Composite Structures 111 (2014) 179–192

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Multi-factorial models of a carbon fibre/epoxy composite subjected

to accelerated environmental ageing
Enrique Guzmán, Joël Cugnoni, Thomas Gmür ⇑
Faculté des Sciences et Techniques de l’Ingénieur (STI), Ecole Polytechnique Fédérale de Lausanne, Station 9, CH-1015 Lausanne, Switzerland

a r t i c l e i n f o a b s t r a c t

Article history: Among materials being introduced in the aerospace industry, the carbon fibre reinforced plastics (CFRP)
Available online 6 January 2014 have a place of privilege because of their exceptional stiffness-to-mass ratio. However, the polymer-
based matrix is vulnerable to damages by environmental conditions. This work exposes the experimental
Keywords: results of several accelerated environmental ageing protocols on CFRP panels. The main concern is to jus-
Carbon fibre epoxy composite tify or reject by statistical means that a significant degradation of mechanical properties does occur over
Accelerated ageing protocol the time, and to establish a basic model to quantify the effects of different environmental factors of the
Aero-structures
composite ageing. The results considered here are the elastic properties evaluated over several weeks of
Moisture absorption
Design of experiments
accelerated artificial ageing. The stiffness degradation of the samples subjected to the aforementioned
Structural health monitoring ageing protocols is statistically described by a non-linear multi-factorial model inspired by the Design
of Experiments (DoE) theory. The evolution of constitutive properties (namely mass and elastic proper-
ties) over the time exhibits an asymptotic exponential increasing (or decreasing) pattern over the time.
The usefulness of these mathematical models is their predictability, based only on theoretical consider-
ations on moisture absorption. This path is further investigated in this paper, clearing up the way to a
methodical prediction of ageing models.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction such as heat, moisture or solar radiation, when applied cyclically

Composite materials are ultra-light structural materials, mas-
sively introduced in the recent years in aeronautical applications.
In spite of their exceptional mechanical performances, they are
vulnerable to aggressive natural ageing factors such as rough tem-
perature changes, chemical corrosion, moisture and solar radiation.
It is widely accepted that special caution needs to be observed
when using these materials to manufacture mechanically critical
airframe components inside a full-scale structure. Although an on-
line health monitoring is strongly suggested by authors [1–3] to
verify the state of a composite structure, a model could be useful
to estimate the degradation extent when subjected to some fre-
quently confronted weathering agents. Carbon fibre reinforced
plastics (CFRP) are currently the most used composites in the aero-
nautical industry (for example in the Boeing 787 and the currently
in development Airbus 350 XWB).
There is currently in scientific literature some lack of under-
standing of polymer-based composite materials behaviour to
weathering. Research has been mainly centred about the effects
of mechanical fatigue [4] and chemical corrosion [5,6]. A quantita-
tive model of the material ageing under other usual natural agents
⇑ Corresponding author. Tel.: +41 21 693 2924; fax: +41 21 693 7340.
E-mail address: thomas.gmuer@epfl.ch (T. Gmür).
on a sample, has not yet been established. This is frequently due
to the elevated number of factors that can potentially affect the life
cycle of polymers. Considering the case of CFRP [7–10], the poly-
mer-based matrix and the fibre/polymer interface are the most
vulnerable components of the composite. Specialised researchers
in the field of composite ageing have described qualitatively the
failure mechanisms [6,9,10] or presented empirical quantitative
evidence of changes in constitutive properties of polymers
[7,8,11,12] when subjected to natural and/or artificial weathering.
In more extensive treatises by Carraher [13] and Brinson [14],
some chemical mechanisms that explain the degradation are
reviewed. These changes lead to a progressive macroscopic degra-
dation of the elastic properties of the composite, which could turn
out to be critical if the proper safety precautions are not considered.
A method to evaluate systematically the ageing of polymer-
based composites and establish a mathematical model is proposed,
which can be interpreted to understand the contributions of iso-
lated or combined factors on the ageing process of aerospace com-
posite. It will be verified if it is possible to estimate quantitatively
the extent of the ageing by estimating the parameters of the math-
ematical model, based only on theoretical assumptions and basic
information about the subject material. According to the specia-
lised literature in the design of experiments theory [15,16], a mul-
ti-factorial model is useful to compare quantitatively the influence

