; (c) intra-bundle void in the 1–3 plane.

micromechanics point of view, which gives insights into where l is the sample length, w its width, D? the lamin-
the reasons for anisotropic diffusion behavior. ate diffusivity in the direction perpendicular to the
fibers and Dk the laminate diffusivity in the direction
parallel to the fibers, as shown in Figure 5.
Macroscopic model
For the composite samples with transverse configur-
The 1-D water uptake from the macroscopic diffusion ation, equation (3) can be used, while for the samples
through the specimen thickness can be described by the with parallel configuration, equation (3) must be mod-
method given in the ASTM Standard,11 as written in ified to take into account the diffusion in the direction
equation (1). If the diffusivity of a composite laminate parallel to the fibers going through the thickness, as
is measured, this diffusivity obtained is typically in the follows
direction perpendicular to the fibers. When knowing
"  3 X 2
the specimen dimensions and the saturation level of 8 1 X 1 X 1 ð2iþ1Þ2 ðh Þ Dk t
e
the water uptake, the diffusivity D? can be found by MðtÞ ¼ M1 1 
2 i¼0 j¼0 k¼0 ð2i þ 1Þ2
obtaining the best fit for the mass uptake data vs. time ð4Þ
2  2 2  2
#
using equation (1). In the above equation, the effect of eð2jþ1Þ ðwÞ D? t eð2kþ1Þ ð l Þ D? t
volumetric changes (swelling) during the diffusion pro- 
ð2j þ 1Þ2 ð2k þ 1Þ2
cess is neglected. In some polymers immersed in water,
a relatively small volume change can cause deviation
from the Fickian diffusion.13 Figure 5 shows the diffusivities for each direction in
The above method is based on the 1-D diffusion and both transverse and parallel configuration. For the neat
requires the edges of the samples to be sealed. Sealing resin, due to the assumption of isotropy,
the edges with a diffusion-tight material can be difficult. D? ¼ Dk ¼ DM. Equation (3) becomes
Often metal foils are used, but they sometimes do not
"  3 X 2
bond easily to the polymer. When the foils fall off 8 1 X 1 X 1 ð2iþ1Þ2 ðh Þ DM t
e
during the experiment, the data can be severely dis- MðtÞ ¼ M1 1  2
 i¼0 j¼0 k¼0 ð2i þ 1Þ2
turbed. A simpler way to measure diffusivity of aniso- ð5Þ
2  2 2  2
#
tropic composites is to make samples without sealing eð2jþ1Þ ðwÞ DM t eð2kþ1Þ ð l Þ DM t
the edges. In that case a 3-D version of equation (1) is 
ð2j þ 1Þ2 ð2k þ 1Þ2
needed. The 3-D solution-to-mass uptake in an aniso-
tropic plate, following the Fickian diffusion, was
reported by Crank12 where DM is the matrix diffusivity.
"  3 X 2
1 X 1 X 1 ð2iþ1Þ2 ðh Þ D? t
8 e Microscopic model
MðtÞ ¼ M1 1  2
 i¼0 j¼0 k¼0 ð2i þ 1Þ2
ð3Þ The anisotropic diffusivity can also be predicted by
2  2 2  2
#
eð2jþ1Þ ðwÞ D? t eð2kþ1Þ ð l Þ Dk t using a micromechanics model based on mainly the
 diffusivities of the resin and fiber bundles in the com-
ð2j þ 1Þ2 ð2k þ 1Þ2
posite. A micromechanics model is formulated
6 Journal of Composite Materials 0(0)

Figure 5. Configuration of samples used for parallel and transverse diffusion measurement.

Figure 6. A simplified microstructure of FRP composite.

FRP: fiber reinforced polymer.

to determine the overall (homogenized) anisotropic dif-  is formulated based on the volume average of the flux
f,
fusivity of FRP composites, comprising unidirectional in the four subcells
fibers embedded in matrix. The micromechanics model
is based on a simple unit cell with four fiber and matrix 1X 4   1X 4

