Accepted Manuscript

Title: Simulation of the development of local panel distortions

due tohot-curing adhesives

Author: Konstantin Priesnitz K.M.B. Jansen Jos Sinke Rinze

Benedictus

PII: S0924-0136(15)30024-8
DOI: http://dx.doi.org/doi:10.1016/j.jmatprotec.2015.06.013
Reference: PROTEC 14448

To appear in: Journal of Materials Processing Technology

Received date: 14-1-2015

Revised date: 8-6-2015
Accepted date: 9-6-2015

Please cite this article as: Konstantin Priesnitz, K.M.B. Jansen, Jos Sinke,
Rinze Benedictus, Simulation of the development of local panel distortions
due tohot-curing adhesives, Journal of Materials Processing Tech. (2015),
http://dx.doi.org/10.1016/j.jmatprotec.2015.06.013

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
*Manuscript

Simulation of the development of local panel distortions

due to hot-curing adhesives

t
ip
Konstantin Priesnitza,b,∗, K.M.B. Jansenc , Jos Sinkeb , Rinze Benedictusb
a
Materials innovation institute (M2i), Mekelweg 2, 2628CD Delft, the Netherlands

cr
b
Faculty of Aerospace Engineering, Delft University of Technology (TU Delft), 2629HS
Delft, the Netherlands
c
Faculty of Industrial Design Engineering, TU Delft, Landbergstraat 15, 2628CE Delft,

us
the Netherlands

an
Abstract

A model was developed and applied to describe the development of local

M
panel distortions caused by hot-curing adhesives. The cure evolution, vol-
ume changes and evolving mechanical properties of a commercial hot-curing
epoxy system were integrated into the simulation model; material parame-
d

ters were determined experimentally. Predictions of distortions for different

te

temperature cycles were compared with experimental data. The model pre-
dicts well the deformations that occur over an entire cure cycle including the
p

heating and isothermal phase.

ce

Keywords: Adhesive bonding, cure, chemical shrinkage, bond-line

read-through
Ac

1. Introduction

Adhesive bonding is known to offer certain advantages over other joining

techniques: Barnes and Pashby (2000) name the ability to join dissimilar

∗
Corresponding author, Tel.:+31152788673; fax: +31152781151.
Email address: k.w.priesnitz@tudelft.nl (Konstantin Priesnitz)

Preprint submitted to Journal of Materials Processing Technology June 7, 2015

Page 1 of 46
materials, the increase in structure stiffness because of large area bonds in-
stead of localized contacts and additional functionality such as vibration

t
damping or sealing as benefits. Adderley (1988) points out that one-part,

ip
hot-curing epoxies are often chosen in automotive or aerospace applications

cr
for structural bonds. The cure at elevated temperature, however, can cause
distortions, especially if thin-walled structures like e.g. car body shells are

us
involved. Fuchs et al. (2010) divided panel distortions into local and global
ones. While global ones describe the overall deformation of the bonded struc-

an
ture, local distortions appear along the bond line close to the adhesive.
These distortions are caused by the evolving properties of the adhesive
over a cure cycle: the adhesive transforms from a viscous liquid to a vis-
M
coelastic solid. At the same time, the cross-linking process causes a volume
reduction of the polymer, often referred to as chemical shrinkage. In addition
d

both the adhesive and the adherends expand and shrink due to a changing
te

temperature. These simultaneous processes can lead to panel distortions.

In several studies, the development of panel distortions is investigated by
p

continuous measurements over a cure cycle. Hahn et al. (1993) measured the
ce

shear modulus of an epoxy during cure. The displacement curve of a steel

strip during the bonding process in different cure cycles was determined by
Ac

Priesnitz et al. (2014a); Chudaska and Hahn (1993) monitored the position
of a composite panel. These publications show that distortions can occur
early in the cure cycle when the structure is still heating up.
Hahn and Jendrny (2003) predicted the displacement of a steel strip dur-
ing cooling. Fuchs et al. (2010) predicted the curvature of a steel and a
composite panel. Patankar et al. (2013) calculated residual stresses in a bi-

Page 2 of 46
material strip under cyclic thermal loading. These models presume a stress-
free state until the adhesive is cured completely. Adolf and Martin (1996)

t
present a small-strain constitutive model for stresses in cross-linking poly-

ip
mers, whereas Lion and Höfer (2007) propose a model for finite strain. Those

cr
models are suitable to describe the complete cure cycle, but have not been
used for an in-depth analysis of the development of local panel distortions.

us
In this work, the development of local panel distortions is simulated and
the simulation results are compared with experiments. To that end, the

an
material constants of a commercial epoxy-based hot-curing adhesive system,
Betamate— 1496V (BM1496V), are determined. The simulation model is
based on an earlier work by Priesnitz et al. (2014b) and allows for chemically
M
and thermally induced property changes of the adhesive. Priesnitz et al.
(2014a) provide experimental data on distortion development in different
d

cure cycles, which is used as a comparison for the simulation results.

