Mechanical Systems and Signal Processing 99 (2018) 534549

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Model of moisture absorption by adhesive joint

Veronica Bonilla Mora a, Magdalena Mieloszyk b,, Wieslaw Ostachowicz b,c
a
Gdansk University of Technology, Narutowicza 11/12 Str., 80-233 Gdansk, Poland
b
Institute of Fluid Flow Machinery, Polish Academy of Sciences, Fiszera 14 Str., 80-231 Gdansk, Poland
c
Warsaw University of Technology, Narbutta 84 Str., 02-524 Warsaw, Poland

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

Article history: Adhesive joints offer many advantages over traditional mechanical joining systems.
Received 13 February 2017 Nonetheless, their use is limited since they can be adversely affected by extreme temper-
Received in revised form 23 June 2017 atures and humidity conditions. Moisture contamination (even 13% of the sample weight)
Accepted 28 June 2017
in an adhesive can alter its tensile strength and compromise the structural integrity of the
joint. Moisture absorption processes can be monitored using methods based on fibre Bragg
grating sensors embedded in the adhesive material.
Keywords:
In the present paper, a finite element model of an adhesive joint between composite ele-
Moisture absorption
Hygroscopic strain
ments was analysed using the commercial code AbaqusTM. The investigation contains two
Mass diffusion main parts: a thermal analysis and a hygro-mechanical analysis. The achieved results were
Finite element model verified using experimental investigation results for a sample with embedded fibre Bragg
Hygro-mechanical analysis grating sensors that were applied to monitor the moisture-induced strains in the adhesive
Fibre Bragg grating joint.
Composite The achieved numerical results show good agreement with the experimental ones for all
considered analyses. The presented models can also be used for the determination of mois-
ture content in an adhesive layer especially in a range of 1.52.5% of the water content.
2017 Elsevier Ltd. All rights reserved.

1. Introduction

Adhesive joints offer many advantages over traditional fastening methods, such as a more homogeneous stress distribu-
tion, aesthetic appeal, higher stiffness, high fatigue strength, low weight, the possibility to join dissimilar materials and cor-
rosion prevention [1]. They are also preferable to traditional methods like mechanical fasteners that may require the drilling
of the component. Drilling holes can be a source of damage for the material and they introduce undesired stress concentra-
tions near the affected area [2]. Nonetheless, the use of adhesive joints is still limited since many of its properties are neg-
atively affected by environmental conditions.
One of the most important causes for loss of mechanical strength in an adhesive joint and composite materials is moisture
absorption [39]. By virtue of its polymeric nature, the ingress of water in an epoxy is associated with an increased separa-
tion between the molecular chains which causes expansional strains. This phenomenon is referred to as plasticization and it
can alter the chemical structure of the component [10]. Moisture has important effects on composite performance because it
causes degradation, especially in the polymeric matrix of composites [11]. Variations of moisture concentration results in
changes in the dynamic characteristics of composite elements [12]. For the majority of composite structures, the increase
of hygroscopic concentrations results in the decrease of mechanical properties of the material. For hygrothermal effect, it

Corresponding author.
E-mail address: mmieloszyk@imp.gda.pl (M. Mieloszyk).

http://dx.doi.org/10.1016/j.ymssp.2017.06.042
0888-3270/ 2017 Elsevier Ltd. All rights reserved.
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 535

is noted that the variation of vibration frequency is caused by both the hygrothermal deformation and the change of material
properties [13].
Moisture concentration up to 1% has fairly small effects on the dynamic characteristics of composite elements although it
is noticeable [12]. Due to this, different models [14,12,15,13,16] have been developed to analyse the influence of moisture,
temperature, fibre orientation and loading conditions on structure dynamic properties of composite laminates.
Moisture absorption by an adhesive leads to several changes in its physical and mechanical structure. It reduces the glass
transition temperature T g , tensile strength and lowers the ultimate elongation of the adhesive [8]. Thus, moisture absorption
and moisture-induced strain monitoring is an area of high interest in the field of structural health monitoring (SHM).
The most important degradation processes due to moisture contamination (even 13% of water mass gain) in adhesive
layers are: plasticisation [1719] cracking [18], hygroscopic expansion of the adhesive [17,19] and deterioration of the
adherend/adhesive interface [1719]. The moisture distribution inside an epoxy component is necessary to fully assess
the adhesive joints mechanical response under known environmental conditions. The diffusion characteristics of moisture
in an epoxy adhesive are critical factors to predict the moisture profile [8].
One of the experimental methods for determining the amount of moisture in an epoxy is based on FBG (fibre Bragg grat-
ing) sensors embedded into structural element during the manufacturing process [3,20,21]. FBG sensors written on silica
fibre optics have many advantages: small size and weight, high corrosion resistance (both water and chemicals) and no cal-
ibration requirements [22]. The FBG sensor wavelength is linearly dependent on strain [23] and temperature [24]. Since the
process of moisture absorption results in an increase of material volume reflected by internal strain changes, it can be mea-
sured by FBG sensors embedded into the analysed material layer [3,20,21]. Numerical simulations can be used to predict the
mass gained by water intake and the mechanisms of diffusion through polymeric materials [25,26].
In the paper, several finite element models were created to simulate the thermal response, diffusion dynamics and strain
development of the tested components and serve as a methodology for future studies in the area. The numerical analysis
results were compared with experimental results performed on composite samples with adhesive layers with embedded
FBG sensors. The experimental investigation was described in detail in a previous paper [20].