0263-8223/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruct.2013.12.028
180 E. Guzmán et al. / Composite Structures 111 (2014) 179–192
of the existent ageing factors on the composite panels. The elastic
properties, which are the object of this study, are measured by
indirect methods, using modal testing to obtain the natural fre-
quencies of the working specimens [17], which in turn are
processed by a mixed numerical–experimental identification algo-
rithm to obtain the experimental elastic properties [18]. Physically,
the modal testing is done using mainly integrated piezoelectric
sensors and accelerometers, which can be inscribed in a global
structural health monitoring (SHM) method, which in summary
allows surveying the state of a structure using networks of
dynamic sensors. The choice of a dynamic measurement method
is motivated by its flexibility, robustness, readiness and
non-destructiveness when compared to more classic static
methods (tensile tests, bending tests, ultrasound), qualities that
are appreciated in aeronautical applications.
The mathematical pattern of the aforementioned model can be
hinted by visual inspection of the experimental pre-modelling
plots after the identification of the constitutive properties. The no-
tion of Prony series and asymptotically exponential increase or de-
crease are inspired from previous works on composite testing
[11,13,14]. A deeper statistical analysis shows a correlation be-
tween the absorbed moisture mass and the loss of stiffness. This
statement leads to the replacement of the number of cycles by
the water concentration as the state variable in the model. The use-
fulness of this reasoning is evident in the final lines of this paper: it
leads to a generalised model for mechanical parts with more
complex shapes, and contributes to the future research on CFRP
with an early estimation of the extent of their ageing using only
some basic information about the material’s initial properties.
2. Accelerated ageing protocol due to environmental conditions
In order to reveal any changes in stiffness due to exposition to
aggressive ageing factors, progressively demanding experimental
campaigns took place. The ageing factors included temperature,
relative humidity (RH) and ultraviolet (UV) radiation, since they
are frequently met by full-scale CFRP aeronautical structures dur-
ing a life cycle. The ageing protocols were inspired from previous
works [7–10] as well as in ASTM standard guidelines for cyclic age-
ing protocols [19], for the combined hygro-thermal testing [20–22]
and for UV radiation testing [23,24]. These factors, alone or com-
bined, have several effects on polymer-based pieces inside long-
time operating aircraft: water diffusion, polymer molecules
cross-linking/de-linking, alternate dilatation/contraction cycles,
photo-oxidation, post-curing, residual stress, temperature gradi-
ents, and many others. In summary, a suitable ageing protocol
would have the following characteristics:
a. Cyclic conditions are more likely to induce damage than
steady conditions. Indeed, at a microscopic level, cyclic ther-
mal gradients and water concentration gradients continu-
ously contribute to the rise of mechanical stress around
the fibres and between the composite layers. Cyclic changes
induce thus additional damage due to the mechanical fati-
gue of the components. Moreover, this is far closer to reality
since aerospace components are constantly exposed to cyclic
environments.
b. Temperature peaks are usually fixed above the start point of
the glass transition zone. In this paper, this range of tempera-
tures was estimated from the supplier’s data sheet and after
the corresponding curing process (curing rate is higher than
99%). Physically, dilatation, vitreous transition and thermal
oxidation are likely to occur at elevated temperature.
c. Highly humid environments can contribute to composite age-
ing as well, facilitating fibre de-bonding, de-lamination,
embrittlement, polymer chemical weakening, inner stressing,
etc. The mass absorption is to be verified by weighing period-
ically the total mass of the samples, in order to evaluate the
diffusion coefficient.
d. High UV radiation is generally more specific to structures
continuously dwelling at high altitudes, leading to photo-
oxidation and polymer chain dissociation, among others.
Consequently, the specimens were subjected to cyclic environ-
ments, with the following factors controlled: the surrounding
temperature (T), the relative humidity (RH) and the intensity of
an A-class ultraviolet radiation (UV) lamp on one face of the plate
samples (Fig. 1). The cycles are longer for the humid protocols due
to technical reasons. The scheduled campaigns and the series, with
the respective codes and samples included, the protocol parame-
ters and the durations are summarised in Table 1. Hardware in-
cluded a Weiss TechnikÒ WK180/40 climatic chamber, with the
required control and data acquisition software.
Normalised variables are worked with: for a real factor u (that
can be either T; RH or UV), the corresponding normalised factor x
is given by
u  umin
x¼ ð1Þ
umax  umin
where umin and umax are respectively the minimum and maximum
value adopted by u. Thus, the value of x is always between 0 and
1, inclusively. The normalised temperature, relative humidity and
UV radiation are denoted x1 ; x2 and x3 respectively. These are shown
along with the complementary time parameters for each ageing
protocol in Table 1.
After an estimation based on data provided by the supplier and
in the literature, the thermal and the water diffusion coefficients,
the thermal equilibrium due to convective transfer inside the cli-
mate chamber is reached after between 1 and 2 min in the range
of temperature 25–125 °C, so the saturation is easily reached. On
the other hand, the moisture saturation is not reached in one cycle,
in the case of 4 mm-thick plate samples. In fact, it is not reached
before 200 h in steady-state conditions. This is expected, because
of the tests carried out prior to the cyclic ageing tests (as detailed
in Section 6.1). However, the water concentration is accumulative,
since most of the water absorbed in each cycle remains inside the
structure during the dry phase of the cycle. At the end of the pro-
tocol, there is a significant amount of water inside the bulk body of
the sample (as shown in Section 6). Concerning the penetration of
UV radiation, the same problem can be observed. After the infor-
mation available in the literature, the equilibrium of radiant heat
on plates by UV radiation (340 nm wavelength, 0.35 W/m2/nm
intensity, as recommended by the G155 ASTM standard, corre-
sponding to a direct sunlight beam at sea level and 0° latitude) is
estimated to take about 20 min. This time period is less than any
of the periodic exposure times faced in the protocols (the shortest
is 1800 s = 30 min).
3. Sample manufacturing
3.1. CFRP specimens
The material for experimentation is the Carbon-PrePreg PR-UD
CST 125/300 FT109, supplied by Suter-KunstoffeÓ AG (Switzer-
land). It is originally a scroll of unidirectional (UD) carbon fibre
tissue (ToraycaÓ T700S carbon fibre), pre-impregnated in unhard-
ened epoxy polymer (PREDOÓ FT109) with an areal weight of
125 g/m2 (60% of fibre volume fraction). The nominal after-curing
elastic properties of this material are given in Table 2. A total of
six different 30  30 cm2 surface plates were manufactured in
autoclave (curing at 85 °C and under 5 bar for 10 h, followed by
curing at 90 °C and under 5 bar for 4 h). Each one of these square
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192 181

Table 1
Summary of ageing protocol parameters for all the series: nominal values (T; RH and UV), normalised values (x1 ; x2 and x3 ), the number of cycles N cyc , the time period
Dt ¼ Dtþ þ Dt  and the number of measurements q.

Series Samples Temperature Relative humidity UV radiation N cyc Dt (min) q Dt  a

Dt þ
2
T (°C) x1 RH (%) x2 UV (W/m ) x3
Min Max Min Max Min Max
D DI-DVI 5 95 0 0 95 1 0 300 1 800 90 8 2:1
EF1 EI, FI 45 135 1 0 70 0.73 0 300 1 900 90 9 1:1
EF2 EII, FII 45 135 1 0 70 0.73 0 0 0 900 90 9 1:1
G1 GI,GIII 5 95 0 0 0 0 0 0 0 1100 60 12 1:1
G2 GII 5 95 0 0 0 0 0 300 1 1000 60 11 1:1
H1 HI,HIII 45 135 1 0 0 0 0 300 1 1100 60 12 1:1
H2 HII 45 135 1 0 0 0 0 0 0 1100 60 10 1:1
J JI-JV 5 95 0 0 95 1 0 0 0 800 90 8 1:1
a
Dt þ is the part of the cycle at high temperature and Dt is the part at low temperature.

Table 2
Nominal elastic properties after layup and curing (1: longitudinal direction, 2: transverse direction, 3: normal direction). Over 63 measurements, l is average value and r is the
corresponding standard deviation.

Young’s modulus (GPa) Poisson’s ratio (–) Shear modulus (GPa)

E1 E2 E3 m12 m13 m23 G12 G13 G23
l 96.00 7.67 8.70 0.38 0.3 0.03 3.60 3.59 2.24
r 0:067 0:013 –a –a –a –a 0:021 –a –a
a
Nominal values as supplied in the data sheet.

Table 3
3.2. Sensors and integration
Specimen dimensions and initial mass.

