You are on page 1of 28

Journal of Adhesion Science and Technology

ISSN: 0169-4243 (Print) 1568-5616 (Online) Journal homepage: https://www.tandfonline.com/loi/tast20

Evaluation of an elastic meshless formulation


to adhesive joints’ strength prediction against
established methods

I.J. Sánchez-Arce, L.D.C. Ramalho, R.D.S.G. Campilho & J. Belinha

To cite this article: I.J. Sánchez-Arce, L.D.C. Ramalho, R.D.S.G. Campilho & J. Belinha
(2019): Evaluation of an elastic meshless formulation to adhesive joints’ strength prediction
against established methods, Journal of Adhesion Science and Technology, DOI:
10.1080/01694243.2019.1702829

To link to this article: https://doi.org/10.1080/01694243.2019.1702829

Published online: 23 Dec 2019.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tast20
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY
https://doi.org/10.1080/01694243.2019.1702829

Evaluation of an elastic meshless formulation to adhesive


joints’ strength prediction against established methods
I.J. Sanchez-Arcea , L.D.C. Ramalhoa , R.D.S.G. Campilhoa,b and
J. Belinhaa,b
a
INEGI, Instituto de Ci^encia e Inovac~ao em Engenharia Mec^anica e Engenharia Industrial, Porto,
Portugal; bDepartment of Mechanical Engineering, Instituto Superior de Engenharia do Porto (ISEP),
Instituto Politecnico de Porto, Porto, Portugal

ABSTRACT ARTICLE HISTORY


Adhesive joints are widely used in the production of goods, Received 2 July 2019
mainly in the transport industry. However, their industrial applica- Revised 6 November 2019
tions often have non-standard complex shapes. Computer simula- Accepted 1 December 2019
tion, like the finite element method (FEM), is widely used for their
KEYWORDS
analysis but limitations still exist. Meshless methods have been in Finite element method
development and offer an option to overcome some limitations (FEM); Natural Neighbour
of the FEM; however, these are still in development. In this work, Radial Point Interpolation
a recent meshless method, the NNRPIM, has been applied to the Method (NNRPIM); joint
analysis of adhesive joints. First, experimental data corresponding strength; analytical models;
to four overlap lengths and three different adhesives were meas- meshless method;
ured, as a benchmark. Afterwards, joint strength was analytically single-lap joint
obtained as a second benchmark. Then, all the joint geometries
were simulated utilising the FEM and NNRPIM methodologies,
joint strengths were calculated from those simulations. Finally, the
results were compared against the first and second benchmarks.
Meshless method proved to be a good alternative to the FEM,
providing similar strength prediction with differences less than
3% between them. Moreover, the stress distribution curves were
compared, differences of 5% in the peak shear stresses were
found. In conclusion, the NNRPIM provides accurate results and
could be utilised for further study of adhesively-bonded joints.

1. Introduction
Most of the objects used in our daily life are composed of several components, regard-
less their size, joint together. Throughout the years, several mechanical ways to joint
materials have been developed and perfected (e.g. riveting, bolting, welding, brazing).
Although adhesives have been used in human life since thousands of years [1,2], they
have become more ubiquitous in recent years with applications to the aerospace, auto-
motive, energy, and construction fields, to name a few [3,4]. In general, the types of
joint (geometries) are similar to those used for welding, such as lap joints, butt-joint,

CONTACT I.J. Sanchez-Arce isidrodjsa@gmail.com INEGI, Instituto de Ci^encia e Inovac~ao em Engenharia


Mec^anica e Engenharia Industrial, Porto 4200-465, Portugal
ß 2019 Informa UK Limited, trading as Taylor & Francis Group
2 I.J. SÁNCHEZ-ARCE ET AL.

Nomenclature
E Young’s modulus of the substrates L0 Overlap length
Ea Young’s modulus of the adhesive c Half overlap length
G Shear modulus of the adhesive P Tensile force applied to the joint
t1 Thickness of the upper substrate x Distance from the overlap centre
t2 Thickness of the lower substrate towards the joint ends
t3 Adhesive layer thickness X Domain for the meshless method
t Substrate thickness when t1 ¼ t2 n Number of nodes in the domain X.
b Out of plane width of the joint

‘T joint’, and their combinations are used. Although adhesive joints are expected to
work only in shear, it is not always achieved [1].
The single-lap joint (SLJ) is the most used type for analysis and experimental test-
ing being followed by the double-lap joint (DLJ) [1]. SLJs are more used because of
their easiness to manufacture. However, the SLJ has a disadvantage, the load acts
eccentrically because of the lap. Conversely, in DLJs, the load acts concentrically, but
it is more complex to be executed for repair work. The strength of an adhesively-
bonded SLJ can be influenced by several variables such as substrate and adhesive
mechanical properties, substrate thickness, overlap length, adhesive layer thickness,
surface preparation, and adhesive curing time [5]. In experimental tests of brittle
(AralditeV AV138) and ductile adhesives (similar to the AralditeV 2015) the joint
R R

strength was inversely proportional to the adhesive layer thickness [6]. Similarly, for
tougher adhesives (SikaForceV 7888) the strength decreases as the adhesive layer
R

thickness increases [7]; in all cases the minimum tested adhesive layer thickness was
0.2 mm. Moreover, it is important to maintain a thin layer thickness to reduce voids,
which lower the joint strength [6,7]. In thin adhesive layers the cohesive fracture is
observed at the centre of the adhesive layer whilst in thicker layers the fracture has
been observed closer to the interfaces [7].
To analyse both SLJ and DLJ analytical models have been derived, being the one
developed by Volkersen [8] amongst the first. This model took into account the differ-
ences in substrate thickness and neglected the load eccentricity, and so only ‘shear’
stresses can be determined [1,9]. Subsequently, Goland and Reissner [10] developed a
model which considered the load eccentricity; thus, both ‘shear’ and ‘peel’ stresses
could be determined; however, it was only valid for short overlaps [11], and the stress
values are estimated only in the adhesive mid-thickness [12]. Goland and Reissner’s
model was later improved and expanded by other authors (e.g. [11,13]), being the
works of Hart-Smith remarkable because of their deep analysis. Hart-Smith developed
models for both elastic and elastic-plastic behaviour of the adhesive [11]. It is import-
ant to note that all these models considered an elastic-only behaviour of the substrates
and a shear-only failure of the adhesive [11]. Crocombe proposed that the joint’s max-
imum strength is achieved when the whole adhesive layer has yielded (s  sy) while the
adherends have not. This model is known as ‘Global yielding’ (GY) [14]. It is important
to note that all these analytical models consider the SLJ as a two-dimensional geometry
in a ‘plane strain’ condition. Analytical models like Hart-Smith (in the elastic-plastic
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 3