2. Adhesive joint - moisture contamination detection

The numerical analysis was performed on a sample composed by two elements of GFRP (Glass Fiber Reinforced Plastic)
composites with a stacking sequence of (0/90/0/90/90/0/90/0) joined together by an adhesive layer with a thickness of
0.2 mm as it is presented in Fig. 1. The adhesive used was a two-part structural epoxy paste adhesive produced by Henkel
Corporation and commercially known as Loctite EA 9394 Aero or Hysol EA 9394.
In the adhesive layer two FBG sensors were embedded parallel to main axis of the sample to measure volumetric change
of the sample induced by moisture influence. The first one was located near the samples center while the second one - more
closely to the edge of the sample (see Fig. 1).
The soaking process was performed under stable temperature conditions (333.15 K 2 K). The sample was kept inside a
box filled with water, up to 0.5 mm on top of the sample, for 2 weeks (336 h) (Fig. 1).
As it is visible in Fig. 2, the maximum amount of moisture that can be absorbed by a GFRP sample during 14 days of soak-
ing is less than 1%. As the adhesive layer absorb more water than composite the numerical analysis was concerned on this
part of the adhesive joint. However the water diffusion throughout composite elements was taken into consideration in mass
diffusion analysis. The relative water gain M(t) was determined using the following relationship

wk t  wref
Mt 1
wref

where wref is the samples weight when dry and wk t its weight at time t.

Fig. 1. Sample dimensions with two FBG sensors location in adhesive layer and measurement set-up.
536 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Fig. 2. Comparison of water content [%] for the adhesive and GFRP composite experimental results.

The experimental data was recorded as a percentage of weight gain, Mt. However, the numerical analyses are run using
average concentration, Ct, as the main variable. To be able to compare the simulation results with the experimental data,
the weight-gain data was converted to average concentration. The average concentration was calculated through the rela-
tionship [3,21].

qepoxy
Ct Mt 2
qwater

where qepoxy 0:00136 g/mm3 [27] and qwater 0:0009832 g/mm3 [28] are the densities of the epoxy and water at 333 K,
respectively. The experimental results for average concentration are shown in Fig. 3(a). The concentration at saturation is
3:126% (2:13% of weight gain). Fig. 3(b) shows the moisture-induced strains measured using FBG sensors.
Material properties of the adhesive and composite used in the numerical analyses are given in Table 1. It was assumed
that the adhesive behaves isotropically. Density and thermal conductivity of the adhesive were given by the manufacturer
[27]. Its specific heat was taken from Lai et al. [3] in which a similar epoxy was employed. The coefficient of hygroscopic
swelling was determined by experimental data, since it is given by the slope when plotting strains values against concen-
tration as shown in Fig. 3(b). The GFRP composites were considered as homogeneous, isotropic materials with properties
assumed similar to those stated by Keller et al. [29] and Bai et al. [30].
Viscoelastic properties of the adhesive were not taken into consideration in the simulation analyses. The tensile strain
values achieved during the experimental investigation [20] were small enough to claim that the material is in its elastic
range according to the stress-strain curve for Hysol EA-9394 adhesive [31].

Fig. 3. Experimental data: (a) water content, (b) strain induced by moisture measured using FBG sensor embedded into adhesive layer.
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 537

Table 1
Material parameters of the adhesive layer.

Material Parameter Value Unit

Density 0.00136 g/mm3
Thermal conductivity 0.000331 W/mm K
Specific heat 1.00 J/g K
Adhesive Youngs modulus 5.49e16 (gmm/s2)/mm2
Poissons ratio 0.4
Hygroscopic swelling coeff. 0.004030
Density 0.00187 g/mm3
Thermal conductivity 0.00035 W/mm K
Composite Specific heat 1.17 J/g K
Youngs modulus 4.17e17 (gmm/s2)/mm2
Poissons ratio 0.3

2.1. Numerical model

A finite element model was created using the commercially available finite element method (FEM) software AbaqusTM. The
numerical simulation comprises two main parts: a thermal analysis and a hygro-mechanical analysis. The hygro-mechanical
analysis contains two steps: a mass diffusion analysis and a mechanical analysis.
The sample contained an adhesive layer and two GFPR composite skins. The analyses were focused on the adhesive layer
where the FBG sensors were embedded. However, since the composites had an active role in the heat transfer process and
the interaction between them and the adhesive was important for the mechanical analysis, they were also taken into account
on the 3D-models of the sample. The FEM model details are presented in Fig. 4. Due to the symmetric properties (both geo-
metric and loading-wise) of the samples, to simplify the model and the computational effort, only one fourth of the total size
of the sample (Fig. 4(a)) was taken into consideration. During the simulation the adhesive was divided into three elements
through its thickness and the sensors (sensing point) were located in the center - see Fig. 4(c).
Sets of elements were created for the areas where the FBG sensors were located (Fig. 4(b)). The FBG sensors dimensions
(125 lm) are assumed to be small enough in comparison with the adhesive dimensions to be neglected and the material
parameters of silica glass were not included in the numerical models.
During this numerical investigation a comparison of results for two different global size meshes (1 mm and 0.5 mm) was
performed see Fig. 4(d) and (e).

Fig. 4. FEM model: (a) the sample scheme with marked modeled area, (b) top view of the sample with highlighted FBG sensors location, (c) cross-section of
the adhesive layer with marked FBG sensors location, (d) 1 mm size mesh, (e) 0.5 mm size mesh.
538 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