te

2. Material modelling and characterization

p

2.1. Cure kinetics

ce

The cross-linking process of polymer chains in the adhesive can be char-

acterized with a single variable q, the degree of cure. In the uncured state
Ac

before any cross-links are formed, q has the value 0. During a cure cycle, q
increases steadily until no reactive groups are left over. In that fully cured
state, q has the value 1.
A well-known equation for describing thermoset cure was proposed by

Page 3 of 46
Kamal and Sourour (1973):
 
Ea m
q̇ = k0 exp − q (1 − q)n , (1)

t
RT

ip
where R = 8.314 J/(mol K) is the ideal gas constant. Fitting the equation
to experimental data determines the material parameters k0 , Ea , m and n.

cr
Kinetic parameters can be obtained with a differential scanning calorimeter

us
(DSC). A DSC monitors the heat flow dH/dt over time for a given temper-
ature profile. Since each epoxy group reacts exothermically and, therefore,
releases a specific amount of heat, the degree of cure can be linked to the
heat flow:
dq(t) an
1 dH(t)
M
= (2)
dt H∞ dt
with H∞ being the (total) heat of reaction. Here, uncured samples were
d

exposed to different heating rates while the heat flows were monitored over
time. Similar to the approach of Jansen et al. (2007), the kinetic parameters
te

were obtained as follows: the activation energy Ea was determined with a

p

modified Kissinger approximation referred to as the type B-1.92 method by

Starink (2003), which is based on the linear relation
ce

β Ea
ln 1.92
= −1.0008 + C, (3)
T RTf
f
Ac

where ln is the natural logarithm, β is the temperature rate, Tf is the isocon-

versional temperature and C is a constant independent of β and Tf . Thus,
the slope of the plot ln(β/Tf1.92 ) versus 1/Tf determines Ea .
Modifying eq. 1 to
q̇
 Ea  = k0 q m (1 − q)n (4)
exp − RT

Page 4 of 46
and fitting that equation to DSC data determines the remaining parameters
k0 , m and n.

t
ip
2.2. Volume changes
The change in the specific volume v of a material point in a continuum

cr
is described by

us
v = v0 J, (5)

where v0 is the specific volume in a reference state and J is the Jacobi de-

an
terminant of the deformation gradient. For small deformations, that is when
second-order and higher-order terms of the displacement gradient ∇u can be
neglected, J can be written as
M
J = 1 + εkk (6)

with εkk being the trace of the linear strain tensor.

d

An isotropic material is assumed. Liebl et al. (2012) decomposed, like

te

many others, the strain in several parts. Here, a distinction is drawn between
mechanical, thermal and chemical contributions:
p

εij = εme th ch me th ch
ij + εij + εij = εij + ε δij + ε δij , (7)
ce

where δij is the Kronecker Delta. In case an isotropic material undergoes no

other volume changes than thermal expansion or contraction (i.e. no pressure
Ac

or other stresses are applied, nor are volume changes chemically induced),
then, from eq. 5 and 6 follows:

v = v0 (1 + 3εth ) (8)
ZT
th
ε = α(T̃ )dT̃ (9)
T0

Page 5 of 46
εth is the thermal strain, T0 is the reference temperature at which the material
has the volume v0 , T is the current temperature and α is the linear coefficient

t
of thermal expansion (CTE). A volumetric CTE can be defined as

ip
1 dv(T )
αvol (T ) := = 3α(T ). (10)

cr
v0 dT

If only thermal and chemical deformations occur,

us
v = v0 (1 + 3εth + 3εch ) (11)

an
holds.
Li et al. (2004) describe the volume changes in an idealized cure cycle, in
which the adhesive is heated up, held at a cure temperature Tc and cooled
M
back down to ambient temperature T0 , see fig. 1. The epoxy expands ther-
d

1
v
p te

0 2
3
ce

T0 Tg Tc T
Ac

Figure 1: Volume changes over an idealized cure cycle (Li et al., 2004).

mally while heated up (0-1), shrinks chemically at cure temperature (1-2),

then shrinks due to temperature decrease, first with a high, rubbery CTE αr

Page 6 of 46
(2-3), then, below the glass transition temperature Tg , with a lower, glassy
CTE αg (3-4). It is noteworthy that in this cycle a glassy state trajectory in

t
the heating phase, as described by Gromala et al. (2011), is not accounted for.

ip
It was pointed out by Li et al. (2004) that standard procedures to measure

cr
cure shrinkage are performed at ambient temperature, i.e. these methods
measure an effective shrinkage

us
1 v4 − v0
ε̃ch
tot = (12)
3 v0
(from point 0 to 4 in fig. 1) rather than the real chemical shrinkage εch

an
tot (1-2).

