Journal Pre-proof

Geometrical optimization of adhesive joints under tensile impact loads using cohesive
zone modelling

J.P.A. Valente, R.D.S.G. Campilho, E.A.S. Marques, J.J.M. Machado, Lucas F.M. da
Silva
PII: S0143-7496(19)30241-6
DOI: https://doi.org/10.1016/j.ijadhadh.2019.102492
Reference: JAAD 102492

To appear in: International Journal of Adhesion and Adhesives

Please cite this article as: Valente JPA, Campilho RDSG, Marques EAS, Machado JJM, da Silva LFM,
Geometrical optimization of adhesive joints under tensile impact loads using cohesive zone modelling,
International Journal of Adhesion and Adhesives, https://doi.org/10.1016/j.ijadhadh.2019.102492.

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. 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.

© 2019 Elsevier Ltd. All rights reserved.

 Geometrical optimization of adhesive joints under tensile impact
loads using cohesive zone modelling

J.P.A. Valente1, R.D.S.G. Campilho1,2∗∗, E. A. S. Marques2, J. J. M. Machado2, Lucas F. M. da

Silva3
1
Departamento de Engenharia Mecânica, Instituto Superior de Engenharia do Porto (ISEP), Rua Dr. António
Bernardino de Almeida, 431, 4200-072 Porto, Portugal
2
Instituto de Ciência e Inovação em Engenharia Mecânica e Engenharia Industrial (INEGI), Rua Dr. Roberto
Frias, 4200-465 Porto, Portugal
3
Departamento de Engenharia Mecânica, Faculdade de Engenharia (FEUP), Universidade do Porto, Rua Dr.
Roberto Frias, 4200-465 Porto, Portugal

Abstract

Industrial developments have led to an increasingly wide implementation of adhesive

bonding. Due to the variability of adhesive bonding performance caused by different adhesive
properties, it is necessary to validate Finite Element Method (FEM) tools. It is possible to
increase the performance of adhesive joints when subjected to impact loadings, without
making complex design changes, with the variation of geometric parameters or by modifying
the adherends’ geometry. This work compares the results of different geometric changes
applied to a single-lap joint (SLJ), when subjected to impact, through Cohesive Zone Models
(CZM). Geometry modifications of the SLJ are made by introducing outer and inner chamfers
into the adhesives, as well as adding adhesive fillets, to observe the effects of these
modifications with different types of adhesives. The combination of the geometric changes
that produce the best result is subsequently made. As a result of this work, the CZM technique
was validated for the impact strength prediction of adhesive joints and the optimal joint
geometries were defined as a function of the adhesive.

Keywords

Adhesive joint; Cohesive zone models; Geometrical optimization; Impact loading.

∗
Corresponding author. Tel: +351939526892. Fax: +351228321159. Email: [email protected]

1
1. Introduction

Vehicle manufacturers are in a constant search for lighter and higher performing structural
construction methods [1], enabling the creation of structures with lower weights and higher
strength, possessing the capability to undergo severe dynamic loads without catastrophic
failure. Adhesive bonding is a key joining technique for achieving this type of performance,
but it requires and extensive study of the effect of impact loads on the joint strength [2]. In
fact, many of the materials used in these structures are known to have some degree of strain
rate sensitivity which, when in conjunction to the inertial loads derived from large
accelerations, causes large differences in adhesive behaviour. To design bonded structures
under these conditions, it is therefore important to study in detail the strain rate dependency of
adhesives and adherends, and to use this knowledge to assess the behaviour of the complete
joint [3-5]. In SLJ there should ideally exist a uniform shear stress in the adhesive layer, to
provide a maximum joint efficiency [6, 7]. Such behaviour is therefore ideal and rarely
achieved in real structures, once stress concentrations can arise from: differential straining in
the adherends, referring to the shear-lag effect; bending introduced by the non-axial loading;
and end effects caused by the free surfaces at the edges of the adhesive layer. Differential
straining in the adherends gives rise to a non-uniform shear stress distribution in the adhesive,
with the maximum shear stress occurring at the ends of the overlap [7, 8].

Hart-Smith [9] indicated that, when using composite adherends, the design process of bonded
joints should consider chamfered ends to prevent the generation of peel stresses, without any
loss of shear strength. Vallée and Keller [10] performed experimental and numerical studies
(quadratic shear-tensile failure criterion) on the effect of outer chamfers in the pultruded glass
fibre reinforced polymer (GFRP) adherends of double-lap joints (DLJ), observing a reduction
of through-thickness tensile and shear stress peaks towards the chamfered joint edges.
However, the joint strength was not significantly improved by this geometrical modification
of the adherends. The influences of both the outer and inner chamfer angle at the free end of
adherend on the stress distribution in aluminium single-lap crippling joints were individually
investigated by You et al. [11] by the use of an elastic-plastic FEM formulation. Considering
the outer chamfer configuration, the results showed that the different stress distribution along
mid-bondline would not vary significantly, and the peak value of peeling stresses increased
with the increase of the angle when the free end of the adherends was chamfered. For the
inner chamfer configuration, the peak value of longitudinal stress and the shearing stress in
the end-bondline were lower than the ones found for the outer chamfer configuration,

2
although the peak value of peeling stress was higher for the inner chamfer configuration.
Apart from these chamfer modifications, real adhesive joints are usually formed with a fillet
of adhesive spew which is squeezed out under pressure while the joint is being manufactured.
The assumption that the adhesive layer has a square shape end is thus unlikely to be realistic.
Mylonas [12], using a photo elastic technique, has investigated the stresses induced at the end
of an adhesive layer for several adhesive edge shapes, and has shown that the position of the
maximum stress is dependent on the edge shape. Tsai et al. [13] have also demonstrated that
the spew fillet has a drastic effect in the stress distribution of the adhesive.