3. Thermal analysis

The main goal of the thermal analysis was to determine the time it took the sample to reach an isothermal state, both
when in dry conditions as well as when submerged.
The convection boundary conditions were imposed in the model through a surface film condition interaction only on the
top surfaces and external faces. The bottom was considered as insulated since it is in direct contact with the floor of the box,
and the remaining faces are omitted because of symmetrical properties. A tie condition was also used between the adhesive
and the composites as it was assumed that there is no thermal resistance between both materials. The tie condition equates
temperatures at the matching nodes.
The model was meshed using elements with an approximate global size mesh of 1 mm, and a constraint was placed to
have three elements across the thickness of the adhesive. During the thermal analysis 20-node quadratic heat transfer bricks
(DC3D20) elements were applied. Overall, a total of 15 625 elements and 93 788 nodes were employed.
The sample was first at a room temperature of 304 K 0.5 K and then placed inside the temperature chamber which was
kept at 333 K 2 K. During the experimental investigation, the sample was in dry conditions inside the chamber, and another
set of measurements were performed in wet conditions.
Thus, the mathematical model for the analysis involves heat convection between water or air and the corresponding sam-
ple, and heat conduction within the sample.
Two simulations were performed for the sample. One when the sample is introduced in the temperature chamber in dry
conditions and the second when the sample is submerged in water. In all cases, there is heat transfer through natural con-
vection with laminar flow. The corresponding convective heat transfer coefficients for air and water were calculated using
the correlation equation for natural convection on a flat plate [32,33]
!1=4
hL q2 g c bDTL3 lcP
Nu C Gr  Pr n 0:54Ra1=4 0:54  3
k l2 k

where Nu, Gr, Pr and Ra stand for Nusselt, Grashof, Prandtl and Rayleigh numbers respectively.
The meaning of each parameter and its corresponding value is listed in Table 2. Properties for both water and air were
 
taken at the average temperature value, T T f T p =2, from [28,34]. The characteristic length for a horizontal plate is
defined as L surface area=perimeter [35].
The heat convection coefficients were calculated using Eq. (3) and were equal to 4.93 W/Km2 and 819 W/Km2 for air and
water, respectively. These values are within the expected ranges for natural convection in gases (5 30 W/Km2) and in
liquids (20 1000 W/Km2) [33].

4. Hygro-mechanical analysis

The hygro-mechanical analysis contains two steps: a mass diffusion analysis and a mechanical analysis.

4.1. Mass diffusion and hygroscopic strains

The absorption of water by a polymer causes a swelling in the material known as hygroscopic swelling and the deforma-
tion of the material is referred to as hygroscopic strains. It has been found that hygroscopic strains, ehygro , follow a linear rela-
tionship with the change in concentration in the material DC. The behaviour is in analogy to how thermal strains, ethermal ,
linearly depend on the temperature difference, DT. Both dependencies are mathematically shown in the equations below
[36]
ehygro b  DC 4

Table 2
Parameters of air and water for Eq. (3).

Symbol Parameter Air Water Unit

q Density 1.109e0 9.901e2 kg/m3
cp Specific heat constant pressure 1.007e3 4.181e3 J/(kg K)
k Thermal conductivity 2.699e2 6.370e1 W/m K
l Dynamic viscosity 1.941e5 5.96e4 Pa s
b Coefficient of thermal expansion 3.200e3 4.150e4 1/K
g Gravitational acceleration 9.81 m/s2
L Characteristic length 10.42 m
Tf Temperature of fluid 333.25 K
Tp Temperature of plate 304.15 K
DT Temperature difference 29.1 K
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 539

ethermal a  DT 5
The symbols b and a stand for the coefficients of hygroscopic swelling and thermal expansion, respectively. Both can be
determined experimentally.
Ficks laws are the basis for understanding diffusion processes in different media. They accurately help quantify the con-
centration to be expected in many materials. However, it has been noted in multiple studies [10,3740] that the majority of
polymers show a non-Fickian nature, often named as anomalous diffusion. The most widely accepted diffusion models to
reproduce the water absorbed in polymers can be divided into two groups: multiphase diffusion and time dependent
diffusion.
Non-Fickian behaviour is observed in polymers when their temperature is less than the glass transition temperature (T g ),
as it is in this case. This anomalous behaviour could be the consequence of the relaxation process and/or chemical interac-
tions between the water molecules and the polymer chains [41]. Under the glass transition temperature, many properties of
polymers are time-dependent, e.g. the stress may be slow to decay after the polymer has been stretched [42]. Anomalous
diffusion occurs when the diffusion and relaxation rates are comparable, i.e. the polymer chains do not adjust as quickly
to the presence of the penetrant.
A diffusion analysis with a finite element model was created to simulate the water ingress inside the epoxy joint of the
samples used in the experiment. Two different time dependent models (the Time-Varying Boundary Conditions and the
Dual-Stage model) for non-Fickian diffusion were considered and compared. The progression of the average concentration
across the sample with time was calculated and compared to experimental results. The results for the distribution of the
water concentration Ct was then used to quantify the hygro-mechanical behaviour of the analysed sample.

4.2. Time-Varying Boundary Conditions

The viscoelastic nature of polymers can affect the diffusion process by increasing the concentration of moisture at the
exposed surfaces of the material with time [39]. In order to take into account this viscoelastic effect, Weitsman and Cai
[43] introduced a model with time-varying boundary conditions using Prony series described by the equation
X
C 0 t Ai Br 1  expt=sr  6
r

where t means time, Ai and Br are Prony coefficients, and sr is the rth time constant controlling how the concentration is
allowed to vary at the boundaries [39]. The variable C 0 stand for the concentration at the boundary.
b and M.
For the general case with time-varying conditions, the solution is given by a linear combination of C b Cb and M
b are
solutions of a one-dimensional case of Ficks second law for uniform diffusivity with an initial concentration, Cx; 0 0, and
boundary conditions equal to Eq. (6) with C i 0 and one exponential term.
It can be present as equation [43]

C x; t C i C H x; t X R
Cr b
C x; t; br 7
C1 C1 C
r1 1

M t Ci b
M t; br 8
M1 C1
where Mt, is a weight gain.
The variables C 0 ; C 1 and M 1 stand for the concentration at the boundary, concentration at saturation and total amount of
solute at saturation, respectively.
The material properties considered for the analyses are given on Table 3. The density is given by the manufacturer [27].
The diffusion coefficient was determined by conducting a linear fit on the first four points of the experimental data (Fig. 5)
and using equation [44]
 2  2
2h M2  M1
Dp p p 9
4M 1 t2  t1

where h is thickness of the analysed material, M 1 ; t 1 and M 2 ; t 2 are data for two points of the considered line.