Sample Length l Width w Thickness h Mass m In SHM, dynamic tests are preferred since they are non-destruc-
(mm) (mm) (mm) (g)
tive and easily applicable. A combination of accelerometers and
DI-II-III-IV-V-VI 150 99.0 4.0 83:49  0:43 polyvinylidene fluoride (PVDF) integrated film sensors has been
EI-II 300 98.5 4.4 188:5  4:5
used for obtaining the dynamic signals. The minimum number of
FI-II 300 81.6 4.0 134:0  2:0
GI-II-III 300 92.0 3.7 153:0  3:0 modes to be observed is determined using a FE model of the sample,
HI-II-III 300 95.4 4.2 179:0  2:0 as described in Section 4. While accelerometers are fixed and
JI-II-III 300 98.2 4.3 181:0  1:5 removed between two measurements, PVDF sensors are glued to
or embedded inside the composite plate, implying that they are part
of the ageing process as well (as in Fig. 2). It has been demonstrated
plates was divided, giving six different groups of specimens, named that PVDF sensors are able to survive the ageing process without
D–J and scheduled to be tested following the ageing protocols seeing their dynamic measurement capabilities affected [25].
described in Section 2. The layup configuration was purely
UD, with the carbon fibres in the longitudinal direction. The 4. Monitoring tools and methods
dimensions and corresponding nominal masses of the plates were
measured using an electronic Vernier and Mettler ToledoÓ weigh 4.1. Non-destructive test: modal analysis
balances respectively (a 0.01 g resolution for the D-series, a 0.1 g
resolution for all the others, see Table 3). There is a wide range of solutions to measure directly or indi-
rectly the elastic properties of a structure. In the interest of SHM,
it is imperative to restrict the choice to non-destructive, fast, ro-
bust techniques. Modal analysis is a simple but powerful tool to
obtain information about the state of a structure, based on the
measurement of the natural frequencies and the corresponding
modes. Among the available technologies, it was found that piezo-
electric transducers provide remarkable quality signals, suitable
for dynamic analyses [26–28]. For this reason, an adequate modal
testing bench was chosen for experimentation. Hardware includes
a Hewlett Packard HP 35670 Dynamic Signal Analyser, which is
capable to directly compute and display the FRFs from time wave-
forms supplied by sensors. The complete measurement chain hard-
ware is shown in Fig. 3. It is important to remember that an
additional parameter to deduce the sample stiffness from its natu-
Fig. 1. Example of ageing protocol profile: the temperature T (in blue), the relative ral frequencies is the mass, which would potentially change in case
humidity RH (in green) and UV radiation intensity UV (in red). The mean values are of moisture absorption/evaporation between the environment and
in dashed line. Temperature and relative humidity profiles exhibit finite slopes in
the transient phases, while UV radiation can be turned on/off instantaneously. (For
the structure. As the expected change in the mass is not negligible,
interpretation of the references to colour in this figure legend, the reader is referred it is necessary to monitor it carefully in order to avoid systematic
to the web version of this article.) measurement errors. The modal extraction software ME-ScopeÓ,
182 E. Guzmán et al. / Composite Structures 111 (2014) 179–192
Fig. 2. Schemes of sensor mounting: (a) glued PVDF sensor, (b) embedded PVDF
sensor and (c) accelerometer with wax.
developed by Vibrant TechnologiesÒ, is used to find the modal
parameters from the supplied FRFs.
4.2. Finite element model and structural identification algorithm
A finite element (FE) model provides accurate results when the
dynamic behaviour of real plates is simulated. This approach was
used to build a mixed numerical–experimental identification algo-
rithm, based upon a non-linear least squares method combined to
ABAQUSÓ FE simulations in order to minimise the error between
estimated and experimental frequencies, and to find the most suit-
able values for the elastic properties.
A structured mesh with C3D20R (in ABAQUSÓ nomenclature)
quadratic hexahedral elements of 2 mm was used, in order to have
Fig. 3. Equipment used for the modal analysis: (1) composite plate, (2) acceler-
ometer, (3) PVDF integrated sensor, (4) impact hammer, (5) charge amplifier B&K
2635 and (6) HP 35670 Dynamic Signal Analyser.
at least 2 elements along the thickness (5 nodes). The stiffness ma-
trix is computed using the reduced integration technique. The UD
plates were modelled assuming a homogeneous orthotropic elas-
ticity model in which the parameters E1 (longitudinal Young’s
modulus), E2 (transversal Young’s modulus) and G12 (transversal
shear modulus) are to be identified, while the remaining parame-
ters have been assumed as constant. Thus, it is not necessary to in-
clude in the variable vector all of the nine independent engineering
constants that determine the elastic behaviour of an orthotropic
material. A simple sensitivity study based on a FE model links
the lowest frequency modes to the elastic parameters having a sig-
nificant influence on them. A sensitivity parameter is defined as:
Df =f
S¼ ð2Þ
DE=E
S is a normalised derivative of the frequency f with respect to the
elastic parameter E. For example, in Fig. 4, it can be seen that the
basic vibration modes (flexural and torsional modes) for a D-series
specimen depend essentially on two tensile moduli (E1 and E2 ) and
one shear modulus (G12 ).
An algorithm for the minimisation of an objective function
based on the discrepancy between the estimated and experimental
natural frequencies has been developed. This algorithm is based on
the one proposed by [18], which uses a Levenberg–Marquardt opti-
misation strategy. The initial guesses for E1 ; E2 and G12 are esti-
mated by the values provided in CFRP prepreg supplier’s data
sheet, and shown in Table 2.
The descending direction is obtained by the finite difference
method on the FE model. Only positivity of the elastic parameters
is a constraint for this optimisation problem.
5. Results: setup of a multi-factorial statistical model
5.1. Introduction to experimental results
For the sake of uniformity and clarity, the changes in the consti-
tutive properties of the samples are tracked with respect to their
initial value:
m E1 E2 G12
l¼ ; 1 ¼ ; 2 ¼ ; c12 ¼ ð3Þ
m0 E10 E20 G120
Fig. 4. Example of sensitivity analysis between engineering constants and natural
frequencies for a D-series sample. The latter are mainly influenced by the
longitudinal and transverse tensile moduli E1 and E2 , and the transverse shear
modulus G12 .
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192 183