Several other authors explored the FEM to model impact loadings in adhesives and adhesive
joints, usually employing constitutive models to represent the strain rate sensitive behaviour
of the adhesives and adherends [14]. For example, Sawa et al. [15], in 2003, researched the
influence of stress-wave propagations and stress distributions in SLJ using a model with
dynamic effects, identifying the locations of the maximum principal stress. An alternative
method to establish constitutive models is the use of CZM, which combines the strength of
materials and fracture mechanics approaches [16]. These models are especially well-suited for
adhesive joints, as they work best for simulating failure in thin planes of materials [17].
Traction-separation laws, introduced for each material, are used to model stiffness and
degradation of the element. These laws can exhibit several different shapes, the most common
being the triangular and the trapezoidal laws. Although the use of CZM is extensive in
adhesive modelling, the work of Carlberger and Stigh, published in 2007 [18], was the first to
demonstrate the validity of using this type of approach to predict impact strength. Other
authors, such as Haufe et al. [19], May et al. [20], Clarke et al. [21], Avendaño et al. [22] and
Neumayer et al. [23] have used this approach and achieved accurate prediction of failure loads
using complex dynamical cohesive models with strain rate dependent data.

This work compares the results of different geometric changes applied to a SLJ, when
subjected to impact, through CZM. Geometry modifications of the SLJ are made by
introducing outer and inner chamfers into the adhesives, as well as adding adhesive fillets, to
observe the effects of these modifications with different types of adhesives. The combination
of the geometric changes that produce the best result is subsequently made.

3
2. Numerical model

2.1. Model configuration

The models were created using Abaqus® (Dassault Systèmes, Suresnes, France) to obtain the
load-displacement curves (P-δ) and a prediction of the maximum load (Pm). All models
created use CZM, which can represent the fracture process and location, advancing beyond
the classical continuum mechanics modelling. The numerical models for quasi-static
simulations typically consist of an implicit model, computed in a general/quasi-static analysis,
with cohesive elements applied to the bondline. To simulate the impact conditions, the impact
models followed the same approach, consisting of a deformable part with cohesive modelling
of the adhesive layer. Mesh element types were changed to explicit elements with the same
controls as for the quasi-static simulation. The explicit type of formulation enables the model
to consider the inertial effects, which can be an important factor for the joint strength. Also,
the material properties were altered to match the strain rate dependent material properties. To
replicate the impact’s kinetic energy, a 26 kg mass block (the same mass used in the
experimental tests) was added to the last 25 mm length of the free adherend, as shown in
Figure 1. The main geometric parameters are the total length between grips (LT), overlap
length (LO), adherends’ thickness (tP) and adhesive thickness (tA). x is the horizontal
coordinate beginning at the leftmost overlap edge. The volume of the block is 1250 mm3 and
to simulate an impact with 38.3 J of energy and an initial velocity of 1.75 m/s, the density was
set to 2.08E+7 kg/m3. A predefined velocity field was applied to this mass, replacing the 4
mm of horizontal displacement boundary condition used in the quasi-static simulation.

Figure 1 – Model geometry used for impact conditions.

For the impact analysis, the quasi-static step was replaced with a dynamical explicit
calculation with a period of 0.005 seconds and considering non-linear geometry effects.
Element stiffness degradation as well as the reaction forces and displacements of the
boundary condition nodes were requested to extract the P-δ curves and determine the status of
the cohesive elements.

4
2.2. Cohesive zone modelling

Figure 2 shows the pure-mode (traction or shear) and mixed-mode traction separation laws,
where tn0 and ts0 are the cohesive strengths in tension and shear, respectively, and δn0 and δs0
are the respective displacements, and δnf and δsf are the tensile and shear displacements at
failure, respectively.

Figure 2 – Traction-separation law with linear softening: pure and mixed-mode models.

The cohesive elements used for this work employ a quadratic stress criterion for representing
damage onset in mixed-mode. The quadratic stress criterion is given by Equation 1.

〈 〉
1 Eq. (1)

tn and ts are the current cohesive stresses in tension and shear, respectively, and 〈〉 are the
Macaulay brackets, emphasizing that a purely compressive stress state does not initiate
damage. After the mixed-mode cohesive strength is attained (tm0 in Figure 2) by the fulfilment
of Equation 1, the material stiffness is degraded. Complete separation is predicted by a linear
power law form of the required energies for failure in the pure modes

1 Eq. (2)

GI and GII are the current strain energy release rates in each loading mode, while GIC and GIIC
are the respective critical values.

2.2. Materials and properties

2.2.1. Adhesives properties

Three distinct adhesives were considered in this work, namely a stiff (Araldite® AV138), a
ductile (3M® DP8005), and a high toughness (Nagase Chemtex® XNR6852 E-2) adhesive.

5
These adhesives were selected with the aim of assessing the behaviour of a SLJ under impact
condition using adhesives with very different mechanical properties.

In Table 1 it is possible to observe the mechanical properties for the three adhesives
considered in this work, determined at quasi-static conditions (ρ is the specific weight and ν
the Poisson’s coefficient). The values of Young’s modulus (E) and tn0 were experimentally
determined by performing tensile bulk tests; and ts0 by performing thick adherend shear tests
(TAST). The shear modulus (G) was estimated by the Hooke’s law using E and ν. GIC and
GIIC were respectively determined by double-cantilever beam (DCB) and end-notched flexure
(ENF) tests.