Table 3
Material parameters for mass diffusion analysis.

Parameter Time-varing BC Dual stage Init

1st stage 2nd stage
Diffusion 0.00018325 0.00021023 0.00022764 mm2/h
Solubility 1.00 1.00 1.00
540 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Fig. 5. Determination of dual-stage using experimental data.

Solubility is defined as the concentration at saturation, which is the maximum amount of a substance that can be dis-
solved per the amount of a solvent [5]. AbaqusTM uses solubility to calculate the normalized concentration / equation [45]

C
/ 10
s
where C means concentration of the diffusing material and s - solubility in the base material. All boundary conditions must
be given as normalized concentrations as well. This means that if a surface is in full contact with the penetrant a boundary
condition equal to one should be applied. In the model, due to the nature of the boundary conditions used, it was preferable
to choose a solubility of 1 and boundary concentrations as concentration at saturation when necessary.

4.3. Dual stage model

The dual-stage model is characterized by diffusion profiles that appear to have two levels of saturation concentration.
Some studies have been able to model this behaviour by the use of two parallel Fickian processes, others have used two
sequential Fickian processes instead. For their study, Shirangi & Michel [46] compared both approaches and found that
the two sequential Fickian processes adapted better to experimental data.
The main parameters of the dual-stage model are: diffusion coefficients for the first and second stage, D1 D2 , concentra-
tions of saturation at the first and second stage, M11 M 12 and the time at which the change is made. All of this information
can be obtained from the experimental data directly and by using Eq. (9) for the diffusivity coefficients.
The diffusivity coefficient, D, and the concentration at saturation, C sat , are important parameters to characterize diffusion
in a material. Respectively, they determine the rate of diffusion and the absorption capacity [6]. Both can be obtained by
experimental data. In the case of the diffusivity coefficient, it is given by the initial slope when plotting the weight gain,
Mt, versus the square root of the time (Eq. (9)) [44].
At first glance, the weight-gain data for the sample (Fig. 3) might not give the impression of a dual stage diffusion. How-
ever, by taking a closer look, it is possible to distinguish two stages with different saturation points. Both of these stages are
pointed out on Fig. 5, where the first three points of the experimental data were omitted.
The mass saturation concentrations for each stage are 0.0211 and 0.0226, respectively. The corresponding diffusion coef-
ficients were calculated by Eq. (9), doing a linear fit on each of the two slopes. It is important to note that for the second
stage, the concentration used in the equation is 0.0015, the difference between both saturation concentrations. As for the
solubility, it was preferred to keep it as unity and account for the different saturation concentrations through the boundary
conditions.

5. FE models and results

5.1. Thermal analysis

The temperature distribution for dry and wet conditions for the sample are presented in Figs. 6 and 7, respectively. A
comparison of temperature measured by FBG sensor and determined numerically in its location for the same conditions
are shown in Fig. 8.
The obtained results showed that, when in dry conditions, the sample is able to reach thermal equilibrium during the first
hour (Fig. 6). The temperature profile for the sample was obtained by monitoring the temperature of a node during the
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 541

Fig. 6. Temperature distribution on the sample under dry condition.

simulation. This node was located in the area where the FBG sensors were placed in the sample (Fig. 4). This is in a good
agreement with experimental results (Fig. 8(a)). The higher curve at the beginning of the experimental results is due to
the fact that the temperature chamber needs to go to a higher temperature than the input setting before it can stabilize
at the correct temperature. When submerged, the sample reaches an isothermal state within the first 5 min (Fig. 7). These
results confirm the experimental assumption that during the hygroscopic strain measurements there were no temperature
changes. Higher differences are visible for the sample in wet conditions (Fig. 8(b)). This could be due to the fact that the
water temperature was lower than 333 K when placing the sample into the box with water. The oscillations visible on
the experimental values, once the system has stabilised, are because of the temperature oscillations ( 2 K) in the heating
chamber.

5.2. Mass diffusion analysis

The main focus of the analysis was the evolution of the moisture distribution within the adhesive with time. The type of
elements chosen were first-order brick elements (type DC3D8).
Fig. 9 shows the average concentration obtained by each anomalous diffusion approach, compared to the experimental
data. In both cases, it is clear that the use of a 1 mm mesh is sufficient enough to get a reliable concentration profile.
The numerical data for 1 mm mesh density were compared with experimental data and an analytical solution based on
multiphase diffusion - Langmuir model. The results are presented in Fig. 10.
The Langmuir diffusion model assumes that water can diffuse into the material, but some water molecules are also
trapped inside the epoxy microstructure [25]. As for Fickian model, the solution can be approximated by [39]
( "  0:75 #)
Mt f Dt f
expct 1  exp 7:3 expft  expct 1  expft 11
s fc 4h
2 fc

where c; f are constant referred to probability of water in each state. The calculated parameters were assumed as follow
D 1:5e  12m2 =s; c 2e  1; f 4:1e  2 [20].
As it is visible the process is very fast during the first 50 h and then the water content increases very slowly. When com-
paring the experimental results with the numerical and analytical models, the Langmuir model shows higher concentrations
between 50 h and 200 h and lower after 200 h. Both time dependent diffusion models used in the FE models adjust better to
542 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Fig. 7. Temperature distribution on the sample under wet condition.

Fig. 8. A comparison of temperature determined numerically and experimentally measured by FBG sensor for (a) dry and (b) wet conditions.

the experimental data curve. However, between both, the Time Varying model presents the saturation process more
precisely.

5.3. Hygro-mechanical analysis

In order to quantify the deformation suffered by the material due to moisture absorption, the spatial moisture concen-
tration profile obtained from the previous mass difussion analysis was sequentially combined with a mechanical deforma-
tion analysis.
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 543

Fig. 9. Comparison of average concentration received from experiment (exp) and simulation (with 1 mm and 0.5 mm mesh density): (a) Time Varying,
(b) Dual Stage.