The 0 index denotes the nominal initial values of the elastic Y X b

y¼ ¼1þ ai /i ðx1 ; x2 ; x3 Þð1  en=10 i Þ ð4Þ
properties, summarised in Table 2. At first glance, the graphs sug- Y0 i
gest an exponential asymptotic behaviour (Fig. 5). An increase in
the mass can be observed in the ‘‘humid’’ protocols (D-, EF1-, /0 ¼ ð1  x1 Þð1  x2 Þð1  x3 Þ
EF2- and J-series). As stated before, in a UD configuration, the influ-
/1 ¼ x1 ð1  x2 Þð1  x3 Þ
ence of the polymer matrix and the fibres on the elastic properties
are partially decoupled. Among the identified properties, the longi- /2 ¼ ð1  x1 Þx2 ð1  x3 Þ
tudinal tensile modulus E1 is mostly associated to the carbon /3 ¼ ð1  x1 Þð1  x2 Þx3
fibres, while E2 and G12 are related to the epoxy matrix. On one /12 ¼ x1 x2 ð1  x3 Þ
hand, the value of 1 exhibits a rather more irregular variation /13 ¼ x1 ð1  x2 Þx3
around a steady value. Globally, it was observed a slight increase
/23 ¼ ð1  x1 Þx2 x3
in the longitudinal tensile modulus, at least during the first phase
of the ageing process, although not statistically representative (er- /123 ¼ x1 x2 x3
ror bars represent the standard deviation). On the other hand, an In Eq. (4), each function /i ðx1 ; x2 ; x3 Þ represents a series of sam-
opposite evolution of 2 and c12 can be observed in Fig. 5, and ples under a given set of environmental conditions x1 ; x2 and x3
shows a decrease of stiffness. However, the extent of the loss de- (see Table 1), Y represents either m; E1 ; E2 or G12 , and y represents
pends on the ageing protocol parameters. In all cases, the loss of l; 1 ; 2 or c12 . Physically, the identified constant coefficients ai
transverse bending and shear stiffness is much more significant are the final asymptotic values of the loss/gain, and 10bi are the
that the scattering. corresponding time constants (in number of cycles). In other
words, this expression is basically a model expressing an exponen-
5.2. Mathematical model inspired from the Prony series tial decay (or rise) of the constitutive properties of the samples
subjected to an accelerated ageing. Given the strong non-linearity
From the preliminary observations in the previous section, a of the model, the result of the identification process can be sensi-
multi-factorial mathematical model following the pattern of a tive to the initial estimations. The Levenberg–Marquardt least-
Prony series is proposed. The mathematical expression is given square optimisation algorithm was used to find the best fitting
by the following equation

Fig. 5. Example of evolution of relative mass l, Young ’s moduli 1 ; 2 and shear modulus c12 for (a) D-, (b) EF1-, (c) EF2- and (d) H1-series.
184 E. Guzmán et al. / Composite Structures 111 (2014) 179–192

parameters. The initial conditions were chosen on a basis of multi-

ple Monte Carlo simulations, in order to find a global minimum for
the least-square fitting error.
The fitting results are shown in Fig. 6, with the corresponding
residuals and coefficients of determination R2 . As it can be seen,
the fitting quality is very high for l; 2 and c12 , while it is consid-
erably lower in the case of 1 . As it is shown in the following
sections, 1 (which represents the stiffness along the carbon fibres
direction) shows a behaviour that cannot be statistically
determined.

5.3. Validation of the mathematical model

The validation of a model has been carried out by graphical and

analytical means. For the former, a simple way to do so is by estab-
lishing a scatter plot displaying a graphic comparison between the
estimated curve and the experimental results of aged samples as
shown in Figs. 7 and 8. Visually, it can be observed that the exper-
imental results are globally well predicted by the model, especially
concerning the ageing of the polymer matrix (tensile modulus 2
and shear transverse modulus c12 ).
For the analytical validation, a lack-of-fit test is also performed
on the global data, in order to show the goodness of fit of the
mathematical model, as in the supplement document about the
Lack-of-fit (LoF) test, of which the test hypothesis can be formulated
as follows:

H0 : The model does not induce any significant error:
H1 : The model induces a significant error:
In other words, if H0 is accepted, the model pattern (in this case
given by Eq. (4)) is appropriate for the measured dataset. If H0 is
rejected, the systematically induced error is inherent to the chosen
model pattern.
The parameters of the dataset for the hypothesis test are:

1. A total of N ¼ 193 measurements (one measurement per series

over 100 cycles).
2. The averaged measurements (one averaged measurement per
series over 100 cycles) were Q ¼ 79.
3. The model in Eq. (4) has an order p ¼ 16.
4. m1 ¼ Q  p ¼ 63 degrees of freedom (DoF) for the ‘‘lack-of-fit’’
error.
5. m2 ¼ N  Q ¼ 114 degrees of freedom (DoF) for the experimental
measurement error.

The quotient k between the squared lack-of-fit normalised error

and the squared experimental normalised error serves as observed
value for a Fischer’s distribution law Fðm1 ; m2 Þ. For such a distribu-
tion and a significance level of a ¼ 1%, the critical value of lambda
is kcrit ¼ F 0:01 ð63; 114Þ ¼ 0:583. If k is above this value, the null
hypothesis H0 is rejected with a fail probability of 1%. It can be seen
in Table 4 that H0 is accepted in all cases except for the observed k
corresponding to 1 . This confirms the statement made about the
R2 values for the dependent variables: for l; 2 and c12 , which
exhibit high values for the coefficient of determination, it can thus
be concluded that the models are accurate enough to represent the
ageing of T700S/FT109 specimens, in the frame of the given condi-
tions, while the model is not appropriate to predict 1 .

5.4. Interpretation of the validation

A statement after inspecting Fig. 5 is that there is indeed a tem-

poral evolution of the elastic properties and the mass of the sam- Fig. 6. Bar diagrams for identified model coefficients ai for relative (a) mass l, (b)
Young’s modulus 1 , (c) Young’s modulus 2 and (d) shear modulus c12 .
ples. It is clear that over a long-term weathering, there is a limit
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192
Fig. 7. Scatter plot for measurements of D-series samples: (a)
for the degradation of properties. The exhibited trend of 1 is not
very clear. On the contrary, l shows a clear increase, and 2 and
c12 show a clear decrease.
Physically, 1 is associated to the stiffness of the carbon fibres in
a UD composite. The carbon fibres are rather chemically and ther-
mally inertial, thus the observed increase of 1 would not be due to
a change in the inner structure of the carbon. The changes in 2 and
c12 are less surprising since they were expected, as stated in the re-
view of the state-of-the-art in accelerated ageing. The loss of stiff-
ness by the polymer matrix is a combination of several phenomena
detailed in the literature, in particular due to polymer chain de-
linking, photo-oxidation, water corrosion, etc.
From the observations made on the scatter plots, the determi-
nation coefficient R2 and the corresponding Fischer quotients F in
the previous section, it is safe to state that the evolution of l; 2
and c12 can be explained by such a mathematical model. On the
other hand, the model is not an accurate predictor for 1 .
6. Equivalent moisture absorption in cyclic ageing protocols
Unfortunately, mathematical models based on statistical treat-
ment remain uninteresting from the experimental point of view.
Indeed, they cannot be directly applied to more general cases since
the geometry plays a determining role in several of the ageing
mechanisms (moisture absorption, heat diffusion, penetration
depth of UV radiation, etc.).
To remove this dependency, it is proposed to replace time by a
state variable, such as moisture concentration. To do so, it is neces-
sary to establish a link between theory and experiments in the
185
l, (b) 1 , (c) 2 and (d) c12 .
moisture absorption, for determination of ‘‘equivalent’’ cyclic con-
stants of diffusivity and saturation mass.
6.1. A model for water diffusion in isothermal conditions
Water absorption is possible only in the presence of some humid-
ity in the surrounding atmosphere (RH – 0%). Thus, to simplify the
calculation, the following hypothesis is going to be adopted: any
change of mass in a ‘‘dry’’ protocol (RH ¼ 0%) is neglected.
Concerning the water absorption, a simple mathematical rea-
soning can be adopted in order to estimate the water concentration
absorbed by a sample when subjected to accelerated ageing condi-
tions. This amount is generally not equal between periodic or iso-
thermal conditions. The mathematical approach to model the
water absorption comes from Fick’s law of diffusion in combina-
tion with the mass conservation principle, as expressed in the fol-
lowing equation:
 