version) provide a good tool for parametric studies [1]. More recently, this concept was
retaken by Her [15] to parametrically analyse SLJs and DLJs with different adhesive
thicknesses, adhesive/adherend (G/E) ratios using his own analytical model, validated
also via the finite element method (FEM) [15].
Although analytical models provide an insight about the stresses in an adhesive
joint, the continuous development of numerical and computational technologies had
proven to provide a better mean to analyse stress and strains in the joints [1].
Analytical models were developed for joints with certain characteristics, both geomet-
rical and physical (i.e. materials) while the numerical techniques do not possess such
constraints [1]. The most known and used technique is the FEM in which any geom-
etry is discretised by a set of small elements. Moreover, FEM allows investigating the
effect of geometrical changes in the adhesive and adherends with the aim of reducing
stress concentration at the joint ends [14,16,17]. Although adhesive joints as the SLJ
are three-dimensional, the use of two-dimensional models to analyse them numerically
has been demonstrated to be a good alternative for their analysis [16,18].
The FEM is also used as a benchmark for newer analytical formulations
[14,15,19,20]. Together with experimental work, it has been also used to show the val-
idity of newer numerical techniques [19–22]. It is important to note that experimental
data provides the true joint strength and numerical or analytical formulations allow to
estimate the stress and strain distributions in the joint.
Meshless methods have been evolving throughout the years. There are two main
classifications for such methods, those using approximation functions and those
using interpolation. Interpolating meshless methods possess Kronecker’s delta prop-
erty, as the FEM. It is a property that allows to directly impose the essential and
natural boundary conditions. Among those methods is the ‘natural neighbour radial
point interpolation method’ (NNRPIM), which has been reported to be as accurate
as of the FEM, regardless whether or not the mesh is structured or uniform.
Moreover, the stress patterns obtained with the NNRPIM are smoother than those
from the FEM [23].
In the FEM, nodes are connected by elements which should obey shaped poly-
gons. Under large deformation cases, the elements can be heavily distorted causing
convergence issues. Conversely, meshless methods and NNRPIM in particular do
not have those restrictions, which makes them more suitable to large deformation
analyses [22–25]. In the adhesively-bonded joints case, their strength is determined
by deforming the joint until failure; this process causes a large-deformation state to
the adhesive.
Regarding to adhesive joints, meshless methods have been successfully applied in
few cases. The Element Free Galerkin Method (EFGM) was applied to layer-composed
plates and cantilever beams; afterwards, the results were compared with FEM solutions
for the same cases [19]. The double cantilever beam (DCB) test was analysed using
‘symmetric smoothed particle hydrodynamics’ (SSPH) and a cohesive zone models; the
results were compared with both experimental and numerical data [26]. In the afore-
mentioned cases, good agreement was found between all the methods. Subsequently,
the ‘smoothed particle hydrodynamics’ (SPH) was applied to adhesive joints with the
standardised geometry and an overlap length of 12.5 mm, aiming to evaluate fracture
4 I.J. SÁNCHEZ-ARCE ET AL.

of the joint. Peel and shear stress distributions were determined and compared with
FEM solutions; although this is the first simulation of an SLJ using meshless methods,
the results qualitatively agreed; however, the stress distributions presented a rather
heavy numerical noise [20]. Later, a combination of cases SLJ, bonded, bolted, and
bonded-bolted were modelled using a ‘radial point interpolation method’ (RPIM) being
the first work including non-linear material behaviour and the bolt to adherends contact
[21]. Although these studies provide the bases to use meshless methods in the analysis of
adhesive joints, they only provide an insight into the method and they are not compar-
able amongst them. Moreover, these studies only considered the ‘standard’ SLJ, in which
case also analytical models provide good results; in addition, the adhesive type varied
amongst the studies, some used brittle [20,26] whilst other applied ductile [21].
Recently, the NNRPIM has been extended to elastic-plastic analysis of metallic speci-
mens under tensile tests providing promising results [22]. On the other hand, adhesive
joints are often analysed using elastic-plastic models which provide a closer estimation
to the experimental work. However, it is necessary to investigate the NNRPIM suitabil-
ity to be applied for analysing adhesive joints under elastic behaviour to subsequently
apply the elastic-plastic models.
Joint strength, however, is dependent on the ductility or brittleness of the adhesive
in particular, and so the prediction methodology should be chosen accordingly [14].
Nevertheless, the consideration of elastic-plastic adhesive and adherends in those mod-
els leads to more accurate predictions.
This paper aims to evaluate the NNRPIM method for the elastic analysis of adhesive
joints. The suitability of the method was evaluated by comparing it with the known
analytical models and numerical methodologies as FEM. This evaluation could be the
basis for further analyses of adhesive joints such as elastic-plastic and joint geometries
with industrial applications.

2. Experimental work
The substrates were manufactured from an aluminium AW6082-T651 sheet. The
mechanical properties of the substrates’ material were determined following the guide-
lines and dimensions from the ASTM-E8M-04 standard [27] and it was performed in a
previous work [28]; further description of the properties in § 2.1.
Afterwards, the tensile mechanical properties of each adhesive were determined by
following the guidelines and dimensions from the NF T 76-142 standard [29].
Adhesive’s shear properties were determined via the TAST test. The critical strain
energy release rates were obtained via the DCB, and ENF tests, respectively, following
the ISO 11003-2 standard, as described in previous work [30–35].
Subsequently, single lap joint (SLJ) test samples were prepared by following the
ASTM D1002 standard. In this work, three adhesive types were considered, and so in
order to obtain strong data for the analysis five samples were prepared for each adhe-
sive type and overlap length giving a total of 60 samples.
The adhesives tested in this work were AralditeV AV138 and 2015, and
R

SikaForceV 7888.
R
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 5

Table 1. Mechanical properties of the adhesives [30–33].