Table 1 – Mechanical properties of the three adhesives used in quasi-static conditions.

AV138 [24] DP8005 [25, 26] XNR6852 E2 [27]

E (MPa) 4890 590 1742
G (MPa) 1811 219 622
tn0 (MPa) 41 6.3 42.9
ts0 (MPa) 30.2 8.4 28.7
GIC (N/mm2) 0.35 1.1 1.68
GIIC 0.6 6 18
2
(N/mm )
(g/cm3) 1.7 1.06 1.5
υ 0.35a 0.35b 0.4c
a
manufacturer value; b from reference [25]; c typical value for epoxy adhesives [28].

The mechanical properties at different testing velocities were previously determined by Silva
et al. [27] and can be found in Table 2. The mechanical properties presented for 105000
mm/min (corresponding to an initial impact velocity of 1.75 m/s) were determined by a
logarithmic extrapolation [18] by using the existing experimental values.

6
 Table 2 – Mechanical properties of the three adhesives used at high strain rates [27].

Adhesive Strain rate tn0 (MPa) ts0 (MPa) GIC GIIC

(mm/min) (N/mm) (N/mm)
1 41 30.2 0.35 4.949.1
AV138 100 49.1 36.2 - -
105000 70.2 * 51.7 * 0.35 0.6
1 6.3 8.4 1.1 6
DP8005 10 13 17.4 - -
105000 27.5 36.7 1.1 6
1 42.9 28.7 1.68 18
XNR6852 E2 10 46 33.6 - -
105000 53.7 45.8 1.68 18
*Extrapolated values.

2.2.2. Adherend properties

To achieve maximum strength of the adhesive and minimize the influence of adherends’
deformation, high strength steel adherends (DIN 55 Si7) were considered, to ensure the
absence of plastic deformation during testing. Thus, for the numerical model, the plastic
deformation properties were not considered. The measured properties of this material are as
follows [27]: E=210 GPa, tensile yield stress σy=1078 MPa, tensile strength σf=1600 MPa,
tensile failure strain εf=6%, ρ=7.8 g/cm3 and ν=0.3.

2.3. Model validation process

To validate the model, experimental and numerical data published in a previous work [27]
was compared with the results obtained with the numerical models developed for this work.
Comparisons were performed for the three adhesives under study.

2.3.1. Joint geometry

In the previously performed experimental tests, SLJ based on the ASTM D1002 and ISO
4587 standards were used (Figure 3). To directly compare with the experimental values
obtained [27] the same joint geometry was considered in the numerical simulations performed
for validation purposes. The values of LO=25 mm and tA=0.2 mm were kept constant with the
introduction of outer chamfers.

7
 Figure 3 - Geometry of the SLJ used (dimensions in mm).

2.3.2. Dynamic testing procedure

The experimental impact tests (used as a reference) were performed in a Rosand®
Instrumented Falling weight impact tester, type 5 H.V. (Stourbridge, West Midlands, U.K),
with a 60 kN load cell. The specimens were tested at a speed of 1.75 m/s and a mass of 26 kg
was used for the impactor, corresponding to an impact energy of 40 J, which was found to be
enough to break all specimen configurations. Although no correction is made for the inertial
effects, the load cell is located directly between the impactor tip and the added mass, with
only the much smaller mass of the impactor contributing to the inertial error.

2.3.3. Validation results

This Section aims at validating the proposed numerical methodology for the impact strength
prediction of bonded joints. Figure 4 (a) presents the validation results for the three tested
adhesives. This study consists of comparing the values of Pm and displacement at which Pm is
attained (δPm) between experimental data [27], numerical simulations from a previous
reference (addressed as numerical reference) [27] and numerical simulations carried out in the
present work (addressed as current numerical).

30 3
26.92 28.39 2.63
25 24.00 2.5
19.93 19.90 2.53
20 2
δ Pm [mm]
Pm [kN]

16.79 14.43 1.56

15 13.61 1.5
11.75 0.64 0.67
10 1
0.59
5 0.5 0.58
0.31 0.27
0 0
Experimental Numerical Current numerical Experimental Numerical Current numerical
reference reference
AV138 DP8005 XNR6852 E-2 AV138 DP8005 XNR6852 E-2
a) b)
Figure 4 - Pm (a) and δPm (b) values obtained in the experiments [27], numerical simulations from a previous reference
(numerical reference) [27] and in this work (current numerical).

8
Compared to the experimental data, the numerically obtained Pm values for the three
adhesives always revealed to be higher, either for the numerical reference or current
numerical values (Figure 4 a). The maximum deviations to the experimental data, by
adhesive, were 22.8% for the AV138 (numerical reference), 18.7% for the DP8005
(numerical reference) and 18.3% for the XNR6852 E-2 (current numerical). It is considered
that the lower experimental Pm, compared to both numerical results, is acceptable given the
fact that CZM modelling does not account for imperfections induced during fabrication nor
slight misalignments in the testing machine that can occur in the experiments and
consequently affect the results by altering the theoretical optimal testing conditions. The
current numerical data (achieved during the course of this work) slightly deviated from the
numerical reference, and it was possible to identify the following discrepancies compared to
the numerical reference: smaller Pm by 5.7% for the AV138, smaller Pm by only 0.1% for the
DP8005, and higher Pm by 5.5% for the XNR6852 E-2. The differences between the two
numerical results have some significance despite all the efforts to reproduce the numerical
simulations of reference [27], and this can be related to inconsistencies of mesh refinement
and topology, increment time and frequency of data acquisition.