As an approximation, the thermal expansion of the adhesive during heat-

ing up is assumed to occur with the same CTE αr as in the rubbery state
M
during cooling down. Therefore, the total chemical shrinkage can be esti-
mated in case αr , αg and ε̃ch
tot are known:
d

εch ch
tot = ε̃tot + (αg − αr )(Tg − T0 ). (13)
te

De Vreugd et al. (2007) found the chemical shrinkage to increase linearly

p

with the degree of cure. Here, the same relation is assumed:

ce

εch (q) = q εch

tot . (14)

The function
Ac

 
1 1
α(T ) = k1 + k2 (1 + tanh(C0 (T − Tg ))) (15)
3 2
with the material constants k1 , k2 and C0 describes the transition from a
rubbery CTE,
1
αr = (k1 + k2 ), (16)
3

Page 7 of 46
to a glassy one,

1
αg = k1 , (17)

t
3

ip
based on the Tait equation of state. Sadeghinia et al. (2012) used it ear-

cr
lier to characterize epoxy moulding compounds. De Vreugd et al. (2010)
deployed it to predict the warpage of microelectronic products during cure.

us
C0 determines the sharpness of the transition from the glassy to the rubbery
state. The constants are determined by thermomechanical analysis (TMA):

an
a 4.7 x 5.1 x 4.2 mm-sized block of fully cured material is exposed to three
temperature cycles in which the polymer is heated up slowly to about 200
° C and cooled down back to room temperature. The third heating phase is
M
used to determine the CTE. While the curve of the second and third heat-up
show hardly any difference, the first heating phase seemingly shows a nega-
d

tive CTE when approaching Tg . This observation, however, is attributed to

te

a rearrangement of the contact area of the probe and the specimen during
the first heating cycle. Therefore, the first heat-up phase is ignored.
p

2.3. Mechanical properties

ce

The viscoelastic properties of the fully cured material are determined with
a dynamic mechanical analyser (DMA). In a tensile set-up, the elongation
Ac

of a strip of the epoxy (23.0 x 7.58 x 0.34 mm) is monitored under dynamic
loading. A constant frequency sweep from 1 to 60 Hz is performed repeatedly
during continuous heating from 20 to 180 °C with 1 K/min.
A thermorheologically simple material behaviour is assumed such that a
master curve for the tensile storage modulus E 0 can be obtained by shifting

Page 8 of 46
the isothermal curves according to

E 0 (ω, T ) = E 0 (ωa(T ), Tg ) , (18)

t
ip
where ω is the frequency and Tg is chosen as the reference temperature.

cr
The shift factors log a(T ) are fitted to the Williams-Landel-Ferry equation
(Williams et al., 1955) above Tg and to an Arrhenius model below:

us

C1 (T −Tg )
 −C 2 +T −Tg
for T ≥ Tg
log a(T ) = h i (19)
A 1

2.303R T
− T1g for T < Tg ,

where C1 , C2 and A are fitting constants.

an
M
3. Simulation

A model has been developed to simulate the cure processes investigated

d

by Priesnitz et al. (2014a). The constitutive equations for the adhesive in

te

that model coincides with the ones used in a previous work of the authors
(Priesnitz et al., 2014b) with minor modifications. For completeness rea-
p

sons, those are summarized here. The material data obtained in sect. 2
ce

is implemented in that model and allows the comparison of simulation and

experiment.
Ac

3.1. Constitutive equations

An isotropic material is assumed whose deformation can be described by

eq. 7. Increments of thermal strain are calculated by

1
∆εth = (αi + αi+1 )∆T. (20)
2

Page 9 of 46
The CTE at the start (αi ) and the end (αi+1 ) of the increment are calculated
with eq. 15 in case full cure is reached and with eq. 16 otherwise. Eq. 14

t
is used for the chemical strain contribution. Eq. 1 is used to calculate the

ip
cure evolution; in order to ensure a non-zero starting cure rate, q0 = 0.01 is

cr
chosen as an initial value.
To derive the constitutive equations, the stress tensor σij is split in its

us
deviatoric part σijdev = σij − 13 σkk δij and the hydrostatic pressure p:

σij = σijdev − pδij . (21)

an
In this model, volume changes are described by the hypo-elastic equation:

−ṗ = K ε̇me
M
kk (22)

with a constant bulk modulus K. Experimental determination of K can be

d

a difficult task and is not done in this work because of lack of equipment. In-
te

stead, it is assumed that the bulk modulus has less influence on stresses than
the tensile or shear modulus in cases where the adhesive occupies an uncon-
p

fined open geometry. That is the case for the bond line in the experimental
ce

set-up investigated here and and described by Priesnitz et al. (2014a). The
bulk modulus changes by a factor of about 2 to 3 during the glass-to-rubber
transition, which is the much less than the 100- to 1000-fold changes of the
Ac

shear and elongation modulus during that transition. Therefore, to come up

with value for K, a quasi-static approximation is used:
Eg
K= , (23)
3(1 − 2ν)
where Eg is the instantaneous glassy tensile modulus obtained by DMA. For
ν = 0.45 is chosen.