@c @ @c
¼ D ð5Þ
@t @x @x
where c is the volume concentration, D the diffusion coefficient, t
the time coordinate and x the space coordinate along the thickness.
Under the hypothesis of absence of swelling, the volume concentra-
tion c is proportional to the mass concentration M, defined by
M ¼ l  1. This case can be assimilated to an infinite plate with
thickness h with two exposed boundary conditions (since l  h
and w  h). The solution of Eq. (5) in such conditions is obtained
by application of the Fourier’s series theory, giving the following
result
186 E. Guzmán et al. / Composite Structures 111 (2014) 179–192

Fig. 8. Scatter plot for measurements of J-series samples: (a) l, (b) 1 , (c)2 and (d)c12 .

Table 4
Summary of the lack-of-fit test. To accept the null hypothesis, k 6 kcrit ¼ 0:583. ES
Ms ¼ M s0 eRT ð10Þ
Model Mass Elastic moduli
l 1 2 c12 ES
lnðM s Þ ¼ lnðM s0 Þ  ð11Þ
R 2 0.997 0.853 0.966 0.981 RT
k 0.047 9.938 0.515 0.215
Decision H0 H1 H0 H0
where M s0 is the saturation mass when T ! 1 and ES the corre-
sponding diffusion activation energy. Concerning the influence of
the relative humidity RH on the saturation mass, it has been pro-
! posed by some authors [12,29,30] that a power law model can be
2
MðtÞ 8 X
1
1 ð2k þ 1Þ p2 Dt
¼1 2 exp  ð6Þ used to determine the relation between both
Ms p k¼0 ð2k þ 1Þ2 h
2

Ms ¼ M smax ðRHÞf ð12Þ

where Ms is the saturation mass. However, this expression is com-
plicated, and a simpler function is used to approximate it where f is an exponent depending on the temperature and the
! absorbing composite and Msmax the maximum saturation mass pos-
 0:75  
MðtÞ t t sible. After [29], in the case of an epoxy matrix, this value is 1 in the
 1  exp 1:31 ¼H ð7Þ
Ms s s current temperature range, so this hypothesis is adopted in this
paper.
2
where s ¼ h =ðp2 DÞ is a time constant for this specific exponential The thermodynamic constants D0 ; ED ; Ms ; M s0 defined in Eqs.
growth and H symbolises a function. Diffusivity depends on tem- (8)–(11) can be estimated experimentally by linear regression, if
perature following an Arrhenius-type law: datasets of measured (T; M s ) and (T; D) are available (see Fig. 9).
ED In a previous experimental campaign [31], T700S/FT109 samples
D ¼ D0 e RT ð8Þ were subjected to ageing protocols in steady-state isothermal con-
ditions with constant RH  95%, at 60 °C, 80 °C and 100 °C, with the
ED
lnðDÞ ¼ lnðD0 Þ  ð9Þ results summarised in Table 5. From this, and supported by infor-
RT mation in literature [12,30], the associated diffusion activation
where D0 is the diffusion coefficient when T ! 1; ED the diffusion energy ED ¼ 9:067 kJ/mol, and the saturation enthalpy
activation energy and R the universal gas constant. Similarly, the ES ¼ 1:68 kJ/mol can be deduced in a range of temperature be-
saturation mass Ms depends exponentially on the inverse of the tween 5 °C and 135 °C. With these two linear models, it is possible
temperature by Eq. (10) to estimate quite accurately D and the M s in steady conditions.
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192 187

6.2. Diffusion for periodic ageing protocols Table 5

Computed diffusion coefficients in preliminary tests with test series A and B [31].

The behaviour under periodic protocols is different from that Temperature Diffusion Saturation mass
under constant conditions. Despite this, the base hypothesis is that, T (°C) T (K) 2
D (mm /s) Ms (%)
after a ‘‘transient’’ phase, the amount of absorbed water oscillates
60 333 2.21e7 8.84
around a local moving mean value, which is a fraction of M s at stea- 80 353 2.67e7 9.02
dy conditions. For the first charge and discharge, we have 100 373 3.14e7 9.44
 
M 01 Dt 
¼H ð13Þ where dt n is the time it would take to have a concentration of Mn1
Ms s
in isothermal conditions, and it is defined by Eq. (18):
   
M1 Dt þ dt n M n1
¼ 1  H ð14Þ ¼ H1 ð18Þ
M 01 sþ s Ms
2 2
where s ¼ h =ðp2 D Þ and sþ ¼ h =ðp2 Dþ Þ; Dt  and Dt þ are the where H1 is the inverse function of H and Mn1 is the mass left
corresponding durations of the low temperature and the high tem- after the ðn  1Þth discharge. The physical meaning of dt is shown
perature part of the cycle (including the transient state), and D and in Fig. 10.
Dþ are the corresponding diffusion coefficients. Combining Eqs. (13) A mathematical sequence can then be defined to determine the
and (14) gives absorbed mass after the nth cycle:
           
M1 Dt þ Dt  Mn M n1 Dt þ Dt  þ dt n
¼ 1H H ð15Þ ¼f ¼ 1H H
Ms sþ s Ms Ms sþ s
     
Similarly, for the nth charge and discharge, we have Dt þ Dt  M n1
¼ 1H H þ H1 ð19Þ
  sþ s Ms
M 0n Dt þ dtn
¼H ð16Þ
Ms s where f denotes a recurrence relation. The mass concentration M
can be determined recursively after the nth cycle as a function of
 
Mn Dt þ the concentration after the ðn  1Þth cycle. In Fig. 10, the amplitude
¼ 1  H ð17Þ of the oscillation is exaggerated in order to show the concentrations
M 0n sþ
and time variables. From a simulation after Eq. (19), it can actually
be deduced that the absorbed moisture mass depends essentially on
the ratio Dt þ =sþ and Dt  =s . The final value limn!1 Mn ¼ Ms;eq
(where eq stands for equivalent) of the mass absorption is given
by the solution of x ¼ f ðxÞ. After using MATLAB to solve this recur-
sive equation, the results for the ageing protocols are summarised
in Table 6. The ratio Ms;eq =Ms is asymptotic, representing the frac-
tion of water absorbed by the specimen under a cyclic protocol
Ms;eq with respect to the absorbed mass under a steady protocol Ms .
Concerning the results, it is interesting to see in Table 6 that
independently from the thickness, the equivalent saturation con-
centrations M s;eq are similar between the EF1-, EF2- and J-series.
The equivalent saturation mass for D is on the other hand much
higher. The difference comes mostly from the difference in the cyc-
lic timing of the protocols, since the cycles in the D-series have a
2:1 ratio between Dt =Dt þ (instead of 1:1 for EF1-, EF2- and J-
series).