subcells, as depicted in Figure 6. f ¼ fðÞ XðÞ dVðÞ  VðÞ fðÞ ð6Þ

V ¼1 V ¼1
This unit-cell model has been successfully used to
determine the mechanical and non-mechanical response
of FRP composites, such as thermo-viscoelastic, piezo- where f(a) and V(a) are the flux and volume of subcell
electric, and heat conduction behaviors.30,31 The first (a), respectively, and X(a) is the spatial location of each
subcell is the fiber constituent and the remaining sub- point within the unit cell. It is also assumed that field
cells are matrix. It is noted here that the fiber constitu- variables within one subcell are uniformly distributed.
P
ent in case of typical FRP composite is actually a fiber The total volume of the unit cell is V ¼ 4¼1 VðÞ .
bundle that consists of fiber filaments, as shown in When P the fiber volume fraction is considered, then
Figure 4. Thus, fluid can penetrate within the filaments V ¼ 4¼1 VðÞ ¼ 1. The flux in the composite is
inside the fiber bundle, which might explain the rela- formed due to the gradient of fluid concentration and
tively high diffusivity along the fiber direction in FRP a linear relation is considered between the flux and con-
composites. Perfect bonding at the interphases between centration gradient in the homogenized composite
the fiber bundle and matrix is assumed, and a periodic
boundary condition is imposed on the unit cell. f ¼ D
 u;
 u ¼ grad ðCÞ ð7Þ
In determining the diffusivity of the composite, it is
also assumed that volumetric expansion (swelling) is Here D and u are the effective diffusivity and con-
negligible, thus both fiber and matrix diffusion behav- centration gradient of the homogenized composite,
iors follow the Fickian diffusion. The flux of the fluid respectively, and C is the water concentration in the
that passed through the homogenized composite faces, homogenized composite which is time and spatial
Gagani et al. 7

dependent. The concentration gradient of the homoge- ’ð1Þ ð2Þ ð3Þ ð4Þ
1
1 ¼ ’1 ¼ ’1 ¼ ’1 ¼ ’ ð15Þ
nized composite is related to the concentration gradient
of each subcell by formulating a concentration matrix The flux and water concentration across the fiber
for each subcell BðÞ . The idea of formulating the con- direction are expressed as
centration matrix to relate the field variables of the
constituents to the field variable of the composite was f ð1Þ ð2Þ
2 ¼ f2 ð16Þ
originally proposed by Hill32 for a linear elastic
response of composites. The concentration gradient in f ð3Þ ð2Þ
2 ¼ f2 ð17Þ
each subcell is given by
Vð1Þ ð1Þ Vð2Þ
’ðÞ ¼ BðÞ u ð8Þ ’2 þ ’ð2Þ ¼ ’ 2 ð18Þ
Vð1Þ þ Vð2Þ Vð1Þ þ Vð2Þ 2
A linear relation like in equation (7) is adopted for Vð3Þ Vð4Þ
ð3Þ
the constitutive relation in each subcell ’2 þ ’ð4Þ ¼ ’ 2 ð19Þ
Vð3Þ þ Vð4Þ Vð3Þ þ Vð4Þ 2
fðÞ ¼ DðÞ ’ðÞ ð9Þ f ð1Þ ð3Þ
3 ¼ f3 ð20Þ
where D(a) is the subcell diffusivity. Substituting equa-
tion (8) into equation (9) gives f ð2Þ ð4Þ
3 ¼ f3 ð21Þ

Vð1Þ Vð3Þ
fðÞ ¼ DðÞ BðÞ u ð10Þ ’ð1Þ
3 þ ð1Þ ’ð2Þ ¼ ’ 3 ð22Þ
Vð1Þ þV ð3Þ V þ Vð3Þ 3
Substituting equation (10) into equation (6) gives the
effective flux in terms of diffusivity properties of the Vð2Þ Vð4Þ
’ð3Þ
3 þ ð2Þ ’ð4Þ ¼ ’ 3 ð23Þ
constituents and volume content Vð2Þ þV ð4Þ V þ Vð4Þ 3

1X 4
Using the micromechanical relations, equations
f   VðÞ DðÞ BðÞ u ð11Þ
V ¼1 (14)–(23) and constitutive relations for the constituents
in equation (9), the concentration matrix for each sub-
Finally, the effective diffusivity of the composite is cell can be obtained. Finally, the effective diffusivity is
determined by comparing equation (7) and equation determined from equation (12). In this study, the fiber
(11) bundle is assumed transversely isotropic and the poly-
meric matrix is considered isotropic with regard to their
X4 diffusion behaviors. The fiber bundle diffusivity is
 ¼1
D VðÞ DðÞ BðÞ ð12Þ 
V ¼1 DðfÞ ¼ diag Dð11fÞ , Dð22fÞ , Dð22fÞ , while the diffusivity of
matrix is D(m) ¼ D(m) I3, where I3 is the identity
The conservation of mass of the fluid in the compos- matrix. The diffusivity of the composite is given as
ite body leads to  
D ¼ diag D 11 , D 22 , D 33 ; D 33 ¼ D 22 ð24Þ
dC 
¼ DivðfÞ ð13Þ 1  ð1Þ ð f Þ 
dt D 11 ¼ V D11 þ Vð2Þ DðmÞ þ Vð3Þ DðmÞ þ Vð4Þ DðmÞ
V
where t denotes the time variable. ð25Þ
In order to determine the diffusion behavior and dif-  
ðmÞ ð f Þ
fusivity of the homogenized composite, micromechani- 1 ð1Þ D D22 Vð1Þ þ Vð2Þ
D 22 ¼ V
cal relations between the subcells need to be obtained. V Vð1Þ DðmÞ þ Vð2Þ Dð22f Þ
Consider a unit-cell model in Figure 6 placed in the  ð1Þ  !
ðmÞ ð f Þ ð2Þ
Cartesian coordinate, fibers are aligned in 1 direction, D D 22 V þ V
þVð2Þ þ Vð3Þ DðmÞ þ Vð3Þ DðmÞ
and 2 and 3 are the directions lateral and transverse to Vð1Þ DðmÞ þ Vð2Þ Dð22f Þ
the fiber composite. The flux and water concentration
ð26Þ
along the fiber direction are expressed as
In order to solve for the governing equation in equa-
Vð1Þ f ð1Þ
1 þ Vð2Þ f ð2Þ
1 þ Vð3Þ f ð3Þ
1 þ Vð4Þ f ð4Þ
1 ¼ f 1 0 ð14Þ tion (13), the FE method is used and the constitutive
relation in equation (11) is implemented at each
8 Journal of Composite Materials 0(0)