10

Page 10 of 46
 In order to model the shear behaviour, the cure cycle is divided into three
different stages (see fig. 2). In stage I, the adhesive is liquid and cannot exert

t
ip
T I II III

cr
us
t
Gel point Glass transition

an
Figure 2: Three stages of a cure cycle.

permanent stresses on the steel strip. Here, this behaviour is realised by a

M
small but non-zero shear modulus.
In stage II, a rubbery shear modulus develops. A hypo-elastic approach
d

for deviatoric stress-strain relation is chosen:

te

σ̇ijdev = 2 G(q) ε̇me,dev

ij , (24)
p

where the shear modulus G is defined by the equation from Adolf and Martin
ce

(1996):
! 83
q 2 − qgel
2
G(q) = Gfinal , (25)
Ac

2
1 − qgel

with qgel being the degree of cure at the gel point and Gfinal the rubbery shear
modulus in the fully cured state. qgel depends on details of the chemical
formulation and varies from adhesive to adhesive. López et al. (2002) find
values between 0.3 and 0.6 for epoxy-based systems. The value qgel = 0.455,
measured by Menzel (2011) for BM1496V, is used in this model.

11

Page 11 of 46
 During transition to stage III, relaxation behaviour becomes important.
Therefore, in stage III, a linear viscoelastic material law is chosen:

t
M

ip
X
dev
σ̇ij = 2G∞ ε̇ij me, dev
+2 Gm u̇m
ij , (26)
m=1
1

cr
u̇m
ij + um = ε̇me,dev , m = 1, 2, ...M. (27)
a(T )τm ij ij

As stated by Lion (2007), these equations characterize a thermorheologically

us
simple material with a shift factor log a(T ). The shear relaxation modulus
at the reference temperature is given by

an
M
X
G(t) = G∞ + Gm exp(−t/τm ) (28)
m=1
M
with the rubbery modulus G∞ . A quasi-static approach is used to obtain
G(t) from a master curve of the tensile and bulk modulus:
3KE(t)
d

G(t) = . (29)
9K − E(t)
te

E is determined by the previously described DMA.

All mentioned constitutive equations are integrated in Abaqus using the
p

UMAT, USDFLD and UEXPAN user subroutines and applying the simple
ce

but stable increment scheme

∆f ∆f
f˙t+ ∆t = , ft+ ∆t = ft + , (30)
Ac

2 ∆t 2 2
where f is replaced by the corresponding stress or strain component. The
same scheme is used for the tangent moduli ∂∆σij /∂∆εme
kl .

3.2. Finite-element model

Geometry, material and process parameters of this finite-element sim-
ulation are chosen to coincide with the experimental set-up presented by

12

Page 12 of 46
 Table 1: Properties of the steel and aluminium used in the simulation.
Material E (GPa) ν (-) α (10−6 /K)

t
Steel DX54D+Z 210 0.3 13

ip
Aluminum 6082 69 0.33 23

cr
Priesnitz et al. (2014a). In that study, a 0.7 x 50 x 20 mm steel strip is

us
bonded to a 8 x 50 x 20 mm aluminium bulk part with a 10 mm wide and 2
mm thick bond line. Fig. 3 shows the geometry of the finite-element model

an
and a photo of the specimen used in the experiment. Symmetry allows to
reduce the model to a quarter piece of the real part. The marked edges of
the strip and the substructure are not connected; they cannot move in y-
M
direction, but slide in x- and z-direction. The material parameters of both

Adhesive
Symmetry planes
d
bond line
Steel strip
te

Aluminum
substructure
Node N2
Node N1
p
ce

• 2 edges
Ac

• Unconnected
• uy=0

Figure 3: Finite-element model and photo of the specimen to monitor local panel distor-
tions.

metals are listed in table 1. Properties of the adhesive are taken from liter-

13

Page 13 of 46
ature, table 3, or obtained by the characterization tests below. The model
consists of 16.190 brick elements of type C3D20R. Fig. 4 shows the different

t
temperature profiles in experiment and simulation. The simulation temper-

ip
ature profiles are chosen in such a way as to match the heating rate, cure

cr
temperature and cure time in the respective experiment. The initial step in
the experiment, in which the sample is kept at 50 ° C for one hour before the

us
cure cycle is started, is omitted since it does not affect the outcome of the
simulation. The 6 cycles investigated are listed in table 2.

an
M
d
p te
ce

Figure 4: Temperature profiles in experiment and simulation.