Fig. 9. Graphs of linear regressions (1=T; ln M s ) and (1=T; ln D) as in Eqs. (9) and (11) Fig. 10. Cyclic mass absorption. The variables are those described in Eqs. (13)–(19)
for measured values of A- and B-series. The slopes give respectively the values of (a) in Section 6.2: the saturation mass in constant conditions is M s , while M1 ¼ M s;eq is
ED and (b) ES . the equivalent saturation mass in cyclic conditions.
188 E. Guzmán et al. / Composite Structures 111 (2014) 179–192

6.3. Equivalent diffusivity and equivalent saturation mass Table 7

Comparison between theoretical and experimental estimations of the equivalent
diffusivity.
The idea is now to compare the results from the two models for
moisture absorption inside a structure: an experimental empiric Series Fitted model Deq;exp Theoretical Deq;th (from Rel.
model as defined in Section 5.2, and a theoretical analytic recurrent (cyclic test) isothermal data) error %

model as in Section 6.2. To do so, the ‘‘equivalent’’ diffusivity Deq , D 1.27e12 1.54e12 21.5
for cyclic conditions, needs to be defined. EF1 4.45e13 5.08e13 14.2
EF2 4.61e13 4.93e13 7.1
Let us consider the Prony series approximation of the diffusion
0:75 J 4.44e13 5.39e13 21.5
function HðxÞ ¼ 1  e1:311x  1  e1:3181x , which has the same
exponential form as the terms of the ageing mathematical model,
as it is described in Eq. (4). A link between the theoretical diffusion relative humidity, the evolution of the mass depends essentially
and the statistical experimental model can then be established. For on the existence/absence of moisture in the ageing environment.
the ith protocol, we have In the graphs, the ageing track of all of the 8 series can be distin-
b guished. More particularly, it shows that the pattern of the stiffness
1  e1:3181t=si ¼ 1  en=10 i ð20Þ evolution over the time is similar between ageing protocols that
where t is the time (in [s]), n is the corresponding number of cycles, exhibit the same relative humidity profile (closeness between the
si is the diffusion time constant (defined in Eq. (7)) and 10bi is the J- and D-, EF1- and EF2-, G1- and G2-, and H1- and H2-series),
time constant in number of cycles of the ith term in the mathemat- although the intensity of the loss grows deeper when UV radiation
ical model (4). For a single cycle, n ¼ 1 and t ¼ Dt. It can easily be and high temperatures are applied. However, the main result may
deduced the following equation: be the visible correlation between two a priori independent vari-
ables: the water concentration and the elastic moduli. In Fig. 12
si ¼ 1:3181Dt  10bi ð21Þ and Table 8, linear regression models for the series are graphically
After the definition of s, the equivalent diffusivity Deq for the ith represented (by straight lines) and numerically computed (by the
protocol can be defined as: corresponding slopes).

2
h 7.2. Correlation test
Deqi ¼
p2 si
There are now at our disposal two ways to determine the value
of Deq : a theoretical value Deq;th obtained from simulated results
and experimental Deq;exp obtained from the cyclic experimental
results:
2
h a Student’s t-distribution parameter h is carried out to find out if
Deq;exp ¼
13:01Dt  10bi
All taken into account, the theoretical prediction of the satura-
tion mass and diffusivity, and their experimentally determined
counter-parts are quite close. A fair prediction of absorption can
be done only with theoretical considerations and isothermal diffu-
sion data as shown in Tables 6 and 7 and in Fig. 11. The advantage
of these modelling procedure is that the ageing problem can finally
be seen freed from the influence of geometry, and then applicable
to a much wider range of composite parts.
7. Relationship between moisture concentration and loss of
stiffness
7.1. Linear regression between mass and elastic properties
In Fig. 12, a pattern of correlation between the mass variation
and the elastic properties can be observed. Depending on the
ð22Þ
The corresponding Pearson’s correlation coefficients q for each
series is given in Table 8. If the absolute value jqj is close to one,
the linear regression is appropriate (for further detail, see the sup-
plementary document Statistical correlation between water concen-
tration and loss of stiffness). A hypothesis test based on the value of
ð23Þ the correlation, for each series and each elastic variable, is statisti-
cally significant, and can be formulated as follows

H0 : There is no correlation between variables:
H1 : There is a correlation between variables:
qffiffiffiffiffiffiffiffi
q2
The observed parameter is defined as h ¼ jqj 1 q2 , following a
Student’s t-distribution, with q  2 DoF, where q is the number of
observations per series. The null hypothesis is accepted if
h > hcrit ¼ t a=2 ðq  2Þ. In Table 8, it can be seen that the null hypoth-
esis is constantly rejected for 2 and c12 , while it is almost
constantly accepted for 1 . This reinforces our past statements:
the modelling of 2 and c12 is possible, while 1 ’s is statistically
incorrect. To sum up, this shows that the ageing by water absorp-
tion and the decrease of stiffness inside the polymer matrix are
correlated. The details of the computations are in the supplemen-
tary document. Since, by definition, the fitted straight lines (see
Fig. 12) start always at (1, 1), the regression formula takes invari-
ably the form described in Eq. (24):

Table 6
Estimated mass increase for the four series subjected to humid protocols. M s;eq is the equivalent saturation mass for the cyclic protocol and M s the corresponding saturation mass
under isothermal conditions.