Figure 4 (b) summarizes the δPm attained in the experimental tests and both numerical
simulations as a function of the adhesive. The obtained results show that significant δPm
differences were found between the experiments and both numerical analyses, in the sense
that the numerical reference and current numerical δPm values were much smaller than those
obtained in the experiments. The maximum deviations to the experimental condition, for each
adhesive, were 82.4% for the AV138 (current numerical), 76.9% for the DP8005 (numerical
reference) and 75.6% for the XNR6852 E-2 (numerical reference). Thus, notwithstanding the
numerical data, the experimental δPm values are always considerably higher. This
discrepancy, which is not negligible, is considered to occur due to errors introduced by
integration of the accelerometer’s data, leading to an inflated experimental δ. Adding to this,
the testing machine’s compliance may also be responsible for δ measuring discrepancies over
the numerical results, in which the specimen is directly loaded at its ends. On the other hand,
in opposition to Pm, the differences between the numerical reference and numerical results are
negligible. The maximum difference was attained for the AV138, corresponding to the current
numerical δPm giving a negative deviation of 11.8% over the numerical reference.
Nonetheless, these small differences can be due to slight differences in the numerical input
conditions and captured data frequency.

9
2.4. Specimen configurations under study

2.4.1. Model configurations

To perform the geometrical optimization, three different geometrical characteristics were
considered for this study: outer chamfer, inner chamfer and adhesive fillet. Within each of
these cases, different angles were considered, keeping LO constant. Table 3 shows these
configurations, detailing the key geometrical dimensions and the angles under study. The
considered geometrical variables are the outer chamfer angle (α), inner chamfer angle (β) and
adhesive fillet angle (θ).

Table 3- Specimen configurations under study.

Angle
Designation of
Graphical representation and key geometrical dimensions (in values
the
mm) under
configuration
study

Outer chamfer

7.5º
15º
30º
Inner chamfer
45º
60º
90º

Adhesive fillet

10
2.4.2. Model meshes
Representative meshes for three model configurations under study are shown in Figure 5.

Outer
chamfer

Inner
chamfer

Adhesive
fillet

Figure 5 – Representative meshes for the models under study.

In these meshes, the adhesive region is composed by sweep elements of type COH2D4. The
adherends are fully composed by structured elements. In the triangular areas, elements of
CPE3 type are implemented, as for the rest of the elements are of CPE4R type. For all the
models varying the outer chamfer, a uniform mesh was initially used with the same
parameters of the one used in the validation models. Thereafter, in sections with slopes,
vertical subdivisions of 0.2 mm of spacing were created. In vertices created with the exterior
limits, vertical lines were introduced to refine the mesh and allow the reduction of the
implementation of triangular elements. For all the models varying the outer chamfer, a
uniform mesh was initially used with the same parameters of the one used in the validation
models, using triangular elements. Due to errors obtained during the simulations owing to
excessive deformation of adhesive, the results for θ=7.5° were not included in the case of
using the flexible adhesive (DP8005) and the tough adhesive (XNR6852 E-2).

3. Results

3.1. Outer chamfer

3.1.1. Stiff adhesive (AV138)

Figure 6 shows Pm and δPm for the AV138. The highest Pm occurred for the lowest values of
α. The best Pm result, obtained for α=7.5º, is 6% higher in comparison with the value obtained
for the SLJ without chamfer (α=90°). Due to the brittleness of this adhesive, it is expected

11
that Pm is ruled by peak stresses [29]. The reduced Pm improvement showed that the outer
chamfer does not affect stress distributions by a large amount.

Figure 6 – Pm and δPm for the joints with outer chamfer bonded with the Araldite® AV138 as a function of α.

The value of δPm experiences small variations along the tests and suffers a reduction for
α=7.5°. The values of energy absorbed by SLJ using the Araldite® AV138 was quite similar
for all angles under study, ranging from 2.01 J for α=7.5º to 2.03 J for α=90º.

3.1.2. Flexible adhesive (DP8005)

Figure 7 shows Pm and δPm for the SLJ with outer chamfer, bonded with the DP8005.

Figure 7 – Pm and δPm for the joints with outer chamfer bonded with the DP8005 as a function of α.

12
Considering α between 15°≤α ≤90°, there are no significant Pm differences, although there are
some small variations. For the SLJ with α=7.5°, in comparison with α=90°, a 22% Pm
increase was found, which mainly reflects the reduction of peel peak stresses that usually
occur by the outer chamfering technique [11]. As for δPm, no significant changes were
observed, except for the cases where α is of 7.5° or 60°, where such value is marginally
higher. The δPm for α=7.5º is also higher when compared with the other tests performed. The
values of energy absorbed by SLJ using the flexible adhesive (DP8005) exhibited a larger
variation as a function of α, ranging from 6.49 J for α=90º to 8.43 J for α=7.5º.

3.1.3. Tough adhesive (XNR6852 E-2)

In Figure 8 it is possible to analyse Pm and δPm in more detail for the XNR6852 E-2.

Figure 8 – Pm and δPm for the joints with outer chamfer bonded with the XNR6852 E-2 as a function of α.