Gaussian point within continuum 3-D elements. Each

continuum element has eight nodes. The concentration
of the fluid is evaluated at the nodes. The total amount
of fluid concentration in the composite during the dif-
fusion process is determined as follows

1X Ne
1X 8
CðtÞ ¼ CðmÞ VðmÞ ; CðmÞ ¼ CðmÞ ð27Þ
V m¼1 8 i¼1 i

where Ne is total number of FEs used in solving the

diffusion equation in the composite body. The amount
of fluid concentration at each instant of time in equa-
tion (27) is correlated to the fluid concentration
Figure 7. Mesh used for the FE analysis. FE: finite element.
obtained from experiment. The total fluid concentra-
tion is related to the fluid uptake as follows
equation (1), for the hypothesis of an infinitely wide
MðtÞ plate (a very thin plate). Alternatively, it is possible to
CðtÞ ¼ ð28Þ
M1 use 3-D diffusion solution in equation (5), where all
diffusion constants are set equal to DM. The use of
The micromechanics model is implemented in a con- equation (5) instead of the 1-D flow equation, equation
tinuum three-dimensional FE. At each Gaussian inte- (1), yields more accurate prediction of the diffusion
gration point within the element, the constitutive model constant because it accounts for both the finite width
for the diffusion behavior is determined from the micro- of the laminate and the diffusion flow through its edges.
mechanics model. In this study, the user material sub- The experimental results are compared with both equa-
routine within ABAQUS commercial FE, written in tions (1) and (5), in Figures 8 and 9, respectively.
Fortran, is used to integrate the micromechanics In Figure 8 the results are fitted using equation (1),
model to FE analyses. For the post-processing, the valid for 1-D diffusion in an infinitely wide plate.
total amount of fluid in the composites is determined The matrix diffusivity obtained is 0.0068 mm2/h
by integrating the field variables at the nodes obtained (1.89  106 mm2/s).
from the FE analyses. For this purpose, a Matlab code In Figure 9 is reported the fit between experimental
was generated for reading the nodal field variables and measurements and equation (5), valid for 3-D diffusion
numerically integrating them to obtain the overall fluid in plates of finite width.
concentration. The diffusivity obtained by the 3D method described
In order to simulate the fluid sorption mimicking the above is 0.0063 mm2/h (1.75  106 mm2/s). The 1-D
FRP and resin specimens, FE meshes with eight-node diffusion equation yields a higher diffusivity prediction
hexahedral elements were generated for the plate with than the 3-D diffusion equation, as it does not consider
dimension 50  50  1.5 mm. A convergence study was the edge effects as a cause for steeper weight gain curve.
first conducted in order to determine sufficient numbers The difference is quite small, 7.4%
of elements by comparing the overall fluid concentra-
tion to the analytical solution of isotropic diffusion in
resin. In this study, a total of 4000 elements was suffi-
Saturation level of the composite
cient in capturing the fluid sorption response. Figure 7 The saturation level of the neat resin obtained experi-
shows the FE meshes. mentally is 3.18%. The saturation level of the compos-
ite, assuming that no voids are present in the material,
can be obtained by7
Results
  ð1  vf Þm
Diffusivity of the matrix MC M M
1 ¼ M1 1  wf ¼ M1 ¼ 0:774 %
vf f þ ð1  vf Þm
The matrix weight increase at saturation is about ð29Þ
3.18%. Saturation has been defined as the moment
when the difference in two consecutive measurements where wf is the fiber weight fraction, vf the fiber volume
is lower than 0.5%. The diffusivity of the matrix can be fraction, m the matrix density, f the fiber density and
easily obtained for an isotropic polymer, by fitting the MM 1 ¼ 3.18% the matrix saturation moisture content.
experimental results to the 1-D diffusion solution in The predicted moisture saturation content, 0.774%,
Gagani et al. 9

Neat resin
3.5
3

Water uptake (%)

2.5
2
1.5
Experimental
1
0.5 Fit to Eq. (1)
0
0 5 10 15 20 25
time1/2 (h1/2)
Figure 8. Water uptake measured as % weight increase vs. time in hours. 1-D model.