Ac

14

Page 14 of 46
 Table 2: Temperature cycles.
Name Heating rate Cure temp. Holding time

t
L170 1 K/min 170 °C 30 min

ip
M170 7 K/min 170 °C 30 min
170 °C

cr
H170 35 K/min 30 min
L155 1 K/min 155 °C 60 min

us
M155 7 K/min 155 °C 60 min
H155 35 K/min 155 °C 60 min

an
Table 3: Material constants for BM1496V from Menzel (2011).
Tg (° C) v0 (cm3 /g) v4 −v0
v0
(-) qgel (-)
M
85 0.862 -1.44 ·10−2 0.455

4. Results and discussion

d
te

4.1. Cure kinetics

Tables 4 lists the heat of reaction for each DSC run. The respective
p

exothermic peaks in the heat flow curves are shown in fig. 5. With the
ce

procedure described in section 2.1, the parameters in table 5 for eq. 1 are
found.
Ac

Assuming a cure evolution according to eq. 1 leads to an overestimation

of the cure rate or heat flow in case of the highest heating rate (see fig.
5), while good agreement is reached for the lower heating rates. High rates
can lead to an inhomogeneous temperature distribution in the DSC sample
biasing the result. Values in that region may carry more error. Keeping
that in mind, results are considered sufficient for the simulation model. In

15

Page 15 of 46
future work, it is desirable, however, to study the sensitivity of distortion
predictions to small variations in the cure progress.

t
ip
cr
us
an
M
Figure 5: Exothermic peaks from DSC tests with different temperature rates compared to
d

model predictions.
te

Table 4: H∞ for the different heating rates.

p

1 K/min 4 K/min 7 K/min 10 K/min 35 K/min

ce

H∞ (J/g) 224.2 217.8 207.0 210.8 202.0

Ac

16

Page 16 of 46
 Table 5: Cure parameters.
Ea (J/mol) k0 (1/s) m (-) n (-)

t
8.877·104 3.410·108 0.649 1.551

ip
4.2. Volume changes

cr
The TMA curve for the fully cured adhesive is shown in fig. 6. The

us
material shows a linear expansion below and above the glass transition tem-
perature with a greater CTE above. The value Tg = 85 ° C found by Menzel

an
(2011) agrees with TMA data here. The parameters k1 , k2 and C0 in table
6 and the dashed curve in fig. 6 originate from fitting eq. 15 to the TMA
data.
M
d
p te
ce
Ac

Figure 6: Thermal expansion of the fully cured adhesive.

The real chemical shrinkage εch

tot in table 6 is determined by eq. 13, 16, 17,

the effective chemical shrinkage in table 3, presumably measured at T0 = 20

° C, and the other parameters in table 3 and 6.
17

Page 17 of 46
 Table 6: Parameters characterizing expansion and shrinkage of the adhesive.
k1 (1/° C) k2 (1/° C) C0 (1/° C) εch
tot

t
3.840e-04 1.990e-04 0.0871 -9.112e-3

ip
cr
With the constants obtained, eq. 11 can now be used to describe the vol-
ume change of the adhesive in cases in which the stress state stays unaltered.

us
Figs. 7 and 8 show curves of these volume changes for the experimental
temperature cycles in fig. 4.

an
M
d
p te
ce

Figure 7: Predicted volume changes over the cure cycles L170, M170 and H170, and qgel
Ac

from table 3.

It is noteworthy that for all cycles, the adhesives show the same amount
of shrinkage after cycle completion. Differences only lie in cure evolution and,
therefore, in the development of cure-induced shrinkage. Fig. 7 shows that
with a low heating rate (continuous black curve), the cure and corresponding
volume changes start already during the heating phase. In the fast heating

18

Page 18 of 46
 t
ip
cr
us
an
Figure 8: Volume changes over the cure cycles L155, M155 and H155, and qgel from table
M
3.

experiment (dotted red line) on the other hand, the adhesive reaches the pre-
d

set cure temperature essentially in an uncured state. The experiment with

te

the medium heating rate shows a behaviour in between these two extremes.
The difference in cure evolution does not only affect the shrinkage of the
p

adhesive, it also influences the gelation process. While in cycles with medium
ce

or high heating rates the gel point is reached in the isothermal phase at cure
temperature, gelation takes place at much lower temperatures for the L170
Ac

and L155 experiments.