Property AV138 2015 7888
E (GPa) 4.89 ± 0.81 1.85 ± 0.21 1.89 ± 0.81
 0.35a 0.35a 0.35a
ry (MPa) 36.49 ± 2.47 12.63 ± 0.61 13.20 ± 4.83
rf (MPa) 39.45 ± 3.18 21.63 ± 1.61 28.60 ± 2.0
ef (%) 1.21 ± 0.10 4.77 ± 0.15 43.0 ± 0.6
G (GPa) 1.56 ± 0.01 0.56 ± 0.21 0.71b
sy (MPa) 25.1 ± 0.33 14.6 ± 1.3 –
sf (MPa) 30.2 ± 0.40 17.90 ± 1.8 20a
cf (%) 7.8 ± 0.70 43.90 ± 0.02 100
GIC (N/mm) 0.20 0.43 ± 0.02 1.18 ± 0.22
GIIC (N/mm) 0.38 4.70 ± 0.34 8.72 ± 1.22
a
Manufacturer’s data.
b
Calculated by Hooke’s Law.

Figure 1. Experimental and numerical stress-strain diagram for the three adhesives tested.

2.1. Joint materials


2.1.1. Substrates
The aluminium alloy used as substrate was characterised to determine its mechanical
properties being Young’s modulus E ¼ 70.07 ± 0.83 GPa, yield strength
ry ¼ 261.67 ± 7.65 MPa, ultimate strength rf ¼ 324.00 ± 0.16 MPa, and a failure
strain ef ¼ 21.70 ± 4.24%.

2.1.2. Adhesives
The mechanical properties of each one of the adhesives were experimentally deter-
mined in previous works [30–33] and are listed in Table 1. Moreover, their tensile
stress-strain behaviour is shown in Figure 1.

2.2. Geometry, fabrication, and testing


The SLJs tested were manufactured following the ASTM D1002 standard using a sub-
strate thickness tS ¼ 3 mm, an adhesive thickness tA ¼ 0.2 mm, and four overlap
6 I.J. SÁNCHEZ-ARCE ET AL.

Figure 2. Schematic representation of a SLJ and its parameters.

lengths L0 ¼12.5, 25.0, 37.5, and 50 mm (Figure 2). The clamping distance was set to
LT ¼ 180 mm in all cases.
Once all the substrates were cut, their surfaces were prepared by grit blasting the
bonding interfaces. These surfaces were later degreased with acetone and the adhesive
was applied. In order to achieve accuracy in the assembly a custom-made jig was used
to ensure joint uniformity and constant adhesive thickness; the adhesive thickness was
also controlled by placing calibrated metallic strips between one substrate and the jig
(see i.e. da Silva et al. [29]). Subsequently, all the specimens were left curing at room
temperature for a minimum of 48 h. Once all the joints were cured, the excess of adhe-
sive was mechanically removed (milling).
Finally, the tensile properties of the specimens were determined with a Shimadzu
AG-X 1000 Universal Testing Machine (UTM) with a 100 kN load cell and using a
strain rate of 1 mm/min, as recommended by the standard.

3. Analytical and numerical formulations


The analytical models described below were used to estimate the shear and, when pos-
sible, peel stresses in SLJ. These models are linear, and so only the Young’s modulus
(E), Poisson’s ratio () from adhesive and substrates are necessary. The stresses, peel
and shear, were normalised to the average shear stress which is the ratio between
applied force over adhesive area [18,36].

3.1. Volkersen model


The first theoretical model developed for analysing stress distribution in adhesive joints
was developed by Volkersen [8] predicting the shear stresses along the adhesive by con-
sidering the adherends to be loaded only in tension while the adhesive is only in shear
[36]. The model takes into account different thickness of the upper and lower adher-
ends making it suitable to analyse DLJs; however, both adherends should be from equal
materials. The overlap’s centre is considered as the origin from where the stresses are
calculated towards the joint ends (Equation (1)).

P x coshðxXÞ ðt1  t2 Þx sinhðxXÞ


sðXÞ ¼ þ (1)
bL0 2 sinhðx=2Þ ðt1 þ t2 Þ2 coshðx=2Þ
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 7

Where
 
t1 GL2
x2 ¼ 1þ
t2 Et1 t3

x

L0

The Volkersen model provides an insight into the shear stresses in the joint;
however, during tensile loading of an SLJ bending occurs in the joint area caus-
ing ‘peel’ stresses whilst the Volkersen model does not consider such effects, it is
the base for further analytical methods like the Goland and Reissner
described next.

3.2. Goland and Reissner model


The model of Goland and Reissner [10] considers the bending of the adherends
because of a tensile force applied to the joint. This effect was not considered in the
Volkersen model. Goland and Reissner considered factors which represent the bending
moment and transverse force via two constants k and k0, respectively. The consider-
ation of the bending moment and transverse force allows estimating the shear and
‘peel’ stress distributions in the adhesive layer by means of Equations (2) and (3)
[9,37,38].
0   1
bx
P B C
cosh
BbL ð  t  þ 3ð1  kÞC
sðxÞ ¼  B 1 þ 3kÞ C (2)
4L0 b @ 2t bL0 A
sinh
2t

     
4Pt 2k 0 2kx 2kx
rðxÞ ¼ ð Þ ð Þ
R2 k þ kk cosh k cos k cosh cos
bDL20 2 L0 L0
(3)
     
4Pt 2k 0 2kx 2kx
þ R1 k þ kk sinhðkÞ sin ðkÞ sinh sin
bDL20 2 L0 L0

Where

Ga t1
b2 ¼ 8
E t3
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi
3ð1   2 Þ 1 P
u2 ¼
2 t1 bt1 E
8 I.J. SÁNCHEZ-ARCE ET AL.

L0

2

coshðu2 cÞ
k¼ pffiffiffi
coshðu2 cÞ þ 2 2 sinhðu2 cÞ
(4)
rffiffiffiffiffiffiffiffiffiffiffi
Ea t1
c¼ 6
4

Et3

c
k¼c
t1
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kc P
k0 ¼ 3ð1   2 Þ
t Ebt

R1 ¼ coshðkÞsinðkÞ þ sinhðkÞcosðkÞ
(5)
R2 ¼ sinhðkÞcosðkÞ  coshðkÞsinðkÞ

sinð2kÞ þ sinhð2kÞ

2

Therefore, the bending moment (M) and the transverse force (V) are determined as
follows:

Pt
M¼k (6)
2b

Pt
V ¼ k0 (7)
2b

3.3. Hart-Smith model


Hart-Smith [11] developed an analytic method to determine the stress distributions (both
peel and shear) in an adhesive joint. The method considers either elastic or elastic-plastic
behaviour on the shear stress while the peel is always considered elastic [9,37]. For the
purpose of this work, only the elastic behaviour has been considered allowing to compare
the predictions from different methods. The shear stress distribution is calculated by
means of Equation (8); the peel stress distribution by means of Equation (9); both cases
are function of the material properties and geometries of adherends and adhesive, and the
load applied to the joint (Equations (10)–(15)).