Fig. 10. Comparison of average concentration [%] received from simulation and experiment; exp - experiment, L Langmuir model, TV Time Varying,
DS Dual Stage.

AbaqusTM does not offer the possibility to directly use the concentration distribution for a hygro-mechanical analysis. It
does, however, offer the option to perform a thermo-mechanical analysis. Given the close similarities between both phenom-
ena, it was possible to use these capabilities for the hygro-mechanical analysis.
The normalized concentration values at each node were imported into the mechanical analysis as temperature values.
This was achieved through the Fortran code developed by Yoon et al. [47]. The code reads the record key of the mass diffu-
sion results database and changes it from normalized concentration (key 221) to nodal temperature (key 201). Thus, the
results can be imported into the mechanical analysis as a predefined temperature field using the usual AbaqusTM commands.
During the mechanical analysis, the hygro-mechanical strain is calculated as a thermal strain by specifying the coefficient
of hygroscopic swelling as the thermal expansion coefficient in the material properties, as it was done in this case. If one
desires to conduct a coupled hygro-thermal mechanical analysis, that is also possible by importing the concentration values
as a predefined field and using the user subroutine UEXPAN to define the coefficient of hygroscopic swelling and thermal
expansion. An example of this approach is given in Yoon et al. [47]. Another user-defined subroutine was implemented
by Anshari et al. [48] to model the response of moisture concentration changes in woods. Also, Rafsanjani et al. [49] applied
user-defined material routines for the simulation of hygro-mechanical behaviour of a hierarchical cellular material growth
rings of softwood.
Since only one fourth of the sample was used, symmetric boundary conditions were applied on the corresponding faces.
Also, a condition of zero displacement in the y-axis was applied to the bottom surface since it is simply supported by the
tray. The weight of the composites on the adhesive was included in the analysis through the use of a gravity load. Another
external force considered was the water pressure on top of the samples. The samples were submerged in a tray with dem-
ineralised water. The column of water on top of the samples had a maximum height of 5 mm. Thus, the water pressure on top
544 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Fig. 11. Comparison of strain for calculations with (p)/without (np) water pressure influence; TV Time Varying, DS Dual Stage.

Fig. 12. Comparison of hygroscopic strain determined numerically for two frictional coefficients (0.8 and 0.9) applying methods: (a) Time Varying, (b) Dual
Stage.

Fig. 13. Comparison of hygroscopic strain received from experiment (exp) and simulation (with 1 mm and 0.5 mm mesh density): (a) Time Varying, (b)
Dual Stage.
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 545

Fig. 14. Comparison of hygroscopic strain received from simulation (with mesh density of 1 mm) and experiment; exp - experiment, TV Time Varying, DS
Dual Stage.

of the sample had a magnitude of 6.24E+08 g/mm h2. An influence of pressure on strain determined numerically in FBG sen-
sors location is presented in Fig. 11.
The interaction between the composites and the adhesive was very important in the mechanical analysis. This interaction
was defined as a contact interaction with tangential behaviour using a high frictional coefficient. Simulations were per-
formed using a 0.8 and a 0.9 friction coefficient, given the very slight differences between both results a friction coefficient
of 0.9 was arbitrarily chosen. The comparison between the results with both values of friction coefficient is presented in
Fig. 12. The interaction was defined between the top surface of the bottom composite and the bottom surface of the adhe-
sive; and between the bottom surface of the top composite and the top surface of the adhesive.
When a predefined field is imported, nodes in the mesh are assigned the respective values. To this end, it is advisable to
keep the meshes between the analyses as similar as possible. If there is a mismatch between the meshes, the algorithm tries
to interpolate values between nodes. Of course, this can introduce more variability in the study. For this case, it was pre-
ferred to keep identical meshes for the adhesive between analyses. As for the composites, a coarse uniform mesh of
1 mm was imposed on both parts. The composites were not areas of interest in the analysis and a 1 mm mesh provides a
good aspect ratio since both have a thickness of 1 mm. All part instances were assigned 8-node linear brick elements with
reduced integration (type C3D8R).
Strain measurements taken by the FBG sensors correspond to very local areas inside the adhesive. Sets of elements from
the FE model were taken from these areas in the sample and results were compared to experimental values. The results for
the strain profiles for each different mesh per diffusion approach are shown on Fig. 13. For the strain profiles, the dual-stage
approach is able to adapt better to the experimental results than the time-varying BC approach, especially around the 150th
hour. Also, the use of a finer mesh did give more accurate results for the strain profiles for the FBG sensor.
In Fig. 14 a comparison of hygroscopic strain received from simulation (with mesh density of 1 mm) and experiment is
presented. Neither approach, however, follows the experimental profile trends on either one of the samples after the 200th
hour of the experiment. In the sample, the experimental values show a steady increase from an approximate strain value of
0.00012 to 0.00014; with a slight decrease towards the end of the measurements.
During the experiment, all external conditions were kept constant. The only load varying with time was the amount of
moisture entering the sample. The hygroscopic strains follow a linear relationship with the concentration difference, as sta-
ted in Eq. (4). A strain value of 0.00014 in the sample, would then imply a concentration of 3.47%, which is higher than the
3.126% measured from weight-gain absorption data.
A change in temperature could affect the diffusivity coefficient in the samples and provoke thermal strains that were not
accounted for during the measurements with the dry sample. The samples were placed in a temperature chamber controlled
and monitored at 333 K 2 K, throughout the experiment. A 2 K increase in the temperature would account for an increase of
only 0.0000183 in the strain value. The differences in strain values shown are 0.00002945 for the sample.
Experimental strain values were calculated by the changes in wavelength measured by the FBG sensors. It is possible that
during a wavelength measurement, the spectrum recorded by the FBG is not composed of only one clear peak but rather
several peaks. The measurement equipment used, takes the highest of the peaks and presents that as the measurement. This
highest peak may not be representative of the sample. A comparison between spectra and wavelength could help determine
if this is the case for the odd measurements presented in the strain profiles. Unfortunately, the measuring unit is able to
record either wavelength or spectra only, not both simultaneously, so such a comparison was not available.
546 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Fig. 15. Comparison of strain induced by moisture received from simulation (with 1 mm and 0.5 mm mesh density): (a) Time Varying, (b) Dual Stage.