density when filled and as equal to air density when

Neat resin empty, equation (32)
3.5
3 vf þ vm þ vv ¼ 1 ; wf þ wm þ wv ¼ 1 ð31Þ
Water uptake (%)

2.5
2 fv ¼ water ; ev ¼ air ffi 0 ð32Þ
1.5
Experimental
1 where water is water density and air is air density. We
0.5 Fit to Eq. (5)
simplify air & 0.
0 Therefore equation (30) becomes
0 5 10 15 20 25
time1/2 (h1/2)
MM V
1 vm m þ M1 vv water
Figure 9. Water uptake measured as % weight increase vs. time M1 ¼ ð33Þ
in hours. 3-D model. vf f þ vm m

The void volume fraction of the laminate was

is far from the result observed experimentally: 0.96%. 0.44%, the composite predicted saturation content is
This difference could be due to the presence of voids in therefore 0.99%, very close to the value obtained
the laminate. experimentally: 0.96%.
It is possible to extend equation (29) in order to take
into account the presence of voids Diffusivity of the composite along and across the
fiber direction
MM V
1 vm m þ M1 vv v
f
M1 ¼ MM V
1 wm þ M1 wv ¼
The diffusivities of the composite in the transverse and
vf f þ vm m þ vv ev parallel directions were computed using both 1-D and
ð30Þ 3-D diffusion models. The saturation level used was
0.96% for both sample configurations.
where wm is the matrix weight fraction, wv is the void In Figure 10(a) and (b), the experimental results are
weight fraction, MM 1 ¼ 3.18% the matrix saturation fitted using 1-D diffusion equation, equation (1). The
moisture content, MV 1 ¼ 100% the void saturation diffusivity constants obtained are 0.0051 mm2/h
moisture content, vf the fiber volume fraction, vm the (1.42  106 mm2/s) for transverse diffusion and
matrix volume fraction, vv the void volume fraction, f 0.021 mm2/h (5.83  106 mm2/s) for parallel diffusion.
the fiber density, m the matrix density, fv the filled In Figure 11(a) and (b), diffusivity was obtained by
void’s density and ev the empty void’s density. simultaneous fitting of transverse samples and parallel
We impose volume and mass conservation, equation samples weight gain curves with equations (4) and (5),
(31). We also define void density as equal to water respectively.
10 Journal of Composite Materials 0(0)

(a) Transverse (b) Parallel

1.2 1.2
1 1

Water uptake (%)

Water uptake (%)

0.8 0.8
0.6 0.6
Experimental Experimental
0.4 0.4
0.2 Fit to Eq. (1) 0.2 Fit to Eq. (1)
0 0
0 5 10 15 20 25 0 5 10 15 20 25
time1/2 (h1/2) time1/2 (h1/2)

Figure 10. Water uptake measured as % weight increase vs. time in hours, fitted with the 1-D equation.

(a) Transverse (b) Parallel

1.2 1.2
1 1
Water uptake (%)

Water uptake (%)

0.8 0.8
0.6 0.6
Experimental Experimental
0.4 0.4
0.2 Fit to Eq. (3) 0.2 Fit to Eq. (4)
0 0
0 5 10 15 20 25 0 5 10 15 20 25
time1/2 (h1/2) time1/2 (h1/2)

Figure 11. Water uptake measured as % weight increase vs. time in hours, fitted with the 3-D equation.

The regression was done using non-linear least squares

method. The fit is easily performed as the weight gain Table 1. Diffusivities obtained for neat resin and composite
curve of the transverse samples is dominated by the using 1-D and 3-D diffusion equation.
ply transverse diffusivity D? and the weight gain curve Neat resin Composite
of the parallel samples is dominated by the ply parallel
diffusivity, Dk . The transverse diffusivity obtained is D (mm2/h) D? (mm2/h) Dk (mm2/h)
0.0045 mm2/h (1.25  106 mm2/s) and the parallel diffu-
1-D Equation 0.0068 0.0051 0.021
sivity obtained is 0.02 mm2/h (5.56  106 mm2/s).
3-D Equation 0.0063 0.0045 0.020
Hence the parallel diffusivity obtained is 4.44 times
Micromechanical 0.0062 0.0042 0.020
higher than the transverse diffusivity.
FE model
Similar to the neat resin diffusivity, we observe that
in this case also, the 1-D diffusion equation predicts FE: finite element.
higher diffusivity compared to the 3-D diffusion equa-
tion, both for transverse diffusivity (11.8%) and paral- equations is smaller, because the diffusion through the
lel diffusivity (4.8%). One of the reasons for this edges occurs in the direction transverse to the fibers,
difference is the finite width of the sample that is and is therefore slower than the diffusion in the thick-
taken into account by equations (3) and (4). It is inter- ness direction that runs parallel to the fibers. The
esting to observe that for the transverse diffusion the results of the diffusivities characterized using macro-
deviation between 1-D diffusion results and 3-D diffu- scopic analytical model are summarized in Table 1.
sion results is higher than that for parallel diffusion.
This is because the 1-D diffusion results neglect the
Diffusion analyses using the micromechanical model
flow through the edges, which runs in the direction par-
allel to the fibers, and has therefore 4.44 times higher As discussed in the Microscopic model section, a micro-
diffusivity. For the parallel diffusion, the difference mechanical model, which is implemented in FE ana-
between results obtained by 1-D and 3-D diffusion lysis, is presented for determining the homogenized
Gagani et al. 11