Identically to the Araldite® AV138, there is a clear tendency for the Pm increase with the
reduction of α. The introduction of a chamfer with α=7.5° results in an increase of Pm of
11%, in comparison with the values obtained for SLJ with α=90° (without chamfer). On the
other hand, δPm tends not to be affected by α except for α=7.5°, in which a slight increase
was found. The differences registered, in terms of Pm and δPm, are lower than the ones
observed when using the stiff adhesive (AV138), which is due to the lower stiffness of the
tough adhesive (XNR6852 E-2). The values of energy absorbed by SLJ using the tough
adhesive (XNR6852 E-2) showed a steady increase by reducing α, from 11.03 J (90º) to 11.96
J (7.5º).

13
3.2. Inner chamfer

3.2.1. Stiff adhesive (AV138)

Figure 9 addresses the inner chamfer effect, by presenting Pm and δPm for the SLJ bonded
with the AV138 as a function of β.

Figure 9 – Pm and δPm for the joints with outer chamfer bonded with the Araldite® AV138 as a function of β.

It is possible to observe an increase of Pm with the reduction of β up to 15°, followed by a

more pronounced reduction for β=7.5°. Comparing to the SLJ with β=90° (without chamfer),
a Pm increase of 14% was found for β=15°. For β=7.5°, a decrease of the SLJ performance
was measured, which was due to the less transferred loads due to induced compliance at the
overlap due the extremely small β. The evolution of δPm with β closely follows that of Pm, for
the same reason previously pointed. Although the difference in energy absorbed between
different β was found to be small or even null between the tests performed, for the tests with
β=15° and β=30° an increase of 4% was found over SLJ without chamfer (β=90°). The cause
of such increase is the higher Pm and δPm for the respective tests. The absorbed energy was
similar between limit values of β (90º and 7.5º), at 2.03 J, and slightly increased for
intermediate β values, up to 2.12 J for β=15º and 30º.

3.2.2. Flexible adhesive (DP8005)

The Pm and δPm results for the DP8005 (Figure 10) show a Pm improvement with the
introduction of an inner chamfer with β=60° when compared to the absence of chamfer
(β=90°), although limited to 2%. However, from this point on, the β reduction causes a

14
gradual degradation of the joints’ performance, with emphasis to β=7.5º, which suffers a
significant depreciation of Pm from the β=15º condition (32%).

Figure 10 – Pm and δPm for the joints with outer chamfer bonded with the DP8005 as a function of β.

The small variation between the case of SLJ without chamfer (β=90°) and β=45° is because
of identical load transmitting mechanisms between these joints. The significant reduction for
the test with β=7.5° results from the reduction of transmitted loads along the chamfered
portion of the overlap. Compared with the AV138, the expected peak stress reduction
achieved by the inner chamfer [11] was not relevant due to the ductility of the DP8005 and,
here, the stiffness reduction at the chamfered portion of the overlap induced a reduction of
transferred loads through the adhesive. The δPm results followed the same tendency. The
absorbed energy for the base geometry of β=90º was 6.49 J, then it slightly increased up to
6.92 J for β=30º, and following it reduced again, with a major drop from 15º to 7.5º, attaining
the value of 3.69 J for 7.5º.

3.2.3. Tough adhesive (XNR6852 E-2)

In Figure 11 it is possible to analyse Pm and dPm obtained for each model using the tough
adhesive (XNR6852 E-2). The trend for Pm is similar to that found for the flexible adhesive
(DP8005).

15
 Figure 11 – Pm and δPm for the joints with outer chamfer bonded with the XNR6852 E-2 as a function of β.

The introduction of the inner chamfer promotes a Pm increase up to β=45°, achieving an

increase of 5% in comparison with the case of no chamfer (β=90°). Further diminishing β
results in a gradually larger Pm reduction, up to 19% for β=7.5°. Justification for this
behaviour is much identical to that presented for the DP8005, i.e., the higher compliance at
the overlap, which diminishes the capability to transfer loads, while the expected stress peak
reductions are not so relevant for ductile adhesives. The energy absorbed by SLJ using the
tough adhesive (XNR6852 E-2), as a function of β, does not change significantly between 90º
(11.03 J) and 15º (10.43 J), but suffers a drastic reduction for the 7.5º (7.07 J), thus reducing
almost 36%.

3.3. Adhesive fillet

3.3.1. Stiff adhesive (AV138)

Figure 12 presents the Pm and δPm variation with θ for the AV138, showing that the reduction
of θ (corresponding to higher length fillets) is followed by an increase of Pm and δPm. The Pm
results are accordingly to the expected, once the reduction of θ results on a higher total length
of the adhesive layer available for load transfer between the adherends [30], besides the
known reduction of peak stresses at the overlap ends [31]. It was found that Pm increases
sensibly linearly with the θ reduction, up to 54% from 90º to 7.5º. δPm showed a different
behaviour, with no relevant variation between θ=90º and θ=30º, and with a subsequent
increase up to 52% compared to θ=90º for the joint with θ=7.5º.

16
 Figure 12 – Pm and δPm for the joints with adhesive fillet bonded with the AV138 as a function of θ.

For this joint configuration, the absorbed energy is strongly dependent on θ, ranging from a
minimum of 2.03 J for 90º to a maximum of 4.57 J for 7.5º. Due to the higher bearing load,
with higher failure displacements achieved, there is an increase of the energy absorbed during
impact conditions, resultant from the reduction of θ.