Table 4
Comparison of computational time for two mesh sizes.

Mesh density
1 mm 0.5 mm
Mass diffusion analysis
Time-varying (s) 15 44
Dual-stage (s) 17 47
Problem size (elem.) 9375 37 500
(13 104 n.) (51 204 n.)
Hygro-mechanical analysis
Time-varying 16 min 1 h 36 min
Dual-stage 24 min 1 h 22 min
Problem size 15 927 elem. 44 052 elem.
(26 820 n.) (64 920 n.)

5.4. Mesh comparison

A comparison of average concentration and hygroscopic strains received for two mesh densities was presented in Figs. 9
and 13, respectively. In Fig. 15 a comparison of relationship between strain and average concentration is presented. The max-
imum differences between strain for those two mesh sizes are about 4.5e6 and 1.5e6 for Time Varying and Dual Stage
methods, respectively. As it is presented it is clear that the use of the 1 mm mesh is sufficient enough to get a reliable con-
centration profile. A refined mesh is not necessary since the desired result is not focused on a particular section of the adhe-
sive but rather the average concentration across the whole sample (Fig. 9). For the subsequent mechanical analysis, however,
results are wanted from specific areas of the sample where a more refined mesh might prove to be more accurate (Figs. 13
and 15).
The difference in computational time between those two meshes for whole numerical analysis is presented in Table 4.
The differences between two chosen methods (Dual Stage and Time Varying) are small, although the Dual Stage model needs
more time. For the mass diffusion analysis the computation time is 3 times longer for 0.5 mm than for 1 mm mesh size,
although it is still shorter than 60 s. For hygro-mechanical analysis it is about 6 times longer and the time is about 1.5 h,
while whole analysis for 1 mm mesh size is shorter than 0.5 h. As it was mentioned before the results are reasonable so
for the presenting of moisture concentration determination method the 1 mm mesh size was used.

6. Method for determination of moisture concentration

A comparison of the relationships between hygroscospic strains and moisture concentration for both experimental and
numerical analysese is presented in Fig. 16. The strain values calculated using numerical analyses are higher than those
determined experimentally. The highest differences corresponds to moisture concentration in a range of 1.2% 2.3%. For
better visualization a percentage difference was determined according to the relationship
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 547

Fig. 16. A comparison of strain induced by moisture for experimental and numerical results; exp experimental, DS Dual Stage, TV Time Varying.

Table 5
Percentage differences ED for strain induced by moisture.

Moisture concentration [%] Dual stage (%) Time varying (%)

1.1 32 22
1.6 35 26
2.7 17 16

Fig. 17. A comparison of slope values (multiplied by 1e7) calculated for different moisture concentration for experimental and numerical results; exp
experimental, DS Dual Stage, TV Time Varying.

eNM c  eexp c
EDe 100% 12
eexp c
where NM means numerical method Dual Stage or Time Varying and exp - experimental investigation. The percentage dif-
ferences EDe values for three chosen points are presented in Table 5. As it is visible, they vary in a range of 1735% or 1626%
for Dual Stage and Time Varying methods, respectively. The highest value (35%) was reached for Dual Stage method for a
moisture concentration equal to 1.6%.
Because every curve shape can be approximated using an appropriate number of straight lines with different slopes this
procedure can be also implemented for curves presented in Fig. 16. The slope of the straight line through two points Ax1 ; y1
and Bx2 ; y2 , when x1 x2 , is given by a relationship
y2  y1
a 13
x2  x1
548 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549

Table 6
Percentage differences EDa for slope values (multiplied by 1e7).

Slope values (multiplied by 1e7) Dual stage (%) Time varying (%)
5 6 7
15 3 3
25 8 9

A comparison of slope values a calculated for 1 h interval for both numerical and experimental results is presented in Fig. 17.
For better visibility, the slope values were multiplied by 1e7. An application of using experimental relationship between
strain and moisture concentration in the adhesive layer for determination of moisture concentration in different moments
in time of both soaking and drying processes was described in a previous paper [20].
The comparison (Fig. 17) shows good agrement between numerical and experimental values especially in the range of
1.52.5% where all curvatures (numerical and experimental - see Fig. 16) have very similar shape. For better visibility, sim-
ilarly like for strain induced by moisture, a percentage difference was determined according to a relationship

cNM a  cexp a
EDa 100% 14
cexp a

where NM means numerical method Dual Stage or Time Varying and exp - experimental investigation. The percentage dif-
ferences EDa values for three chosen points are presented in Table 6. As it is visible, they vary in a range of 39% for both
methods. The highest accuracy of numerical models was achieved for a moisture concentration in a range of 1.92.4%
Although the relationship of the moisture content and proposed index a for numerical results are not linear the parameter
a can be used for determination of amount of moisture in adhesive joint.