s ¼ A2 coshð2k0 xÞ þ C2 (8)
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 9

r ¼ A coshðvxÞcosðvxÞ þ B sinhðvxÞsinðvxÞ (9)

Where:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  ffi
1 þ 3 ð 1   2Þ
2G
k0 ¼ (10)
4 t3 Et

Et 3
D¼ (11)
12ð1   2 Þ

P
n2 ¼ (12)
bD

Pðt þ t3 Þ 1

b n 2 c2 (13)
1 þ nc þ
6
 
G P 6ð1   2 ÞM 1
A2 ¼ þ (14)
t3 Et b t 2k sinhð2k0 cÞ
0

 
1 P A2
C2 ¼ ð 0 Þ
 2 0 sinh 2k c (15)
2c b 2k

Similarly to the ‘Goland and Reissner’ model, the Hart-Smith model considers adher-
end bending due to the non-aligned loading. However, the method to calculate the factor
k is a function of the adherend’s bending stiffness (n), as shown in Equation (16); never-
theless, Equations (6) and (7) are still valid for such purpose.
 
t3 1
k¼ 1þ
t1 ðncÞ2 (16)
1 þ nc þ
6

3.4. Global yielding criteria


The global yielding criterion is based on the maximum amount of load that a joint can
support before the adhesive yields (sxy ¼ sy). Being Pm the joint strength limited by the
adhesive; this concept considers the adhesive to be in pure shear (Equation (17)).
However, none of the adherends should yield (Equation (18)), in such case, Pa is the
maximum load the joint can withstand prior to adhered yielding, and so the joint
strength is determined by the weakest element of the joint.

Pm ¼ sy bL (17)
10 I.J. SÁNCHEZ-ARCE ET AL.

Pa ¼ ry bt (18)

For all the aforementioned methods, the peak stresses, both peel and shear, are
located at the joint ends. Thus, the joint strength can be predicted by equalling the
shear stress to the adhesive shear strength. This concept is known as the maximum
shear stress criterion (MSSC). Similarly, the peel stress is equalled to the tensile
strength of the adhesive to determine the joint strength under peel, this criterion is
known as maximum peel stress criterion (MPSC).

3.5. Meshless method


3.5.1. Nearest neighbours radial point interpolation method
The NNRPIM method considers that the geometry to be analysed is composed by a set
of nodes N ¼ fx1 , x2 , :::, xm g 2 Rd which can be either randomly or uniformly distrib-
uted in the surface or volume, and ‘n’ corresponds to the number of nodes inside the
domain X: Then, a Voronoï diagram is created from N resulting in the ‘backing’ mesh
for the method and it provides the base for selecting the ‘influence cell’ V i [25]. Taking
a node xI from X as the point of interest, all the nodes inside V i have an effect on xI .
Therefore, the function relating xI can be described as the sum of two shape functions,
one radial based and one polynomial, as shown in Equation (19) [22,25,39].
Pn Pm
uðxI Þ ¼ i¼1 Ri ðxI Þai ðxI Þ þ j¼1 Pj ðxI Þbj ðxI Þ (19)

where Ri ðxI Þ is the ‘radial basis function’ (RBF), and ai ðxI Þ is a non-constant coefficient
of Ri ðxI Þ [23]. Similarly, Pj ðxI Þ is a polynomial basis function and bj ðxI Þ are their non-
constant coefficients [39]. Various RBFs exist, here one related to the ‘Euclidean norm’
(distance between two points in space) ðrIi Þ was chosen, the multi-quadratics radial
basis functions (Equation (21)).

p
RðrIi Þ ¼ ðrIi 2 þ c2 Þ (20)

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rIi ¼ ðxI  xi Þ2 þ ðyI  yi Þ2 þ ðzI  zi Þ2 (21)

Although c and p are parameters that have to be optimised, their values were already
optimised as c  0:0001 and p  0:9999 [22,23]. The non-constant coefficients ai ðxI Þ
and bj ðxI Þ can be obtained by applying the interpolation function [19] to all nodes (n)
in the domain X [22,39].
The interpolation function [19] can be represented as a matrix [25], as follows:

us ¼ Ra þ Pb (22)
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 11

Where:
2 3
r1 ðx1 Þ r2 ðx1 Þ    rn ðx1 Þ
6 r ðx Þ r2 ðx2 Þ    rn ðx2 Þ 7
6 1 2 7
R¼6
6 .. .. .. .. 7 7 (23)
4 . . . . 5
r1 ðxn Þ r2 ðxn Þ    rn ðxn Þ

2 3
p1 ðx1 Þ p2 ðx1 Þ    pm ðx1 Þ
6 p ðx Þ p ðx Þ    p ðx Þ 7
6 1 2 2 2 m 2 7
P¼6
6 . . . . .. 7 7 (24)
4 . .. .. . 5
p1 ðxn Þ p2 ðxn Þ    pm ðxn Þ

Then, Equation (19) can be expressed as:


n o   n o 
us T us
uðxI Þ ¼ r ðxI Þ ; pðxI Þ MT 1
T T
¼ UðxI Þ ; WðxI ÞT
(25)
z z

where MT is the total moment matrix. Subsequently, Equation (25) can be


expressed as:
Pn
uðxI Þ ¼ i¼1 Ui ðxI Þui (26)

Where Ui ðxI Þ is the shape function value at xI [25]. Subsequently, the partial deriva-
tives with respect to x and y of uðxI Þ: Further details of the method in [22,23,25,39].
Subsequently, the weak-Galerkin form is used to apply the material matrix and the nat-
ural and essential boundary conditions [23]. First, the equilibrium condition is estab-
lished considering the stress tensor r and the body forces b, as shown in [27].

rr þ b ¼ 0 (27)

Then, the natural and essential boundary conditions are established:



rn ¼ t
(28)
u¼u

Where n is the normal, t is the tension applied to the natural boundary (Ct ), and u
is the displacement applied to the essential boundary (Cu ) [23]. Then, the Galerkin
weak form can be expressed in a matricial form, as shown in [29], where K is the stiff-
ness matrix. Therefore, the boundary conditions and loading can be applied to it fol-
lowing the Hooke law [23].

du½KuF ¼ 0 (29)
12 I.J. SÁNCHEZ-ARCE ET AL.

Table 2. Number of nodes and elements used for the FEM models.
Overlap length (mm) Number of nodes Number of elements
12.5 7835 7444
25.0 11,035 10,560
37.5 11,835 11,360
50.0 11,151 10,680
Same nodes were used on the NNRPIM models.