Series Cycles N Temperature (K) Time period (s) Thickness (mm) Mass ratio Equivalent sat. mass (cyclic) M s;eq Saturation mass (isothermal) M s
a b c
T Tþ Dt  Dt þ h Ms;eq =M s – – – –a –b –d

D 800 278 368 3600 1800 4.04 0.135 0.0106 0.0100 0.0101 0.079 0.074 0.075
EF1 900 318 408 2700 2700 4.23 0.110 0.0059 0.0066 0.0060 0.054 0.060 0.055
EF2 900 318 408 2700 2700 4.02 0.112 0.0064 0.0067 0.0064 0.058 0.060 0.057
J 800 278 368 2700 2700 4.29 0.124 0.0063 0.0066 0.0061 0.051 0.054 0.049
a
Predicted from the fitted model.
b
Predicted from the theory.
c
Mean residual mass, obtained by drying of the samples.
d
Predicted from the experimental result.
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192
Fig. 11. Graphic comparison between the theoretical and fitted models, for (a) D-, (b) EF1-, (c) EF2- and (d) J-samples.
M ¼ l  1 ¼ Ai ðyi  1Þ
where i ¼ 1; 2; 3, namely yi stands for either 1 ; 2 or c12 and Ai
denotes the slope.
7.3. A linear model for Ai
There is then a linear relationship between the mass absorption
and the ageing of the polymer matrix, represented by a decay of 2
and c12 . Since the time variable is included in l, the proportionality
coefficient Ai should depend only on the weathering conditions x1
(temperature), x2 (relative humidity) and x3 (UV radiation). A linear
model is suggested in the following equation:
Aðx1 ; x2 ; x3 Þ ¼ a0 þ a1 x1 þ a2 x2 þ a3 x3 þ a12 x1 x2 þ a13 x1 x3
þ a23 x2 x3
where Aðx1 ; x2 ; x3 Þ represents any of the coefficients Ai in Eq. (24).
The coefficients Ai identified by linear regression are summa-
rised in Table 9, along with the corresponding Pearson’s q. Under
the hypothesis that the coefficients of the linear regression follow
a normal law, an analysis of variance (ANOVA) can be then carried
out, in order to check the significance of the different factors in the
ageing of the composite. This multi-factorial analysis of variance
can be performed as described by [15,16], and is detailed in the
supplementary document about Statistical correlation between
water concentration and loss of stiffness. As it can be seen in Table 9,
the null hypothesis (the factor is statistically significant) is
189
ð24Þ accepted for a0 ; a1 ; a2 and a12 (in the case of c12 ). Physically, this
can be interpreted as follows:
1. Since a0 – 0, there is a natural ageing in ‘‘normal conditions’’
(no humidity nor UV radiation, temperature is cyclic but always
under the glass transition threshold).
2. The polymer matrix is sensitive to temperature (a1 ) and humid-
ity (a2 ), and eventually a combination of both (a12 ).
3. The part of UV radiation is close to negligible, at least with the
intensity of the lamps in the ageing processes analysed in this
paper.
Conclusively, since ageing and water absorption are correlated,
Eq. (24) shows that the time (or the number of cycles) can be re-
placed by the water concentration as an ageing variable. The time
ð25Þ
variable is then implicitly included in the water diffusion model,
which can be predicted theoretically from the models in Section
6.3.
On the other hand, the Ai coefficients depend only on the mate-
rial and environmental conditions of the ageing protocol (T; RH and
UV) and can be estimated by the linear model in Eq. (25) with the
parameters ai in Table 9.
8. A predicting method for composite ageing
From a scientific point of view, the multi-factorial model
presented in this paper can prove to be useful in two ways: for
190 E. Guzmán et al. / Composite Structures 111 (2014) 179–192

Fig. 12. Correlation plots between the relative mass l and (a) 1 , (b) 2 and (c) c12 .
 E. Guzmán et al. / Composite Structures 111 (2014) 179–192 191

Table 8 9. Conclusions
Results of hypothesis testing on the correlation between l and the elastic properties.
If H0 is accepted, there is no correlation, while if H1 is accepted, there is a correlation.
The hereby presented macroscopic model, based on the Prony
Series q Ai Decision series, proposed a characterisation method for a set of simulta-
1 2 c12 1 2 c12 neous physical/chemical phenomena that intervene on the CFRP
J 8 0.48 0.14 0.13 H0 H1 H1 ageing. The parameters of that model, fitted on the basis of the
D 8 0.62 0.18 0.16 H1 H1 H1 NLLS criterium, were determined from experimental results. How-
EFI 9 0.20 0.06 0.06 H1 H1 H1 ever, the application of these results was restricted to samples with
EFII 9 0.31 0.08 0.07 H0 H1 H1
the same geometrical dimensions and proportions.
GI 12 0.13 0.02 0.02 H0 H1 H1
GII 11 0.12 0.03 0.03 H0 H1 H1 Thus, the main contribution of this document is not only to sug-
HI 12 0.03 0.01 0.01 H0 H1 H1 gest the form of a mathematical model, but also to set this model
HII 10 0.03 0.01 0.01 H0 H1 H1 free from the specific dimensions of the physical samples that were
experimented with. In order to do so, the first step was to study the
link between the theoretical and the experimental results of water
diffusion and mass absorption. Such analysis gave as a result
Table 9
Summary table of data yi ¼ Ai ðx1 ; x2 ; x3 Þ used to set up a linear model with
‘‘equivalent’’ diffusivities and saturation masses. The second step
interactions, along with the corresponding correlation coefficients q. proved by statistical means that there is a correlation between
the absorbed moisture (a state variable) and the ageing in cyclic
ANOVA Effects p-Value p 6 10%
conditions. In this case, the model has gotten free from the dimen-
2 c12 2 (%) c12 (%) 2 c12 sions influence. In other words, based only on the constitutive
a0 0.0305 0.0287 0.51 0.11 Yes Yes properties of a composite material and the features of a cyclic age-
a1 0.0294 0.0252 8.58 3.10 Yes Yes ing protocol, the ageing can be estimated. This method can be used
a2 0.1754 0.1663 0.14 0.03 Yes Yes
in the future to predict the material ageing in more complex CFRP
a3 0.0127 0.0049 69.67 62.72 No No
a12 0.0822 0.0804 12.16 3.33 No Yes demonstrators.
a13 0.0280 0.0132 42.88 50.35 No No
a23 0.0200 0.0168 64.20 52.40 No No
Acknowledgements

The authors would like to acknowledge the partial financial

characterisation and design of experiments, for the experimenter
to follow a similar method to characterise other composite materi-
als using different combinations of fibre/resin, and for prediction,
using this model to estimate the ageing of a composite structure
with variable geometries, since the effect is known.
For characterisation/design of experiments: Based on all that
has been exposed in this paper, a simple method for establishing a
statistical ageing model for a given material is proposed:
support from the Swiss National Science Foundation, Grant No.
200020-143968/1.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.compstruct.2013.
12.028.