7. Conclusions

The paper presents numerical models that could represent the thermal response, mass diffusion dynamics and moisture-
induced strain propagation of the adhesive joint. The achieved results were validated using experimental investigation
results for samples with FBG sensors embedded into adhesive layer.
The thermal model was constructed to simulate the temperature distribution on the sample inside the chamber in dry
and wet conditions. When in dry conditions, the thermal model was able to accurately simulate the temperature profile
obtained by the FBG sensor embedded in the sample. Simulations results show that the sample was able to reach thermal
equilibrium within the first 5 min. There were hours of difference between the weight measurements and the strain read-
ings. Thus, one can conclude that, even when considering the error involved in the calculation of the convection coefficient,
the sample was in an isothermal state when the strain measurements were recorded.
Regarding the mass diffusion analysis, the main drawback from either one of the approaches used for anomalous diffu-
sion, and also from Fickian equations, is that they all use variables obtained from experimental values. When comparing the
overall concentration of the sample from both approaches with experimental data, the time-varying boundary condition
shows a better correlation than the dual-stage model. However, the time-varying boundary condition model requires the
adjustment of variables to provide the best fit. The dual-stage model requires less information from experimental results
and was able to provide a good correlation with results.
The FE models created for the hygro-mechanical analysis show good correlation with experimental results for 60% of the
simulation time. The reasons for the incongruences towards the end of the simulation remain unknown. New experimental
data is needed to verify if the erratic behaviour is due to the model or to unaccounted factors during the measurement of this
particular set of values. The presented models can also be used for the determination of moisture content in an adhesive
layer especially in a range of 1:5  2:5% of water content.
Additionally a comparison of numerical analysis results for two mesh sizes (0.5 mm and 1 mm) was presented. The dif-
ferences between them were not so high, while the computation time for 0.5 mm mesh was 3 times longer than for the sec-
ond one. Due to this the method for determination of moisture concentration was presented for 1 mm mesh only showing
also good agreement with experimental results.
The authors wish to continue their work in this area of research in the future to overcome all the problems and difficulties
encountered during the line of work presented in this paper.

Acknowledgements

The research was supported by the project (PBS1/B6/8/2012) entitled: Non-invasive Methods for Assessment of Physic-
ochemical and Mechanical Degradation granted by National Centre for Research and Development in Poland.
Authors would like to give special thanks to Academic Computer Centre in Gdansk (CI TASK) for access to AbaqusTM
software.
 V. Bonilla Mora et al. / Mechanical Systems and Signal Processing 99 (2018) 534549 549