3.3.2. Flexible adhesive (DP8005)

In Figure 13 it is possible to analyse the Pm and δPm evolution with θ for the joints bonded
with the DP8005. For this adhesive, Pm also increased with the θ reduction, but the tendency
was not a straight line. The maximum improvement over θ=90º was 37% for θ=7.5º. Thus,
comparing with the AV138, the improvement was smaller, which is related to the higher
ductility of the adhesive, making it less prone to be affected by peak stresses [29]. The δPm-θ
plot followed an identical tendency to Pm, and a maximum δPm increase of 23% was found
for the smallest θ.

17
 Figure 13 – Pm and δPm for the joints with adhesive fillet bonded with the DP8005 as a function of θ.

The values of energy absorbed by the SLJ using the DP8005 are again very sensitive to the
value of θ. Actually, there is a strong increase of the energy absorbed by the SLJ with the
decrease of the fillet angle, ranging from 6.49 J for θ=90º to 11.59 J for θ=7.5º (percentile
improvement of 79%).

3.3.3. Tough adhesive (XNR6852 E-2)

The Pm and dPm evolution with q for the XNR6892 E-2 is presented in Figure 14. The same
overall trend observed for the other adhesives is encountered, with an increase of the two
variables by reducing θ. The Pm increase is nearly linear with the reduction of θ, up to an
increase of 39% over the reference value (θ=90°). Here, the difference between limiting θ
values is much similar to that of the DP8005 due to the identical ductile character of the
adhesive, and thus smaller to that of the AV138. δPm shows a similar tendency, although up
to θ=30º the difference is very small. However, it attains 24% for θ=7.5 comparing with the
base geometry.

18
 Figure 14 – Pm and δPm for the joints with adhesive fillet bonded with the XNR6852 E-2 as a function of θ.

As expected, as the Young´s modulus of the tough adhesive (XNR6852 E-2) is intermediate
between the stiff (AV138) and flexible (DP8005) adhesives, an increase of the energy
absorbed by the SLJ is registered with the increase of the adhesive layer by the presence of
adhesive fillets, being these values between the ones registered by the two other adhesives.
The values of energy absorbed by SLJ using the tough adhesive XNR6852 E-2 range from a
minimum of 11.03 J (θ=90º) to a maximum of 21.37 J (θ=15º).

4. Optimal combinations

In this section, combinations of singular geometrical optimizations are tested, being

considered the most suitable ones regarding each type of adhesive tested. Only combinations
possible to implement experimentally were performed. For example, it was not considered the
introduction of an adhesive fillet with an outer chamfer of the adherends, or even the
combination of outer and inner chamfers. The numerical results obtained are therefore
compared with the validation results and the most effective singular optimizations. As for the
mesh parameters, the control and material attribution are considered identical to the
previously analysed cases.

4.1. Stiff adhesive (AV138)

Using the stiff adhesive AV138, the most effective singular optimization is the use of an
adhesive fillet with θ=7.5°. For the combined configuration, this geometric modification was
used together with an inner adherend chamfer in the adherends, using β=15° as the best single

19
result. The numerical P-δ curves obtained for the different joint configurations using the stiff
adhesive (AV138) are presented in Figure 15, including the validation geometry without
geometric changes.

Figure 15 – P-δ curves for the different joint configurations using the stiff adhesive (AV138).

All P-δ curves show an identical behaviour at the initial stages of the tests, although the
curves diverge from approximately δ=0.3 mm. The conventional and inner chamfer (β=15º)
joints fail at this stage with a similar Pm, while the joints with a θ=7.5º adhesive fillet and
with the combined configuration further endure loads and fail only at a significant higher δ.
Between these two joint configurations, the configuration solely with the adhesive fillet
performs slightly better than the combined configuration. Figure 16 presents Pm and δPm for
all tested joint scenarios with this adhesive.

20
 Figure 16 – Pm and δPm for the different configurations using the stiff adhesive (AV138).

Between the individual modifications, the adhesive fillet is much more effective than the
inner chamfer. Actually, the optimal inner chamfer promoted a Pm improvement over the
conventional joint of only 14%, whilst this value increased to 54% for the optimal fillet. It is
clear from Figure 16 that the combined configuration did not work as expected, even
presenting a Pm value 41% higher to the conventional SLJ, since it is 9% lower than the
filleted joint with θ=7.5°. The δPm difference between joint configurations follows the Pm
behaviour, with the maximum value being found for the filleted joint.

The absorbed energies for the four joint configurations was as follows: 2.03 J for the
conventional joint, 2.12 J for the joint with β=15º, 4.57 J for the joint with θ=7.5 and 3.98 J
for the combined configuration. The maximum percentile improvement over the standard
joint was thus 125% for the joint with θ=7.5.

4.2. Flexible adhesive (DP8005)

The adhesive fillet with θ=15° was the singular optimization with higher Pm increase using
the flexible adhesive (DP8005). Through the comparison of all the results obtained in the
previous sections, it is concluded that an optimal configuration could be achieved with SLJ
with β=60° and θ=15°. The P-δ curves obtained for the different configurations using the
flexible adhesive (DP8005) are presented in Figure 17. It is possible to observe that, between
all tested joint configurations, the joint with θ=7.5º gives the best Pm results. Regarding δPm,

21
whilst the joint with β=60º shows no tangible difference, both the combined and θ=7.5º joints
reveal a significant improvement, which will also affect the absorbed energy.

Figure 17 – P-δ curves for the different joint configurations using the flexible adhesive (DP8005).