Table 3. Number of Gauss points in the adhesive layer for each overlap
length case and solution method.
Solution case
Overlap length (mm) FEM NNRPIM
12.5 2880 5760
25.0 5280 10,560
37.5 6240 12,480
50.0 7200 14,400

3.6. SLJ models


Similarly to the experimental work described in §2.2, those four overlap lengths and
three adhesives were analysed numerically by means of various two-dimensional geo-
metrical models solved using the FEM and NNRPIM methods, from which the number
of elements and nodes are listed in Table 2. For comparison purposes the same number
of nodes and node location was used in the models for both FEM and NNRPIM; how-
ever, NNRPIM does not require the elements.
Although the number of nodes and nodal distributions were kept constant between
the FEM and NNRPIM solutions, the latter had more two times the number of Gauss
points in the adhesive layer, providing a more detailed stress distribution, as shown in
Table 3.
The ryy (peel) and sxy (shear) stresses were compared with those obtained with the
FEM (as a benchmark). The combination of method, overlap lengths, and adhesives
gave a total of 24 models.
In order to achieve consistency in the geometry creation, a MATLAB (The
MathWorks; Natick, Massachusetts, USA) script was developed allowing to create all
the geometries parametrically, with similar meshes, loads and boundary conditions.
Subsequently, all the models were solved using both FEM and NNRPIM techniques by
using a custom-written MATLAB script for each method. The substrate length and
thickness, the adhesive thickness, and the overlap length were considered equal to those
analysed experimentally.
In all cases a displacement (d ¼ 0.1 mm) was applied to the right side of the joint,
simulating the pulling exerted by the UTM. The left side of the joint was fixed (Ux ¼
Uy ¼ Uz ¼ 0). The material properties of substrates and adhesive were taken from pre-
vious experimental work (see §2.1); however, material properties were considered to be
linear-elastic, and so, these results can be compared with predictions derived from the-
oretical models. Once all the models were solved the ‘peel’ and ‘shear’ stresses in the
mid-adhesive layer were extracted and analysed.
Every adhesive and L0 combination results in its own stress (both ryy and sxy) mag-
nitude, although the stress patterns should be similar. In order to compare them, the
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 13

Figure 3. Photographs of some cohesive failures observed during the experiments; ordered from
brittle (AralditeV AV138), ductile (AralditeV 2015), and tough (SikaForceV 7888) adhesive types,
R R R

respectively. (a) L0 ¼ 12.5 mm; (b) L0 ¼ 37.5 mm; (c) L0 ¼ 25.0 mm; (d) Adherend plastic deformation
on tests with SikaForceV 7888 and L0 ¼ 50.0 mm. Adapted from Nunes et al. [40]; photographs
R

courtesy from Prof R.D.S.G. Campilho.

data were normalised. The overlap length was considered from 0 to 1 where 0 repre-
sents the left end whilst 1 the right end; the normalisation was achieved by dividing all
the points by L0. Similarly, the stress magnitudes (shear and peel) were normalised
with respect to the average shear stress (sxy).
All the integration (Gauss) points belonging to the adhesive layer were selected.
Subsequently, peel and shear stresses at those points were requested and plotted in con-
tour plots showing the stress variation along and across the adhesive layer. Thickness
and overlap length were normalised from 50% to 50% and from 0% to 100%, respect-
ively. For further clarity, the overlap axis was plotted in logarithmic scale and only the
left side was represented (0–50%); nevertheless, the right side is antisymmetric to left
along the centre of the adhesive layer (L0 ¼ 50%). Contour plots were drawn for both
FEM and NNRPIM solutions.

4. Results and discussion


4.1. Experimental joint behaviour
All the specimens presented cohesive failure of the adhesive (Figure 3). Moreover, no
plastic deformation of the adherends was observed, except for those with the largest
overlap and strongest adhesive (L0 ¼ 50 mm and SikaForceV 7888) (Figure 3(d)), which
R

also presented the maximum strength (Pm) (Figure 4).


The joint strength Pm or load carrying capacity increased proportionally with the
overlap length, such increment followed a ‘linear’ trend (Figure 4). As expected,
14 I.J. SÁNCHEZ-ARCE ET AL.

104
3
AV138
2015
7888
2

Pm (N)
1

0
12.5 25.0 37.5 50.0
L 0 (mm)

Figure 4. Strength determined from the SLJ specimens related to L0 and adhesive type.

adhesive brittleness has an important role in the final joint strength Pm; as shown in
Figure 4, the adhesive AralditeV AV138 showed the lowest Pm while the SikaForceV
R R

7888 presented the highest strength Pm. Nevertheless, in short overlaps (L0 ¼ 12.5 mm)
the joints presented similar strengths regardless of the adhesive used.
The peel (ryy) and shear (sxy) stress distributions, for identical joints, were deter-
mined in a previous work by the authors [40] and were related to the experimental
joint strengths. It can be inferred that adhesive’s E and L0 affect the final strength
(Figure 4). Moreover, ductile adhesives provide a better stress distribution along L0
[41], and they plasticise prior to the fracture allowing a higher joint strength [16].
Conversely, brittle adhesives, as the AralditeV AV138, cause higher-stress peaks which
R

result in lower-joint adhesive offers similar performance as the ductile, with a differ-
ence of 4.9% (Figure 4) because the maximum peak stresses (a function of L0) are lower
and more uniform [42] and the adhesive area is small; therefore, these small overlaps
do not benefit from the adhesive ductility [11]. Nevertheless, as L0 increases these peak
stresses increase, in which case more ductile adhesives perform better by yielding prior
failure. However, with strong adhesives like the SikaforceV 7888, there is yielding in
R

both adhesive and adherend (global yield) [43]. In conclusion, brittle adhesives were
found more suitable for SLJs with short overlaps, whilst ductile adhesives provide bet-
ter joint performance as the overlap increases. However, the adhesive should be strong
enough that does not cause yielding of the adherends.