diffusion behavior in FRP composites. The main using the micromechanics model, parametric studies
advantages of using the homogenization approach are were considered, see Appendix 1. From the analyses,
that it can be used to easily determine anisotropic dif- the diffusivity of the fiber bundle in the lateral direction
fusivities of FRP with different fiber volume contents is Dð22f Þ ¼ 0.0031 mm2/h (0.86  106 mm2/s), while
while incorporating different diffusivities of fiber the one in the longitudinal direction is Dð11f Þ ¼
bundle and matrix, making it easy for material selec- 0.0308 mm2/h (8.55  106 mm2/s). The diffusivity of
tions with desired diffusion performance, it permits pre- the fiber bundle along the longitudinal direction is
dictions of the diffusion behavior in laminated much higher than the diffusivity of fiber bundle in the
composites with different fiber orientations and it lateral direction, and also higher than the diffusivity of
allows performing diffusion of FRP structures with matrix. Figure 13(a) and (b) shows the diffusion behav-
more complex geometries and boundary conditions. ior in FRP composites, tested transverse and parallel to
In order to test the micromechanics and numerical the longitudinal fiber direction. From the micromecha-
methods, the diffusion constant of the resin was first nics analyses, the effective diffusivities can be easily
determined, both from the 1-D and 3-D analyses. In determined from equations (26) and (27). The overall
the 3-D analysis, the micromechanics model was con- diffusivities of the FRP with 59.5% fiber volume
sidered by setting the fiber volume content to zero. content are D 11 ¼ 0.021 mm2/h (5.80  106 mm2/s)
The diffusivity of resin from the 1-D analysis is and D 22 ¼ 0.0042 mm2/h (1.16  106 mm2/s), which
0.0067 mm2/h (1.86  106 mm2/s) and the one from are summarized in Table 1.
the 3-D analysis is 0.0062 mm2/h (1.71  107 mm2/s).
These values are very close to the analytical solutions
Discussion
results. The concentration of water during the diffusion
process is shown in Figure 12. Measuring the diffusivity of a solid immersed in a fluid
The micromechanics model is now used to determine is typically done by exposing specimens with sealed
the anisotropic diffusivity of fiber bundles in the FRP edges to a fluid and measuring the weight gain.
composites. Prior to determining the diffusivities of A simple 1-D formula stated in ASTM D522911 is
fiber bundles in the longitudinal and lateral directions often used to calculate the diffusivity. Testing with
sealed edges is experimentally difficult, as the sealant
may not bond properly giving distorted results.
Neat resin Performing tests with open edges and using 3-D diffu-
3.5 sivity equations makes the testing much easier. The dif-
3 fusivity is found by obtaining the best fit of the 3-D
Water uptake (%)

2.5 equation against the experimental data. The process is

2 not as easy as just looking at the slope of the curve in
1.5 the ASTM approach, but the best fit can be easily
Experimental
1 found using an optimization tool.
Micromechanical FE
0.5 For the water uptake of neat resin and using thin
0 plates the effect of the edges was small and using the
0 5 10 15 20 25
1-D or 3-D equation gave roughly the same results. It is
time1/2 (h1/2)
recommended though to use the 3-D equation as it is
Figure 12. Concentration of water in the resin from the more accurate and can also be used for thicker speci-
micromechanics and numerical analyses. men geometries with more pronounced edge effects.

(a) Transverse (b) Parallel

1.2 1.2
1 1
Water uptake (%)

Water uptake (%)

0.8 0.8
0.6 0.6
0.4 Experimental 0.4 Experimental

0.2 Micromechanical FE 0.2 Micromechanical FE

0 0
0 5 10 15 20 25 0 5 10 15 20 25
time1/2 (h1/2) time1/2 (h1/2)

Figure 13. Micromechanics analyses of the diffusion in FRP composites (a) transverse direction, (b) parallel direction.
12 Journal of Composite Materials 0(0)