In Figure 18 it is possible to compare Pm and δPm for each tested model using the flexible
adhesive (DP8005). It is visible that the combined joint could not to improve the result for the
singular optimization of θ=7.5º, which yielded a 42% Pm improvement over the base joint.
Actually, it was 3% below the best solution. The tendency for δPm was mainly identical to
that found for Pm. Thus, the highest δPm was found for the joint with θ=7.5º, closely followed
by the combined joint.

Figure 18 – Pm and δPm for the different configurations using the flexible adhesive (DP8005).

22
Also for this adhesive, major absorbed energy variations were found between the four joint
configurations. The conventional joint absorbed 6.49 J of impact energy up to failure. This
value was virtually identical to the joint with β=60º (6.52 J). However, large improvements in
the energy were found for the combined joint (11.02 J) and especially for the joint with θ=7.5
(11.83 J).

4.3. Tough adhesive (XNR6852 E-2)

Using the tough adhesive (XNR6852 E-2), the most effective geometric modification is the
adhesive fillet with θ=15°. Analysing the individual results, a combined joint is considered
with β=45° and θ=15°. The P-δ curves obtained for the several tested configurations using the
tough adhesive (XNR6852 E-2) are presented in Figure 19. The combined configuration
achieves the highest value of δPm, although it is very close to the joint with solely θ=7.5º. On
the other hand, and conversely to this result, Pm is highest for the joint with singular
modification of θ=7.5º. For both Pm and δPm these two joint configurations showed much
improved results over the conventional and β=45º joints.

Figure 19 – P-δ curves for the different joint configurations using the tough adhesive (XNR6852 E-2).

In Figure 20 it is possible to compare in more detail Pm and δPm for each model using the
tough adhesive (XNR6852 E-2). It is thus confirmed that the joint with θ=7.5º yields the best
Pm and highest δPm, with a small difference to the combined joint (differences of 6% and
nearly 0% above the combined joins, respectively). The best joint (with singular modification
of θ=7.5º) improved Pm by 42% and δPm by 24% over the conventional joint, whose result
was much similar to that of the joint with β=45º.

23
 Figure 20 – Pm and δPm for the different configurations using the tough adhesive (XNR6852 E-2).

The values of energy absorbed by SLJ using the tough adhesive (XNR6852 E-2) were close
for the conventional and β=45º joints (11.03 J for both cases), and a significant difference was
found for the other two configurations, whose absorbed energy benefitted from the higher
loads and joint deformations before failure. The optimal joint, i.e., with θ=7.5º, failed after an
energy absorption of 27.43 J, while the combined joint failed with 27.32 J.

5. Conclusions

With the aim of performing a numerical optimization of single-lap joints (SLJ) under impact
loads, the cohesive zone modelling (CZM) technique was first successfully validated with the
three adhesives under study, considering a wide range of adhesive properties. This validation
was undertaken considering experimental and CZM results from a previous work, showing
good maximum load (Pm) predictions. A discrepancy was found in the displacement at Pm
(δPm), but this was justified by errors in the experimental measurement of the displacement
(δ). The individual study of parameters showed a reduced influence of the outer adherends’
chamfer, with maximum Pm improvements over the standard condition of 6% for the AV138,
22% for the DP8005 and 11% for the XNR6852 E-2. In all cases these improvements were
achieved with an outer chamfer angle (α) of 7.5º. The inner adherends’ chamfer showed
promising results for the AV138, with a maximum Pm improvement of 14% for the AV138
(inner chamfer angle, β, of 15º). However, for the other adhesives, due to their ductility, Pm
was marginally improved for intermediate β values, whilst it significantly diminished for the

24
smaller β due to the reduction of transmitted loads at the chamfer. Finally, the adhesive fillet
showed the most promising results, for all adhesives, but with best results for the brittle
adhesive, due to its sensitivity to peak stresses. The maximum Pm improvements were always
found for an adhesive fillet angle (θ) of 7.5º, and these were of 54% (AV138), 37% (DP8005)
and 39% (XNR6852 E-2). The optimal combination study that followed enabled merging the
inner chamfer with the adhesive fillet. However, the synergy between these two geometric
modifications could improve the singular parameter adhesive fillet modification, although the
Pm results were always close to the filleted condition. As a result of this work, the best
geometric condition for SLJ was defined as function of the adhesive, showing that good
improvements over the standard geometry can be found.

References

[1] Galvez P, Quesada A, Martinez MA, Abenojar J, Boada MJL, Diaz V. Study of the
behaviour of adhesive joints of steel with CFRP for its application in bus structures.
Composites Part B: Engineering 2017;129:41-6.
[2] Wu Y, Liu Q, Fu J, Li Q, Hui D. Dynamic crash responses of bio-inspired aluminum
honeycomb sandwich structures with CFRP panels. Composites Part B: Engineering 2017.
[3] Machado J, Marques E, da Silva L. Adhesives and adhesive joints under impact loadings:
An Overview. The Journal of Adhesion 2017;DOI: 10.1080/00218464.2017.1282349.
[4] Banea M, Silva Ld, Campilho R. Effect of temperature on tensile strength and mode I
fracture toughness of a high temperature epoxy adhesive. Journal of Adhesion Science and
Technology 2012;26(7):939-53.
[5] Agarwal A, Foster SJ, Hamed E. Testing of new adhesive and CFRP laminate for steel-
CFRP joints under sustained loading and temperature cycles. Composites Part B: Engineering
2016;99:235-47.
[6] Adams R, Peppiatt N. Stress analysis of adhesive-bonded lap joints. Journal of strain
analysis 1974;9(3):185-96.
[7] Adams R, Coppendale J, Peppiatt N. Failure analysis of aluminium-aluminium bonded
joints. Adhesion 1978;2:105-20.
[8] Harris JA, Adams RD. An assessment of the impact performance of bonded joints for use
in high energy absorbing structures. Proceedings of the Institution of Mechanical Engineers,
Part C: Journal of Mechanical Engineering Science 1985;199(2):121-31.