4.3. Mechanical properties

Fig. 9 displays the DMA results in the form of storage modulus curves for
the five different load frequencies versus temperature. By applying the time-
temperature superposition principle, a master curve for the storage modulus
can be created (see fig. 10). In fig. 11, the shift factors used are plotted

19

Page 19 of 46
 t
ip
cr
us
an
M
d
te

Figure 9: The storage modulus E 0 for different frequencies versus temperature.

versus the temperature at which the respective partial curve was measured.
p

The function in eq. 19 with the fitted parameters in table 7 is depicted as

ce

well. Table 8 and 9 list the obtained shear moduli, relaxation times and the
bulk modulus.
Ac

Table 7: Shift factor parameters.

C1 (-) C2 (K) A (J/mol)
7.855 30.56 362900

20

Page 20 of 46
 t
ip
Table 8: Shear moduli and relaxation times.

cr
τm (s) Gm (MPa) τm (s) Gm (MPa)
1E-10 3.228E+01 1E-02 2.535E+01

us
1E-09 1.415E+01 1E-01 2.383E+01
1E-08 2.274E+01 1E+00 1.953E+01

an
1E-07 1.847E+01 1E+01 1.061E+01
1E-06 1.899E+01 1E+02 3.776E+00
1E-05 1.778E+01 1E+03 1.391E+00
M
1E-04 1.822E+01 1E+04 2.307E-03
1E-03 2.126E+01 1E+10 3.254E-02
d
pte
ce
Ac

Table 9: Bulk and rubbery shear moduli.

G∞ (MPa) Gfinal (MPa) K (MPa)
1.608 1.608 2463

21

Page 21 of 46
 t
ip
cr
us
an
Figure 10: Master curve for the reference temperature.
M
d
p te
ce
Ac

Figure 11: The shift factor from DMA data and fitted curves used in the model.

22

Page 22 of 46
4.4. Experiment versus simulation

The simulation results presented here were obtained with the model de-

t
scribed in section 3.2 and the values in table 3, 5, 6, 7, 8 and 9. Displacement

ip
curves depict the difference in out-of-plane displacement of node N1 and N2 :

cr
∆uy = uy (N2 ) − uy (N1 ) (see fig. 3). For better comparison, all curves are
shifted on the time scale so that cooling down begins at t = 0 min.

us
Fig. 12 shows the simulated and measured displacement of the steel strip
for the cycles with a cure temperature of 170 ° C. Curves for the medium and

an
M
d
p te
ce

Figure 12: Measured (solid lines) and predicted (dashed) displacement of the steel strip
for the cycles with 170 ° C as cure temperature.
Ac

high heating rate (blue and red line) resemble each other: the curves start
to drop late in the heating phase or early in the holding phase to a plateau
level where they remain until the end of the isothermal phase. The second
drop sets in with the cooling down. Predictions and experiments show good
agreement for these curves. The simulation somewhat underestimates the
first decrease, whereas it overestimates the second. The fact that the H170-

23

Page 23 of 46
curve shows less absolute final displacement than the M170-curve, however,
is not reflected in the simulation results.

t
The simulation model allows further insights in the cure process. Fig. 13

ip
shows the development of the strain component εth for the steel, aluminium

cr
and the epoxy, and εch and εch + εth for the adhesive. Furthermore, the
black vertical line marks the gel point. Strain before the gel point cannot

us
cause permanent shear stress and, consequently, no displacement of the strip.
The gel point is reached in the isothermal phase; thermal strain components

an
do not change in that phase. Hence, the first drop in strain of the M170-
and the H170-curve in fig. 12 is caused by the chemical strain component
alone. As soon as εch fades out, no further displacement is seen. Strain in
M
the H170-cycle shows similar behaviour; the only difference is that the gel
point is passed later in the holding phase. The second drop in displacement
d

in all cycles is caused by the thermal shrinkage of the adhesive during cooling
te

down, which exceeds that of the steel and aluminium.

The steel strip’s slow rise early in the heating phase (continuous black
p

line in fig. 12) is not to be found in the corresponding simulation curve.

ce

From about t = −60 min on, however, both curves show similar shapes: a
down-up movement before the cure temperature is reached, a constant level
Ac

above zero during the isothermal phase and a larger drop during cooling.
The predicted strain contributions in fig. 13 can help to interpret the
experimental low heating-rate curve L170 in fig. 12. The chemical strain
development takes place entirely in the long heating phase. There, it causes
first an overall shrinking of the adhesive (mechanical strain neglected) and,
consequently, the up-down movement of the steel strip. As displayed already

24

Page 24 of 46
 t
ip
cr
us
an
Figure 13: Predicted strain development in cycle M170 (top left), M155 (top right), L170
(bottom left) and L155 (bottom right).
M
in fig. 7, the gel point is reached during heating. The development of εch +εth
(green line) indicates that the adhesive tends to occupy more space in the
d

holding phase than at the gel point. That explains why the steel strip deflects
te

in the upwards direction during the isothermal phase (see fig. 12).
The experimental M155- and H155-curve in fig. 14 show, similar to the
p