The same can be said for measuring the water uptake of physical level another phase must be present that
composite laminates. allows different diffusion than observed in the resin.
Measuring diffusivity of composites in fiber direction The presented micromechanics model could predict
is inherently difficult. Typical composite laminates are the observed global anisotropy by giving the fiber con-
thin plates and the fibers are only exposed on one thin stituent (fiber bundle) an anisotropic diffusivity. The
surface. Transverse diffusion dominates such samples. diffusivity across the fiber bundles was less than that
Making thick laminates can compensate the problem of of the matrix and the diffusivity along the fibers was
a small exposed surface, but then the diffusion takes a higher than the homogenized diffusivity of the mater-
long time. Making thick laminates and cutting thin ials. While the presented micromechanics approach
slices creating thin specimens with a large exposed sur- seems to capture the anisotropy diffusivity in FRP com-
face for fiber ends solved the problem and good meas- posites with some rationale behind it, the detailed
urements could be taken. mechanisms of the diffusion in FRP composites are
The small amount of voids causes an increase of the very complex and still not fully understood. From an
saturation level of the composite by 24% (from 0.774% engineering point of view, it is sufficient to accurately
to 0.96%), and needs to be accounted for, since the measure the anisotropic diffusion characteristics. From
saturation level is directly related to the diffusivity. a materials point of view another phase allowing higher
Even a small void content has a large influence on the diffusivity must be present in the composite.
saturation level. However, voids are often distributed
nonuniformly, making it complicated in determining
Conclusions
the overall (average) response of composites by exam-
ining a representative microstructure of composites. This study presents experimental investigations and
The nonuniform distribution of voids also yields to modeling of diffusion behavior in FRP composites.
localized fluid concentration within the composites. The experimental part considered thin specimens of
For the glass fiber epoxy system investigated here, pure resin and FRP composites with fibers arranged
the diffusivity in the fiber direction was found to be parallel and transverse to the thickness of the speci-
about four times the diffusivity transverse to the mens. These specimens were immersed in distilled
fibers, see Table 1. This is a significant difference and water, and moisture uptakes were monitored until sat-
anisotropic diffusion should be considered when calcu- uration was reached. Specimens with unsealed edges
lating the fluid concentration profiles of structures and were used, which is experimentally much easier than
components. It is also noted that the diffusivity in the the standard approach of testing with sealed edges.
fiber direction is higher than the diffusivity of pure The experimental results show that Fickian diffusion
resin. These results show that when considering only can be used to capture the diffusion process in these
fiber bundle and matrix as constituents, the diffusivities specimens. The experimental tests also showed that
of the fiber bundle cannot be assumed as zero, espe- the existence of voids, although very small (around
cially for the longitudinal fiber direction. However, 0.44% volume content), can have a significant effect
assuming that individual glass fibers have 0 or very in the amount of water uptakes.
low diffusivity is often considered to predict diffusivity Two modeling approaches were considered in char-
of the composite by knowing the diffusivity of the resin acterizing the diffusion behaviors of FRP composites,
and geometric arrangement of the fibers. Using the with the Fickian diffusion model. The first approach
assumption that fibers have zero or nearly zero diffu- considers a phenomenological diffusion model and
sivity, e.g. the model developed by Shen and Springer14 both 1-D and 3-D diffusion behaviors were considered.
or the FE-based modes developed in Rocha et al.,22 It was found that although the differences in the diffu-
could not model the observed anisotropy. Several pos- sivities characterized using 1-D and 3-D approaches are
sible explanations to the above issue are as follows. rather small, it is recommended to consider 3-D
This discrepancy is most likely because within fiber fila- approach as it represents a more realistic condition.
ments in a single fiber bundle, resin and air pockets are 1-D models are appropriate for samples with sealed
present, which allow water to seep into fiber bundles. edges. It is noted that diffusivity of FRP composite
As a result, the net diffusivity of fiber bundle cannot be along the fiber direction is 4.44 times the diffusivity in
taken as zero. The glass fiber filaments themselves can the transverse directions, indicating the anisotropic dif-
be considered to have zero or nearly zero permeability. fusion behaviors. The second approach is using a
However, considering that the fibers are arranged in micromechanics model to capture the anisotropic diffu-
bundles and the fibers are surrounded by a sizing (inter- sivity in FRP composites. Fiber bundle and matrix are
face layer), diffusion channels may open up between taken as two constituents within the micromechanics
fiber filaments and in the sizing. It is convenient to model. The anisotropic diffusion behavior is captured
model this as diffusion within the fibers, but on a by considering nonzero anisotropic diffusion
Gagani et al. 13