References

[1] L. da Silva, A. chsner, R. Adams, Handbook of Adhesion Technology, Springer Science & Business Media, 2011.
[2] W. De Goeij, M. Van Tooren, A. Beukers, Composite adhesive joints under cyclic loading, Mater. Des. 20 (5) (1999) 213221.
[3] M. Lai, J. Botsis, J. Cugnoni, D. Coric, An experimentalnumerical study of moisture absorption in an epoxy, Compos. Part A: Appl. Sci. Manuf. 43 (7)
(2012) 10531060.
[4] S. Budhe, M. Banea, S. de Barros, L. da Silva, An updated review of adhesively bonded joints in composite materials, Int. J. Adhes. Adhes. 72 (2017).
[5] S. Eslami, A. Honarbakhsh-Raouf, S. Eslami, Effects of moisture absorption on degradation of e-glass fiber reinforced vinyl ester composite pipes and
modelling of transient moisture diffusion using finite element analysis, Corrosion Sci. 90 (2015) 168175.
[6] X. Fan, S. Lee, Q. Han, Experimental investigations and model study of moisture behaviors in polymeric materials, Microelectron. Reliab. 49 (8) (2009)
861871.
[7] C. Browning, G. Husman, J. Whitney, Moisture effects in epoxy matrix composites, in: Composite Materials: Testing and Design (Fourth Conference),
ASTM International, 1977.
[8] R. Morgan, J. Oneal, D. Fanter, The effect of moisture on the physical and mechanical integrity of epoxies, J. Mater. Sci. 15 (3) (1980) 751764.
[9] H. Xin, A. Mosallam, Y. Liu, F. Yang, Y. Zhang, Hygrothermal aging effects on shear behavior of pultruded frp composite web-flange junctions in bridge
application, Compos. Part B: Eng. 110 (2017) 213228.
[10] Y. Weitsman, Fluid Effects in Polymers and Polymeric Composites, Springer Science & Business Media, 2011.
[11] J. Garcia-Espinel, D. Castro-Fresno, P.P. Gayo, F. Ballester-Muoz, Effects of sea water environment on glass fiber reinforced plastic materials used for
marine civil engineering constructions, Mater. Des. (19802015) 66 (2015) 4650.
[12] Y. Qin, Y. Li, Influences of hygrothermal environment and installation mode on vibration characteristics of a rotating laminated composite beam, Mech.
Syst. Signal Process. 91 (2017) 2340.
[13] Y. Dong, Y. Zhang, Y. Li, An analytical formulation for postbuckling and buckling vibration of micro-scale laminated composite beams considering
hygrothermal effect, Compos. Struct. 170 (2017) 1125.
[14] X. Li, Y. Li, Y. Qin, Free vibration characteristics of a spinning composite thin-walled beam under hygrothermal environment, Int. J. Mech. Sci. 119
(2016) 253265.
[15] B. Jiang, J. Xu, Y. Li, Flapwise vibration analysis of a rotating composite beam under hygrothermal environment, Compos. Struct. 117 (2014) 201211.
[16] Y. Qin, X. Li, E. Yang, Y. Li, Flapwise free vibration characteristics of a rotating composite thin-walled beam under aerodynamic force and hygrothermal
environment, Compos. Struct. 153 (2016) 490503.
[17] M. Wahab, A. Crocombe, A. Beevers, K. Ebtehaj, Coupled stress-diffusion analysis for durability study in adhesive bonded joints, Int. J. Adhes. 22 (3)
(2002) 6173.
[18] A. Higgins, Adhesive bonding of aircraft structures, Int. J. Adhes. 20 (3) (2000) 367376.
[19] D. Karalekas, J. Cugnoni, J. Botsis, Monitoring of hygrothermal ageing effects in an epoxy resin using fbg sensor: a methodological study, Compos. Sci.
Technol. 69 (3) (2009) 507514.
[20] M. Mieloszyk, W. Ostachowicz, Moisture contamination detection in adhesive bond using embedded fbg sensors, Mech. Syst. Signal Process. 84 (2017)
114.
[21] D. Karalekas, J. Cugnoni, J. Botsis, Monitoring of hygrothermal ageing effects in an epoxy resin using fbg sensor: a methodological study, Compos. Sci.
Technol. 69 (3) (2009) 507514.
[22] E. Udd, J.W. Spillman, Fiber Optic Sensors: An Introduction for Engineers and Scientists, second ed., Wiley, 2011.
[23] D. Karalekas, J. Cugnoni, J. Botsis, Monitoring of process induced strains in a single fibre composite using fbg sensor: a methodological study, Compos.
Part A: Appl. Sci. Manuf. 39 (7) (2008) 11181127.
[24] Q. Chen, P. Lu, Fibre bragg grating and their applications as temperature and humidity sensors, At. Mol. Opt. Phys. (2008) 235260.
[25] L. Canal, V. Michaud, Micro-scale modeling of water diffusion in adhesive composite joints, Compos. Struct. 111 (2014) 340348.
[26] J. Wilmers, S. Bargmann, Simulation of non-classical diffusion in polymers, Heat Mass Transfer 50 (11) (2014) 15431552.
[27] H. Corporation, Loctite ea 9394 aero, 2013. <http://na.henkel-adhesives.com/product-search-1554.htm?nodeid=8797801054209>.
[28] T.E. Toolbox, Water thermal properties, 2016. <http://www.engineeringtoolbox.com/water-thermal-properties-d_162.html>.
[29] T. Keller, C. Tracy, A. Zhou, Structural response of liquid-cooled gfrp slabs subjected to firepart i: Material and post-fire modeling, Compos. Part A:
Appl. Sci. Manuf. 37 (9) (2006) 12861295.
[30] Y. Bai, T. Valle, T. Keller, Modeling of thermal responses for frp composites under elevated and high temperatures, Compos. Sci. Technol. 68 (1) (2008)
4756.
[31] T. Guess, E. Reedy, M. Stavig, Mechanical properties of hysol ea-9394 structural adhesive, SAND95-0229, Sandia National Laboratories, Albuquerque,
New Mexico, 1995.
[32] R. Simons, Simplified formula for estimating natural convection heat transfer coefficient on a flat plate, Electron. Cooling Mag. 8 (1) (2002).
[33] L. Jiji, Heat Convection, Springer, 2006.
[34] T.E. Toolbox, Air properties, 2016. <http://www.engineeringtoolbox.com/air-properties-d_156.html>.
[35] A. Jaffer, Convection from a rectangular plate, 2016. <http://people.csail.mit.edu/jaffer/SimRoof/Convection/>.
[36] S. Yoon, B. Han, S. Cho, C. Jang, Experimental verification of non-linear finite element analysis for combined hygroscopic and thermo-mechanical
stresses, in: Proceedings of 2005 SEM Annual Conference, 2005.
[37] F. Mller-Plathe, S. Rogers, W. van Gunsteren, Computational evidence for anomalous diffusion of small molecules in amorphous polymers, Chem.
Phys. Lett. 199 (34) (1992) 237243.
[38] A. Mubashar, I. Ashcroft, G. Critchlow, A. Crocombe, Modelling cyclic moisture uptake in an epoxy adhesive, J. Adhes. 85 (10) (2009) 711735.
[39] G. LaPlante, A. Ouriadov, B. Lee-Sullivan, P. Balcom, Anomalous moisture diffusion in an epoxy adhesive detected by magnetic resonance imaging, J.
Appl. Polym. Sci. 109 (2) (2008) 13501359.
[40] H. Carter, K. Kibler, Langmuir-type model for anomalous moisture diffusion in composite resins, J. Compos. Mater. 12 (2) (1978) 118131.
[41] H. Shirangi, J. Auersperg, M. Koyuncu, H. Walter, W. Muller, B. Michel, Characterization of dual-stage moisture diffusion, residual moisture content and
hygroscopic swelling of epoxy molding compounds, in: International Conference on Thermal, Mechanical and Multi-Physics Simulation and
Experiments in Microelectronics and Micro-Systems, 2008 (EuroSimE 2008), IEEE, 2008, pp. 18.
[42] J. Crank, The Mathematics of Diffusion, Oxford University Press, 1979.
[43] L.-W. Cai, Y. Weitsman, Non-fickian moisture diffusion in polymeric composites, J. Compos. Mater. 28 (2) (1994) 130154.
[44] F. Ellyin, R. Maser, Environmental effects on the mechanical properties of glass-fiber epoxy composite tubular specimens, Compos. Sci. Technol. 64 (12)
(2004) 18631874.
[45] D.S.S. Corp., 2.11 heat transfer, Abaqus Theory Guide, 2016.
[46] M. Shirangi, B. Michel, Mechanism of moisture diffusion, hygroscopic swelling, and adhesion degradation in epoxy molding compounds, in: Moisture
Sensitivity of Plastic Packages of IC Devices, Springer, 2010, pp. 2969.
[47] S. Yoon, C. Jang, B. Han, Nonlinear stress modeling scheme to analyze semiconductor packages subjected to combined thermal and hygroscopic
loading, J. Electron. Packag. 130 (2) (2008) 024502.
[48] B. Anshari, Z. Guan, Q. Wang, Modelling of glulam beams pre-stressed by compressed wood, Compos. Struct. 165 (2017) 160170.
[49] A. Rafsanjani, D. Derome, F.K. Wittel, J. Carmeliet, Computational up-scaling of anisotropic swelling and mechanical behavior of hierarchical cellular
materials, Compos. Sci. Technol. 72 (6) (2012) 744751.