170 ° C-curves, two distinguished, sudden drops in displacement. The simu-

ce

lation results reflect that behaviour, but underestimate the depth of the first
drop and, therefore, the absolute final displacement. The observed difference
in displacement between the M155- and H155-cycle is not predicted by the
Ac

simulation.
The L155-curve has a less pronounced first drop than the M155- and
H155-curve, which also starts earlier in the cure cycle. The simulation re-
produces that, but with smaller depth and with a few minutes delay. Also,
the predicted first drops in the cycles M155 and H155 set in later than the

25

Page 25 of 46
 t
ip
cr
us
an
Figure 14: Measured (solid lines) and predicted (dashed) displacement of the steel strip
for the cycles with 155 ° C as cure temperature.
M
ones observed. This might be due to deviations of the simulated gelation
process from the actual one in the experiment. As in cycles with a cure tem-
d

perature of 170 ° C, the first drop in all 155 ° C-curves is linked to the strain
te

components εch + εth (see fig. 13).

While both simulated L-curves show the same amount of final displace-
p

ment, the M170- and H170-graphs show more absolute final displacement
ce

than their 155 ° C-counterparts. The reason for that is the gel point, which
is reached at different temperatures for the M- and H-curves, but not for
Ac

the L-curves (see figs. 7 and 8). Only thermal strain from that point on
contributes to panel distortions.

5. Conclusions

In this work the development of panel distortions due to hot-curing adhe-

sives was simulated. To that end, a model incorporating the evolving prop-

26

Page 26 of 46
erties of an adhesive during cure, namely thermally and chemically induced
volume changes and mechanical characteristics, was implemented in Abaqus

t
subroutines. Material constants of a commercial, epoxy-based system were

ip
determined making it possible to compare predicted panel distortions to ex-

cr
perimental data.
The simulation reproduced many phenomena seen in experiments: the

us
steel strip deforms already before the cooling down sets in; in case of low
heating rate even before the cure temperature is reached. For the application

an
engineer, it is noteworthy that distortions can start to develop early in the
bonding process and are not just determined by the cure temperature. The
final displacement depends on the entire temperature cycle. The simulation
M
model can help making the bonding process more robust to panel distortions.
d
te

Acknowledgements

This research was carried out under project number M22.8.10384 in the
p

framework of the Research Program of the Materials innovation institute M2i

ce

(www.m2i.nl). The support of the industrial partner Tata Steel is greatfully

acknowledged. The authors thank Dow Automotive for providing material.
Ac

References

Adderley, C., 1988. Adhesive bonding. Materials & Design 9, 287 – 293.

Adolf, D., Martin, J.E., 1996. Calculation of Stresses in Crosslinking Poly-

mers. Journal of Composite Materials 30, 13–34.

27

Page 27 of 46
Barnes, T., Pashby, I., 2000. Joining techniques for aluminium spaceframes
used in automobiles: Part II - adhesive bonding and mechanical fasteners.

t
Journal of Materials Processing Technology 99, 72 – 79.

ip
Chudaska, A., Hahn, O., 1993. Untersuchungen zum temperatur- und

cr
reaktionsbedingten Klebstoffschrumpf - Maßnahmen zur Reduzierung
schrumpfbedingter Bauteilschädigungen, in: Wärmearme Fügetechniken

us
- Kleben - Durchsetzfügen - Nieten, pp. 121 – 130.

an
De Vreugd, J., Jansen, K.M.B., Ernst, L.J., Bohm, C., 2010. Prediction
of cure induced warpage of micro-electronic products. Microelectronics
Reliability 50, 910 – 916.
M
De Vreugd, J., Jansen, K.M.B., Ernst, L.J., Pijnenburg, J.A.C.M., 2007.
Modelling of viscoelastic properties of a curing adhesive, in: WIT Trans-
d

actions on Engineering Sciences, pp. 241–251.

te

Fuchs, H., Fernholz, K.D., Deslauriers, P., 2010. Predicted and Measured
p

Bond-Line Read-Through Response in Composite Automotive Body Pan-

ce

els Subjected to Elevated Temperature Cure. Journal of Adhesion 86,

982–1011.
Ac

Gromala, P., Duerr, J., Dressler, M., Jansen, K.M.B., Hawryluk, M.,
De Vreugd, J., 2011. Comprehensive material characterization and method
of its validation by means of fem simulation, in: 2011 12th Int. Conf. on
Thermal, Mechanical and Multi-Physics Simulation and Experiments in
Microelectronics and Microsystems, EuroSimE 2011.