1. By isothermal tests or from providers datasheet, establish the D References

and M s Arrhenius’ laws, as in Eqs. (8) and (10).
2. Determine the parameters of the ageing protocol: time duration
of the cycle (Dt  ; Dtþ ; Dt) and the corresponding temperature
and relative humidity levels, in order to determine Ms ; D and
Dþ by using the model described in Section 6.2.
3. Build a water absorption model using the recursive Eq. (19),
determining by least squares and Eq. (7) the equivalent diffusiv-
ity Deq and equivalent saturation mass concentration M s;eq .
4. Design and carry out an experimental campaign, using 2k sam-
ple series (where k is the number of factors to be considered; in
this paper, k ¼ 3).
5. The parameters a of the simplified ageing model (25) for the
elastic moduli of the composite can be then determined by a
linear regression, while the pertinence of the model can be sta-
tistically assessed by an ANOVA.
Prediction: If the ageing model for a material is available, the
experimenter can use it to predict the ageing of the analysed
structure:
1. Simulate the evolution of water concentration l using an ana-
lytical or numerical model that takes into account the geometry
of the part, since the equivalent diffusion parameters (Deq and
Ms;eq ) are known.
2. Using the linear model (24), determine the level of ageing for a
given water concentration state.
[1] Giurgiutiu V. Structural health monitoring with piezoelectric wafer active
sensors. 1st ed. Elsevier; 2008.
[2] Sohn H, Farrar C, Hemez F, Shunk D, Stinemates D, Nadler B, et al. A review of
structural health monitoring literature: 1996–2001. Tech. Rep. LA-13976-MS,
Los Alamos National Laboratories; 2004.
[3] Lynch J, Loh K. Summary review of wireless sensors and sensor networks for
structural health monitoring. Shock Vib Digest 2006;38(2):91–128.
[4] Pupurs A, Varna J. Modeling mechanical fatigue of UD composite: multiple
fiber breaks and debond growth. IOP C Ser Mater Sci Eng 2009;5(1):12–7.
[5] Liao K, Schultheisz C, Hunston D, Brinson L. Long-term durability of fiber-
reinforced polymer matrix composite materials for infrastructure applications:
a review. J Advan Mater 1998;30(4):3–40.
[6] White J. Polymer ageing: physics, chemistry or engineering? Time to reflect. CR
Chim 2006;9(11–12):1396–408.
[7] Mouzakis D, Zoga H, Galiotis C. Accelerated environmental ageing study of
polyester/glass fiber reinforced composites (GFRPCs). Compos Part B – Eng
2008;39(3):467–75.
[8] Dao B, Hodgkin J, Krstina J, Mardel J, Tian W. Accelerated ageing versus realistic
ageing in aerospace composite materials – IV. Hot/wet ageing effects in a low
temperature cure epoxy composite. J Appl Polymer Sci 2007;106(6):4264–76.
[9] Fox B, Lowe A, Hodgkin J. Investigation of failure mechanisms in aged
aerospace composites. Eng Fail Anal 2004;11(2):235–41.
[10] Bondzic S, Hodgkin J, Krstina J, Mardel J. Chemistry of thermal ageing in
aerospace epoxy composites. J Appl Polym Sci 2006;100(3):2210–9.
[11] Tong M, Singhal S, Chamis C, Murthy P. Simulation of fatigue behavior of high
temperature metal matrix composites. Tech. Rep. NASA-CR-204605, NASA
Glenn Research Center; 1996.
[12] Lin Y, Chen X. Investigation of moisture diffusion in epoxy system:
experiments and molecular dynamics simulations. Chem Phys Lett 2005;
412:322–6.
[13] Carraher C. Polymer chemistry. New York, NY: CRC Press Inc.; 2010.
[14] Brinson H, Brinson L. Polymer engineering science and viscoelasticity: an
introduction. 1st ed. New York, NY: Springer; 2008.
192 E. Guzmán et al. / Composite Structures 111 (2014) 179–192
[15] Box G, Hunter W, Hunter J. Statistics for experimenters, design, innovation and
discovery. 2nd ed. Wyley, NY: Wiley Series in Probability and Mathematical
Statistics; 2005.
[16] Montgomery D. Design and analysis of experiments. 7th ed. Wyley, NY: Wiley
Series in Probability and Mathematical Statistics; 2009.
[17] Ewins D. Modal testing: theory, practice and application. 2nd
ed. Hertfordshire: Research Studies Press Ltd.; 2000.
[18] Cugnoni J, Gmür T, Schorderet A. Inverse method based on modal analysis for
characterizing the constitutive properties of thick composite plates. Comput
Struct 2007;85(17–18):1310–20.
[19] ASTM International. West Conshohocken, PA, ASTM D4762 standard guide for
testing polymer matrix composite materials; 2011.
[20] ASTM International. West Conshohocken, PA, ASTM C581 standard
practice for determining chemical resistance of thermosetting resins
used in glass-fiber-reinforced structures intended for liquid service;
2003(2008).
[21] ASTM International. West Conshohocken, PA, ASTM D1151 standard practice
for effect of moisture and temperature on adhesive bonds ASTM international;
2000(2006).
[22] ASTM International. West Conshohocken, PA, ASTM D5229 standard test
method for moisture absorption properties and equilibrium conditioning of
polymer matrix composite materials; 1992(2010).
[23] ASTM International. West Conshohocken, PA, ASTM G151 standard practice for
exposing nonmetallic materials in accelerated test devices that use laboratory
light sources; 2010.
[24] ASTM International. West Conshohocken, PA, ASTM G154 standard practice for
operating fluorescent light apparatus for UV exposure of nonmetallic
materials; 2006.
[25] Guzman E, Cugnoni J, Gmür T, Bonhôte P, Schorderet A. Survivability of
integrated PVDF film sensors to accelerated ageing conditions in aeronautical/
aerospace structures. J Smart Mater Struct 2013;22(6):065020.
[26] Park G, Farrar C, di Scalea FL, Coccia S. Performance assessment and validation
of piezoelectric active-sensors in structural health monitoring. J Smart Mater
Struct 2006;15(6):1673–83.
[27] Giurgiutiu V, Zagrai A, Bao J. Embedded active sensors for in-situ structural
health monitoring of thin-wall structures. J Press Vess Technol 2002;124(3):
293–303.
[28] Lin B, Giurgiutiu V. Review of the in situ fabrication methods of piezoelectric
wafer active sensor for sensing and actuation applications. In: Tomizuka M,
editor. Smart structure materials: sensors and smart structure technology for
civil, mechanical, and aerospace systems (Proceedings of SPIE 2005), vol. 5765.
Bellingham, WA; 2005. p. 1033–44.
[29] Weitsman Y. Fluid effects in polymers and polymeric composites. London:
Springer; 2012.
[30] Crank J. The mathematics of diffusion. 2nd ed. Oxford: Oxford Science
Publications; 1989.
[31] Guzman E, Cugnoni J, Gmür T. Accelerated isothermal and cyclic ageing of
carbon fibre/epoxy composite panels with integrated PVDF sensors. In: 15th
International conference on experimental mechanics (ICEM15) book of
abstracts; 2013. p. 335–6.