coefficients for the fiber bundle and isotropic diffusion composites – an overview. Mech Time Dep Mater 2000;
for the matrix. The rationale behind choosing nonzero 4: 107–126.
diffusivity for the fiber bundle is occurrence of water 7. Springer G. Environmental effects on composite mater-
seeping within fiber filament in the bundles during dif- ials. Lancaster, PA: Technomic Publishing Company,
fusion process. 1984.
8. Springer G. Environmental effects on composite mater-
ials. Lancaster, PA: CRC Press: 1984.
Highlights 9. Weitsman YJ. Fluid effects in polymers and polymeric
composites. Springer: US, 2012.
. Micromechanical model for anisotropic diffusion in 10. Jacquemin F, Fréour S and Guillén R. Prediction of local
composites. hygroscopic stresses for composite structures – Analytical
. Testing procedure for the determination of anisotropic dif- and numerical micro-mechanical approaches. Compos Sci
fusion constants. Technol 2009; 69: 17–21.
11. Standard Test Method for Moisture Absorption
. Prediction of moisture equilibrium content in the presence
Properties and Equilibrium Conditioning of Polymer
of voids.
Matrix Composite Materials. ASTM International;
2014. West Conshohocken, PA, ASTM International.
12. Crank J. The mathematics of diffusion. Oxford:
Declaration of Conflicting Interests
Clarendon Press, 1956.
The author(s) declared no potential conflicts of interest with 13. Fan Y, Gomez A, Ferraro S, et al. The effects of tem-
respect to the research, authorship, and/or publication of this peratures and volumetric expansion on the diffusion of
article. fluids through solid polymers. J Appl Polym Sci 2017;
134: 45151.
Funding 14. Shen CH and Springer G. Moisture absorption and
desorption of composite materials. J Compos Mater
The author(s) disclosed receipt of the following financial sup-
1976; 10: 2–20.
port for the research, authorship, and/or publication of this
15. Choi HS, Ahn KJ, Nam JD, et al. Hygroscopic aspects of
article: This work is part of the DNV GL led Joint Industry
epoxy/carbon fiber composite laminates in aircraft envir-
Project ‘‘Affordable Composites’’ with nine industrial part-
onments. Compos A Appl Sci Manuf 2001; 32: 709–720.
ners and the Norwegian University of Science and
16. Bao L-R and Yee AF. Moisture diffusion and hygrother-
Technology (NTNU). The authors would like to express
mal aging in bismaleimide matrix carbon fiber compos-
their thanks for the financial support by The Research
ites—part I: uni-weave composites. Compos Sci Technol
Council of Norway (Project 245606/E30 in the Petromaks 2
2002; 62: 2099–2110.
programme). The authors from Texas A&M University
17. Bao L-R and Yee AF. Moisture diffusion and hygrother-
would like to acknowledge the funding supports from the
mal aging in bismaleimide matrix carbon fiber compos-
National Science Foundation (CMMI-1266037) and Office
ites: part II—woven and hybrid composites. Compos Sci
of Naval Research (N00014-13-1-0604).
Technol 2002; 62: 2111–2119.
18. Arnold JC, Alston SM and Korkees F. An assessment of
References methods to determine the directional moisture diffusion
1. Davies, P. Rajapakse, Y. D.S. (eds.) Durability of com- coefficients of composite materials. Compos A Appl Sci
posites in a marine environment. Dordrecht: Springer Manuf 2013; 55: 120–128.
Netherlands, Dordrecht; 2014. 19. Beringhier M, Simar A, Gigliotti M, et al. Identification
2. Davies P. Environmental degradation of composites for of the orthotropic diffusion properties of RTM textile
marine structures: new materials and new applications. composites for aircraft applications. Compos Struct
Phil Trans A Math, Phys Eng Sci 2016; 374: 20150272. 2016; 137: 33–43.
3. Ochoa OO and Salama MM. Offshore composites: transi- 20. Blikstad M, Sjöblom POW and Johannesson TR. Long-
tion barriers to an enabling technology. Compos Sci term moisture absorption in graphite/epoxy angle-ply
Technol 2005; 65: 2588–2596. laminates. J Compos Mater 1984; 18: 32.
4. Mouritz AP. Introduction to aerospace materials. Reston, 21. Bond DA. Moisture diffusion in a fiber-reinforced com-
VA: American Institute of Aeronautics and Astronautics; posite: Part I - Non-Fickian transport and the effect of
Cambridge, UK: Woodhead Pub.; 2012. fiber spatial distribution. J Compos Mater 2005; 39:
5. Chamis CC and Singhal SN. Quantification of uncertain- 2113–2141.
ties of hot-wet composite long term behavior. In: 22. Rocha ICBM, Raijmaekers S, Nijssen RPL, et al.
Thornton EA (ed.) Aerospace thermal structures and Experimental/numerical study of anisotropic water diffu-
materials for a new era. Progress in Astronautics sion in glass/epoxy composites. In: 37th Risø inter-
and Aeronautics: American Institute of Aeronautics and national symposium on materials science; Risø,
Astronautics; 1995, pp.259-272. Denmark: IOP Publishing Ltd; 2016.
6. Weitsman YJ and Elahi M. Effects of fluids on the 23. Gagani A, Krauklis A and Echtermeyer AT. Anisotropic
deformation, strength and durability of polymeric fluid diffusion in carbon fiber reinforced composite rods:
14 Journal of Composite Materials 0(0)