4.2. Stress and strain distribution


Upon close inspection of the shear and peel stresses in the adhesive layers correspond-
ing to the 12 models analysed here, it was found that, qualitatively, the stress distribu-
tions are identical (Figure 5). Nevertheless, small differences were found at peak
stresses; the shear stresses estimated with the NNRPIM were found lower by 5.35%, in
average, (range 4.53% to 6.56%) with respect to those from the FEM (Figure 6(a)).
In the peel stresses case, the peak values were also lower by 3.10%, in average, (range
0.60% to 4.34%) with respect to the FEM estimated values (Figure 6(b)). The minimum
difference, in both peel and shear stresses, was found in the 25 mm overlap for the
three adhesives evaluated here.
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 15

(a) 4 (b) 6
FEM FEM
NNRPIM
3 4

/ avg
/ avg

yy xy
xy xy

2 2

1 0

0 -2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/LO x/LO
Shear stress Peel stress
Figure 5. Normalised ‘shear’ and ‘peel’ stress distributions at the centre of the adhesive layer on a
SLJ with 25.0 mm of overlap bonded with an AralditeV 2015 adhesive.
R

(a) 10 (b) 10
12.5 mm 37.5 mm 12.5 mm 37.5 mm
25.0 mm 50.0 mm 25.0 mm 50.0 mm
8 8
Difference (%)

Difference (%)

6 6

4 4

2 2

0 0
AV138 2015 7888 AV138 2015 7888
Adhesive Adhesive
τxy σyy
Figure 6. Difference, in percentage, between the peak stresses estimated from the NNRPIM
method compared to the FEM.

4.2.1. Stress distribution differences


The stress distributions obtained with both FEM and NNRPIM methods are rather
similar. Nevertheless, upon a close inspection along the whole of the overlap length,
differences were found in both shear and peel stress distributions. The first comparison
was performed at the peak values.
In the shear stress case, the maximum difference was 6.56%, corresponding to L0 ¼
50.0 mm with the adhesive AralditeV AV138; the smallest difference was 4.53% found
R

for L0 ¼ 25.0 mm with the adhesive AralditeV 2015, as shown in Figure 6(a).
R

In the peel stress case, the maximum difference was 4.34%, corresponding to L0 ¼
12.5 mm with the adhesive AralditeV AV138; the smallest difference was 0.60%
R

observed at L0 ¼ 50.0 mm with the adhesive AralditeV AV138, as shown in Figure 6(b).
R

From these data, it can also be observed that in all cases the smallest difference cor-
responded to L0 ¼ 25.0 mm regardless the adhesive (Figure 6), except for the
16 I.J. SÁNCHEZ-ARCE ET AL.

(a) (b)
8

Normalised yy difference (%)


Normalised xy difference (%)
6
0
10
4

0
10 -2
-2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/L0 x/L0
Shear Peel
Figure 7. Normalised differences of shear and peel stress distributions corresponding to
L0 ¼ 50.0 mm and SikaForceV 7888. These differences were calculated in the centre of the
R

adhesive layer.

aforementioned peel difference which looks odd compared with the trends of the data.
Moreover, such small differences indicate the NNRPIM method provide rather similar
results to those predicted by well-established methods as the FEM. Difference between
peak stress values comparing meshless to FEM have been reported before [20]; how-
ever, the difference found by Mubashar and Ashcroft [20] was five times larger than
the one observed here. Such big difference can be attributed to the node density; in this
study, the mean distance between nodes in the adhesive layer was 0.03 mm whilst the
reported distance of the other study was 0.06 mm [20].
Similarly, the differences throughout the full stress distributions, in both shear and
peel, were calculated for all the cases. The maximum difference was always located at
the joint ends (Figure 7); moreover, the maximum differences were found for L0 ¼
50.0 mm for all cases, except for the shear stress with SikaForceV 7888, where the max-
R

imum difference (5.47%) was found with the L0 ¼ 37.5 mm shear stress distribution.
In Figure 7, the normalised differences between FEM and NNRPIM predictions are
presented only using L0 ¼ 50.0 mm and SikaForceV 7888 as an example. The stress dis-
R

tributions corresponding to all cases are rather similar. Again, the smallest difference
was found with L0 ¼ 25.0 mm in all the adhesives. It is important to note that the dif-
ference between stress predictions from both methods (FEM & NNRPIM) was small
indicating that NNRPIM method is a good option for further analysis of adhe-
sive joints.
Stresses along the overlap length and across the layer thickness presented a saddle
shape with peaks at the free edges of the layer in all the cases tested. Moreover, there is
a slight stress variation across the adhesive thickness. Figures 8 and 9 present the typ-
ical stress distributions found in all cases; corresponding to the FEM and NNRPIM sol-
utions, respectively. The major differences between solutions are found between the
ends and 0.02 L0 (Figure 8 and 9). Moreover, the NNRPIM also predicted slight varia-
tions of both peel and shear stresses around 0.1ta from the interfaces, which were not
detected in the FEM solution.
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 17

Peel stress
7
0.4

)
1

yy avg
Normalised thickness position
6

/
2

0
1
2
0.2 5

Normalised peel stress (


4

3
0 3

2
-0.2
4
1
5

3
6

-0.4 0

-2 -1
10 10
Normalised overlap position

Shear stress
3
0.4

) xy avg
1
Normalised thickness position

/
2.5

1
1.5
0.2

Normalised shear stress (


2
1.5

0
1.5
-0.2
2.5

1
2

-0.4

10-2 10-1
Normalised overlap position

Figure 8. Contour plot of peel and shear stresses at the adhesive layer, FEM solution of a
L0 ¼ 12.5 mm and AralditeV 2015 joint; left half of the overlap is shown.
R

Moreover, in all cases, it was observed the presence of peaks in the free ends of the
adhesive layer. These peaks were quantified and are shown in Figure 10. The brittle
adhesive (AralditeV AV138) presented the highest peak values in all cases whilst the
R

ductile adhesives (AralditeV 2015 and SikaForceV 7888) presented rather similar values
R R

in all cases. In the FEM solution, the peak values increased with the overlap length,
whilst in the NNRPIM the peak values stabilised for overlaps greater than 37.5 mm
(difference 6.5% in shear, and 2.0% in peel), as shown in Figure 10.
The stress concentration at the joint’s free ends had been previously observed in a
photoelasticity study [44] and in finite element models of spew effects [16]. Moreover,
for a joint with 12.5 mm of overlap and joint with a brittle adhesive the peak was found
to be greater than 10 times the average shear stress [16]; in this study, for a similar
adhesive (AralditeV AV138) stress concentrations of 12.5 and 5.6 (MPa/MPa) were
R

found in peel and shear, respectively, and so agreeing with Adams and Peppiant [16].
Stress distribution at the interfaces follow the same shape as those at the adhesive layer
centre with the exception of presenting a peak at the joint free end. The magnitude of
the peaks corresponds to those reported in Figure 10. On the other hand, considering
the Saint Venant’s principle, as described in [16], if the Gauss points around 0.2 mm of
the free end are not considered, the stresses follow the same pattern and magnitude
across the whole adhesive thickness.
18 I.J. SÁNCHEZ-ARCE ET AL.