25
[9] Hart-Smith L. Adhesive bonding of composite structures—progress to date and some
remaining challenges. Journal of Composites, Technology and Research 2002;24(3):133-51.
[10] Vallée T, Keller T. Adhesively bonded lap joints from pultruded GFRP profiles. Part III:
Effects of chamfers. Composites Part B: Engineering 2006;37(4-5):328-36.
[11] You M, Yan Z-m, Zheng X-l, Yu H-z. Effect of outer chamfer on stress distribution in
single lap joint. Journal of Aeronautical Materials 2007;27(5):91.
[12] Mylonas C, Drucker D. Twisting stresses in tape. Experimental Mechanics 1961;1(7):23-
32.
[13] Tsai MY, Morton J. The effect of a spew fillet on adhesive stress distributions in
laminated composite single-lap joints. Composite structures 1995;32(1-4):123-31.
[14] Sato C. Design for Impact Loads. In: da Silva LFM, Öschner A, Adams RD, editors.
Handbook of Adhesion Technology: Springer; 2011. p. 743-63.
[15] Sawa T, Higuchi I, Suga H. Three-dimensional finite element stress analysis of single-lap
adhesive joints of dissimilar adherends subjected to impact tensile loads. Journal of Adhesion
Science and Technology 2003;17(16):2157-74.
[16] Cavalli MN, Thouless MD. The effect of damage nucleation on the toughness of an
adhesive joint. The Journal of Adhesion 2001;76(1):75-92.
[17] Campilho RDSG, De Moura MFSF, Domingues JJMS. Stress and failure analyses of
scarf repaired CFRP laminates using a cohesive damage model. Journal of Adhesion Science
and Technology 2007;21(9):855-70.
[18] Carlberger T, Stigh U. An explicit FE-model of impact fracture in an adhesive joint.
Engineering Fracture Mechanics 2007;74(14):2247-62.
[19] Haufe A, Pietsch G, Graf T, Feucht M. Modelling of weld and adhesive connections in
crashworthiness applications with LS-DYNA. Proceedings of the NAFEMS seminar
“simulation of connections and joints in structures” Wiesbaden, Germany2010.
[20] May M, Voß H, Hiermaier S. Predictive modeling of damage and failure in adhesively
bonded metallic joints using cohesive interface elements. International Journal of Adhesion
and Adhesives 2014;49:7-17.
[21] Clarke MI, Broughton JG, Hutchinson AR, Buckley M. Application of the design of
experiments procedure to the behaviour of adhesively bonded joints with plastically
deformable adherends to enable further understanding of strain rate sensitivity. International
Journal of Adhesion and Adhesives 2013;44:226-31.
[22] Avendaño R, Carbas RJC, Chaves FJP, Costa M, da Silva LFM, Fernandes AA. Impact
Loading of Single Lap Joints of Dissimilar Lightweight Adherends Bonded With a Crash-

26
Resistant Epoxy Adhesive. Journal of Engineering Materials and Technology
2016;138(4):041019.
[23] Neumayer J, Kuhn P, Koerber H, Hinterhölzl R. Experimental determination of the
tensile and shear behaviour of adhesives under impact loading. The Journal of Adhesion
2016;92(7-9):503-16.
[24] da Silva LF, De Magalhaes F, Chaves F, De Moura M. Mode II fracture toughness of a
brittle and a ductile adhesive as a function of the adhesive thickness. The Journal of Adhesion
2010;86(9):891-905.
[25] Pinto AMG, Magalhães A, Campilho RDSG, De Moura M, Baptista A. Single-lap joints
of similar and dissimilar adherends bonded with an acrylic adhesive. The Journal of Adhesion
2009;85(6):351-76.
[26] da Silva LF, Da Silva R, Chousal J, Pinto A. Alternative methods to measure the
adhesive shear displacement in the thick adherend shear test. Journal of Adhesion Science and
Technology 2008;22(1):15.
[27] Silva MRG, Marques EAS, da Silva LFM. Behaviour under Impact of Mixed Adhesive
Joints for the Automotive Industry. Latin American Journal of Solids and Structures
2016;13(5):835-53.
[28] Reedy Jr ED. Strength of butt and sharp-cornered joints. The Mechanics of Adhesion:
Elsevier New York, NY; 2002. p. 145-92.
[29] Nunes SLS, Campilho RDSG, da Silva FJG, de Sousa CCRG, Fernandes TAB, Banea
MD, et al. Comparative failure assessment of single and double-lap joints with varying
adhesive systems. J Adhesion 2016;92:610-34.
[30] Taib AA, Boukhili R, Achiou S, Gordon S, Boukehili H. Bonded joints with composite
adherends. Part I. Effect of specimen configuration, adhesive thickness, spew fillet and
adherend stiffness on fracture. Int J Adhes Adhes 2006;26(4):226-36.
[31] Campilho RDSG, de Moura MFSF, Domingues JJMS. Numerical prediction on the
tensile residual strength of repaired CFRP under different geometric changes. Int J Adhes
Adhes 2009;29(2):195-205.

27