experimental, analytical and numerical study. Manuscript h Thickness

submitted for publication. 2018. I3 identity matrix
24. Springer G, Tsai S and Springer G. Thermal conductiv- l length
ities of unidirectional materials (Thermal conductivities M moisture content
of unidirectional composite materials parallel and
M1 composite moisture equilibrium
normal to filaments, using analogy to shear loading
content
response). J Compos Mater 1967; 1: 166–173.
25. Whitney JM. Three-dimensional moisture diffusion in
MM1 matrix moisture equilibrium content
laminated composites. AIAA J 1977; 15: 1356–1358. MV1 void moisture equilibrium content
26. Loos AC and Springer GS. Moisture absorption of gra- Ne total number of finite elements used in
phite-epoxy composites immersed in liquids and in humid FE simulations
air. J Compos Mater 1979; 13: 131–147. t time
27. Humeau C, Davies P and Jacquemin F. Moisture diffu- vf fiber volume fraction
sion under hydrostatic pressure in composites. Mater vm matrix volume fraction
Design 2016; 96: 90–98. vv void volume fraction
28. HEXION. Technical Data Sheet. EPIKOTE Resin MGS V unit-cell total volume
RIMR 135 and EPIKURE Curing Agent MGS RIMH V(a) subcell volume
137 2006. wf fiber weight fraction
29. Standard Test Methods for Constituent Content of
wm matrix weight fraction
Composite Materials. West Conshohocken, PA: ASTM
wv void weight fraction
International: 2015.
30. Muliana AH and Kim JS. A two-scale homogenization W Width
framework for nonlinear effective thermal conductivity of X(a) subcell length in the flux direction
laminated composites. Acta Mechanica 2010; 212: a Subcell
319–347. air air density
31. Muliana AH. A micromechanical formulation for piezo-
e empty void’s density
electric fiber composites with nonlinear and viscoelastic
constituents. Acta Materialia 2010; 58: 3332–3344.
f fluid-filled void’s density
32. Hill R. Elastic properties of reinforced solids: some the- m matrix density
oretical aspects. J Mech Phys Solids 1963; 11: 357–372. water water density
’(a) subcell concentration gradient
’(a)i concentration gradient in direction i
Appendix through the subcell a
’i homogenized concentration gradient
Notation through the composite in direction i
B(a) subcell concentration matrix ’ effective composite concentration
C fluid concentration gradient
D effective composite diffusivity
D(f) fiber diffusivity
D(m) matrix diffusivity
D(a) subcell diffusivity Appendix 1
D|| parallel diffusivity This appendix discusses parametric studies in examin-
Dr transverse diffusivity ing the effect of anisotropic diffusion in the fiber bun-
DM matrix diffusivity dles on the overall diffusion behaviors in FRP
f flux through the homogenized composites. The micromechanics model is used. The
composite purpose of this parametric study is to guide the material
f(a) flux through the subcell parameter characterization, i.e. Dð11fÞ , Dð22fÞ of the fiber
f(a)i flux in direction i through the subcell a bundles. The diffusivity of the isotropic matrix was
fi homogenized flux through the compo- obtained from the experimental test on neat resin.
site in direction i
Gagani et al. 15

Parallel Parallel
1.2 1.2

1 1

Water uptake (%)

Water uptake (%)

0.8 0.8
Experiment
0.6 0.6 Experiment
D11_f = 5 Dm D22_f = 2 Dm
0.4 D11_f = 2 Dm 0.4 D22_f = Dm
D11_f = Dm D22_f=0.5 Dm
0.2 0.2
D22_f=0.2 Dm
0 0
0 100 200 300 400 500 0 100 200 300 400 500
time(h) time(h)

Figure 14. Diffusion response of FRP in the parallel fiber direction: the effect of fiber anisotropy.

Transverse Transverse
1.2 1.2

1 1

0.8 0.8
Water uptake (%)

Water uptake (%)

Experiment
0.6 0.6 Experiment
D11_f = 5 Dm
0.4 D11_f = 2 Dm 0.4 D22_f = 0.5 Dm
D11_f = Dm D22_f = 0.3 Dm
0.2 D11_f=0.5 Dm 0.2 D22_f = 0.2 Dm
0 0
0 100 200 300 400 500 0 100 200 300 400 500
time (h) time (h)

Figure 15. Diffusion response of FRP in the transverse fiber direction: the effect of fiber anisotropy.

Figure 14 illustrates the water uptake response in FRP the diffusion response along the transverse fiber direc-
parallel to the fiber axis. It is seen that the transverse tion. In this case, the transverse fiber diffusivity (Dð22fÞ )
diffusivity of the fiber bundle (Dð22fÞ ) has insignificant has a significant effect while the axial fiber diffusivity
influence on the diffusion response in the parallel fiber (Dð11fÞ ) has insignificant effect, which is expected. This
direction. The diffusion in the parallel direction is gov- information can guide the determination of the aniso-
erned by the axial diffusivity of the fiber, diffusion of tropic fiber diffusivity from the experimental data.
the matrix and fiber volume content. Figure 15 shows