Peel stress
2 1 0 8
0.4

)
3

yy avg
Normalised thickness position

/
6

2
0.2 3

Normalised peel stress (


4
0

-0.2 4 2
5

0
6

-0.4 0

1
10-2 10-1
Normalised overlap position

Shear stress
1 0.5 3
2

)
0.4

xy avg
5
Normalised thickness position

1.
2.5

/
1

0.2

Normalised shear stress (


2
1.5

1
2

1.5

-0.2
1
5

-0.4
1.

0.5 0.5
1
10-2 10-1
Normalised overlap position

Figure 9. Contour plot of peel and shear stresses at the adhesive layer, NNRPIM solution of a
L0 ¼ 12.5 mm and AralditeV 2015 joint; left half of the overlap is shown.
R

4.3. Strength prediction


4.3.1. Analytical methods
In general, three out of four analytical methods analysed in this work under-predicted
the joint strength, whilst the fourth (GY) over-predicted it (Figure 11).
In particular, beginning with the shortest overlap (L0 ¼ 12.5 mm), Vorlkersen’s
method provides the minimum difference with respect to the experimental data with
the adhesive AralditeV AV138 provided a theoretical strength 11.56% lower than the
R

experimental (Figure 11(a)). Conversely, the GY criterion provided better predictions


for the ductile adhesives AralditeV 2015 and SikaForceV 7888 giving differences of
R R

1.32% and 22.83% with respect to the experimental data (Figure 11).
As L0 increased, the difference between the first three analytical methods and the
experimental data increased. However, regardless of the magnitude of L0 the Volkersen
method provided the best strength prediction only for the brittle adhesive AralditeV
R

AV138. The differences with respect to the experimental data were 21.87%, 34.15%,
and 40.06% corresponding to L0 ¼ 25.0, 37.5, and 50.0 mm, respectively (Figure 11(a)).
In all L0 combinations with the brittle adhesive, the Goland and Reissner and Hart-
Smith models predicted around 50% of the experimental joint strength. The best
approximation was with the shortest overlap giving strengths of 50.92% and 52.51%,
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 19

yy
, FEM yy
, NNRPIM
40 40

30 30
Normalised stress

Normalised stress
Peel ( yy / avg)

Peel ( yy / avg)
20 20

10 10

0 0
12.5 25.0 37.5 50.0 12.5 25.0 37.5 50.0
Overlap length (mm) Overlap length (mm)

xy
, FEM xy
, NNRPIM
40 40
2015
AV138
30 7888 30
Normalised stress

Normalised stress
Shear ( xy / avg)

Shear ( xy / avg)

20 20

10 10

0 0
12.5 25.0 37.5 50.0 12.5 25.0 37.5 50.0
Overlap length (mm) Overlap length (mm)

Figure 10. Normalised peak peel and shear stresses at the adhesive layer free ends. Comparison
between FEM (left) and NNRPIM (right) solutions.

respectively. The largest overlap also gave the smallest predicted strength, being 42.36%
and 45.50%, respectively (Figure 11(a)). Conversely, for the ductile adhesives, strength
prediction was found lower with these two analytical methods.
For the ductile adhesives, AralditeV 2015 and SikaForceV 7888, the GY model always
R R

provided the best strength estimation, even though it was mostly over-predicting. For
L0 ¼ 12.5 mm and the adhesive AralditeV 2015 strength prediction was rather close,
R

being 1.32% higher than the experimental (Figure 11(b)). For the other three overlap
lengths, the best predictions were found with the adhesive SikaForceV 7888 which pro-
R

vided differences lower than 10% in these three cases.


These results indicate that the Volkersen method provides the best strength pre-
diction for the ‘standard’ overlap and brittle adhesives. It has been suggested that
analytical methods are only valid for a short spectrum of the adhesive joints [1].
Moreover, these models have been developed having the ‘standard’ geometry as a
reference, and so its direct application to other overlaps is not always recom-
mended [11]. It is important to note that the main difference between the experi-
mental strengths and the analytical results is that the first three models neither
consider any non-elastic behaviour of the adhesive or the adherends nor the
brittleness or ductility of each adhesive.
20 I.J. SÁNCHEZ-ARCE ET AL.

(a) 10 4 (b) 10 4
3 3

2.5 2.5

2 2
Pm (N)

Pm (N)
1.5 1.5

1 1

0.5 0.5

0 0
12.5 25.0 37.5 50.0 12.5 25.0 37.5 50.0
L 0 (mm) L 0 (mm)

Experimental Hart-Smith Experimental Hart-Smith


Volkersen Global Yield Volkersen Global Yield
Goland & Reissner Goland & Reissner

Araldite ® AV138 Araldite ® 2015


(c) 10 4
3

2.5

2
Pm (N)

1.5

0.5

0
12.5 25.0 37.5 50.0
L 0 (mm)

Experimental Hart-Smith
Volkersen Global Yield
Goland & Reissner

SikaForce ® 7888
Figure 11. Comparison between experimental strength and analytical strength for each adhe-
sive tested.

The effect of these properties in the joint strength was pointed out by Hart-Smith
[11], and recently analysed by Nunes and colleagues [40], indicating that adhesive
toughness is necessary for larger overlaps [40]. Conversely, the GY criterion over-pre-
dicts the results in most cases but estimates strengths close to the experimental in those
adhesives with greater toughness. It can be attributed to the better load distribution
along the overlap [40] which is important for the application of this criterion [14].

4.3.2. Meshless method


The joint strength calculated from the FEM and NNRPIM predictions were com-
pared with the joint strength determined experimentally (Figure 4). It was found
that the experimental strength is higher than the predicted by the numerical work,
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 21

(a) 104 (b) 104


2.5 2.5
Exp Exp
2 FEM 2 FEM
NNRPIM NNRPIM
1.5 1.5
Pm (N)

Pm (N)
1 1

0.5 0.5

0 0
12.5 25.0 37.5 50.0 12.5 25.0 37.5 50.0
L 0 (mm) L 0 (mm)

Araldite ® AV138 Araldite ® 2015


4
(c) 2.5 10
Exp
2 FEM
NNRPIM
1.5
Pm (N)

0.5

0
12.5 25.0 37.5 50.0
L 0 (mm)

SikaForce ® 7888
Figure 12. Predicted joint strength for each adhesive based on the MSSC.