28

Page 28 of 46
Hahn, O., Chudaska, A., Hüsgen, B., 1993. Influence of the process parame-
ters on the quality of thermosetting adhesive joints. Welding in the World,

t
Le Soudage Dans Le Monde 31, 277–280.

ip
Hahn, O., Jendrny, J., 2003. Evaluation of simulation models for the es-

cr
timation of deformation of adhesively bonded steel sheets during curing.
Welding in the World 47, 31 – 38.

us
Jansen, K.M.B., Qian, C., Ernst, L.J., Bohm, C., Kessler, A., Preu, H.,
Stecher, M., 2007. Kinetic characterisation of molding compounds, in: Eu-

an
roSime 2007: International Conference on Thermal, Mechanical and Multi-
Physics Simulation Experiments in Microelectronics and Micro-Systems,
M
2007.

Kamal, M., Sourour, S., 1973. Kinetics and thermal characterization of

d

thermoset cure. Polymer Engineering and Science 13, 59–64.

te

Li, C., Potter, K., Wisnom, M.R., Stringer, G., 2004. In-situ measurement of
chemical shrinkage of MY750 epoxy resin by a novel gravimetric method.
p

Composites Science and Technology 64, 55 – 64.

ce

Liebl, C., Johlitz, M., Yagimli, B., Lion, A., 2012. Three-dimensional chemo-
thermomechanically coupled simulation of curing adhesives including vis-
Ac

coplasticity and chemical shrinkage. Computational Mechanics 49, 603–

615.

Lion, A., 2007. Einführung in die Lineare Viskoelastizität.

Lion, A., Höfer, P., 2007. On the phenomenological representation of curing

phenomena in continuum mechanics. Archives of Mechanics 59, 59–89.

29

Page 29 of 46
López, J., Ramı́rez, C., Torres, A., Abad, M., Barral, L., Cano, J., Dı́ez, F.,
2002. Isothermal curing by dynamic mechanical analysis of three epoxy

t
resin systems: Gelation and vitrification. Journal of Applied Polymer

ip
Science 83, 78–85.

cr
Menzel, S., 2011. Zur Berechnung von Klebverbindungen hybrider
Karosseriestrukturen beim Lacktrocknungsprozess. Ph.D. thesis. Technis-

us
che Universität Dresden.

an
Patankar, K.A., Dillard, D.A., Fernholz, K.D., 2013. Characterizing the
constitutive properties and developing a stress model for adhesive bond-
line readout. International Journal of Adhesion and Adhesives 40, 149–157.
M
Priesnitz, K., Sinke, J., Benedictus, R., 2014a. Influence of the tempera-
ture cycle on local distortions in car panels caused by hot curing epoxies.
d

International Journal of Adhesion and Adhesives 50, 216 – 222.

te

Priesnitz, K., Sinke, J., Benedictus, R., 2014b. On the simulation of panel
p

distortions due to hot curing adhesives. International Journal of Solids and

ce

Structures 51, 2470 – 2478.

Sadeghinia, M., Jansen, K.M.B., Ernst, L.J., 2012. Characterization and

Ac

modeling the thermo-mechanical cure-dependent properties of epoxy mold-

ing compound. International Journal of Adhesion and Adhesives 32, 82–88.

Starink, M., 2003. The determination of activation energy from linear heating
rate experiments: a comparison of the accuracy of isoconversion methods.
Thermochimica Acta 404, 163 – 176.

30

Page 30 of 46
Williams, M., Landel, R., Ferry, J., 1955. The temperature dependence of
relaxation mechanisms in amorphous polymers and other glass-forming

t
liquids. Journal of the American Chemical Society 77, 3701–3707.

ip
cr
us
an
M
d
p te
ce
Ac

31

Page 31 of 46
 M
Figure

ed
pt
ce
Ac

Page 32 of 46
Figure

ed
pt
ce
Ac
Page 33 of 46
Figure

ed
pt
ce
Ac
Page 34 of 46
 M
Figure

ed
pt
ce
Ac

Page 35 of 46
 u
Figure

an
M
ed
pt
ce
Ac

Page 36 of 46
Figure

ed
pt
ce
Ac
Page 37 of 46
 M
Figure

ed
pt
ce
Ac

Page 38 of 46
Figure
T I II III

c e
Ac
Page 39 of 46
t
Gel point Glass transition
 u
Figure

an
M
ed
pt
ce
Ac

Page 40 of 46
Figure

ed
pt
ce
Ac
Page 41 of 46
Figure Adhesive

M
Symmetry planes bond line
Steel strip

d
Aluminum
substructure

te
Node N2
Node N1

cep
Ac
• 2 edges
• Unconnected
• uy=0
Page 42 of 46
Figure

ed
pt
ce
Ac
Page 43 of 46
Figure
1

e
v

pt
0 ce 3
2
Ac
4

T0 Tg Tc T
Page 44 of 46
 M
Figure

ed
pt
ce
Ac

Page 45 of 46
 M
Figure

ed
pt
ce
Ac

Page 46 of 46