as it was also observed with the analytical models. Nevertheless, the lowest differ-
ence was observed with L0 ¼ 12.5 mm which is the overlap used by the ASTM
standard. In all cases, both experimental and numerical, the joint strength
increased proportionally with the overlap length (L0), as it can be observed in
Figures 4, 12, and 13.
Considering the MSSC, the minimum difference between experimental and numer-
ical data was found at L0 ¼ 12.5 mm where the numerical techniques under-predict the
joint strength by 52.86% and 55.69% (FEM and NNRPIM, respectively) corresponding
to the adhesive AralditeV AV138; while the maximum difference was observed at L0 ¼
R

50.0 mm with the adhesive SikaForceV 7888 where the numerical methods provide
R

strengths equivalent to 18.26% and 19.06% (FEM and NNRPIM, respectively) corre-
sponding to the experimental data (Figure 12).
Simarly with the MPSC, the numerical methods under-predict the joint strength with
respect to the experimental data. The minimum differences were observed with L0 ¼
12.5 mm and the adhesive AralditeV 2015 were 43.85% and 45.46% corresponding to the
R

FEM and NNRPIM, respectively. The maximum differences were again found with the
most ductile adhesive tested, SikaForceV 7888, corresponding to L0 ¼ 50.0 mm being
R

18.83% and 19.27% for the FEM and NNRPIM, respectively; as shown in Figure 13.
22 I.J. SÁNCHEZ-ARCE ET AL.

(a) 104 (b) 104


2.5 2.5
Exp Exp
2 FEM 2 FEM
NNRPIM NNRPIM
1.5 1.5
Pm (N)

Pm (N)
1 1

0.5 0.5

0 0
12.5 25.0 37.5 50.0 12.5 25.0 37.5 50.0
L 0 (mm) L 0 (mm)
AV138 2015
(c) 4
10
2.5
Exp
2 FEM
NNRPIM
1.5
Pm (N)

0.5

0
12.5 25.0 37.5 50.0
L 0 (mm)
7888
Figure 13. Predicted joint strength, based on the Maximum Peel Stress Criterium, for
each adhesive.

In reference to the analytical models, the numerical results from both methods
(FEM and NNRPIM) follow a similar trend in all cases. The maximum difference
between strength predictions from the Hart-Smith model and the FEM was found in
the joints with the AralditeV 2015 for all the overlaps; the maximum difference was
R

13.31% corresponding to L0 ¼ 37.5 mm. In the NNRPIM case, the maximum difference
observed was 9.69% corresponding to L0 ¼ 25.0 mm.
These differences can be attributed to the differences between the experimental and
numerical cases, being these differences summarised as follows. Experimental data took
in account the maximum load a joint holds prior its failure; when it occurs the adhe-
sive yielded; while the numerical data was considered, in this first stage, as linear
elastic, and so the final joint strength differs. The use of numerical models in the lin-
ear-elastic region has been reported before [20]. In that case, both FEM and meshless
method were compared, the meshless solution was threefold the FEM solution [20].
From a reported comparison between the FEM and the Goland and Reissner model,
the strength determined numerically was found around 20% lower than the analytical
for an SLJ [16]. Such differences can be expected because the joint strength is inversely
proportional to the peak stresses, numerical models capture better stress concentrations
at the joint ends which result in higher stresses on such points.
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 23

6000
FEM
NNRPIM

4000

Time (s)
2000

0
12.5 25.0 37.5 50.0
L 0 (mm)

Figure 14. Time difference between the FEM and NNRPIM methods for all the overlap lengths.

It is important to note that the reported models only considered one adhesive and
one overlap, and so cannot be directly compared with the present work. To summarise,
the agreement found between the models developed here and the analytical models
supports the hypothesis that the NNRPIM is a suitable tool to analyse adhesive joints.

4.4. Computational cost


In all cases, NNRPIM computational cost was higher than FEM time. The largest dif-
ference was observed for L0 ¼ 12.5 mm requiring four times more CPU-time when
compared to FEM. The minimum difference was observed for L0 ¼ 37.5 mm requiring
1.7 times more CPU-time (Figure 14).
Although there is a difference in computational cost, for this particular case, it can
be considered low; differences around 20 times with respect to FEM had been reported
in the analysis of an SLJ with L0 ¼ 12.5 mm by using another meshless method [20]. It
has also been suggested that meshless methods, in general, require more computational
time to solve the problem [39,45]. As it is already known, the computational time
depends on the mesh density. In the cases analysed in this work, the same node distri-
bution was used for both FEM and NNRPIM aiming to obtain similar stress results
with a comparable domain; therefore, time differences were expected because it is also
known that meshless methods as NNRPIM could provide good results with coarser
and even non-uniform mesh distributions [39].
It can be concluded that NNRPIM is a good alternative method to analyse adhe-
sively-bonded joints.

5. Conclusions
Numerical analysis of adhesively bonded has proven to be the most accurate method to
predict joint strength when considering non-linear material properties. Current estab-
lished numerical analyses require fine meshes in order to provide accurate results; how-
ever, large deformation analyses may require even finer meshes. The NNRPIM is a
meshless method that does not require a previous mesh or uniformly distributed nodes;
in addition, it allows large deformation analyses making it suitable to analyse adhesive
24 I.J. SÁNCHEZ-ARCE ET AL.

joints but its application still is in an early stage. In order to determine the suitability
of NNRPIM for this application, the strength of adhesive joints with various overlap
lengths and adhesive types was evaluated in this work. As a first stage, only linear elas-
tic properties were used and the predictions were compared against analytical models
and the finite element method.
The analytical models estimated joint strengths as close as 11.56% to the experimen-
tal data only for short overlaps and brittle adhesives. Longer overlaps increase the joint
strength when the adhesive has good toughness. Hence, the ‘global yielding’ criterion is
the best alternative, with estimated strengths as close as 5.28% to the experimental
data. In consequence, the application of analytical models is also governed by the adhe-
sive’s characteristics. Two of the analytical models evaluated in this work, Volkersen’s
and Goland and Reissner, were developed to consider only the elastic region of the
materials. The third one, Hart-Smith has been developed to consider elastic adherends
and elastic or elastic-plastic adhesives; for comparison purposes, only the elastic model
was used in this work. Accordingly, the same properties were used in the numerical
cases to ease comparison. Similar to the analytical models, the numerical predictions
under-predicted the joint strength by approximately 50% using the brittle adhesive.
Likewise, as the overlap and adhesive toughness were increased, the predicted strength
was found lower than the experimental around 80%. Therefore, a similar trend as the
analytical models was observed.
The NNRPIM strength estimations were found similar to those obtained from the
finite element method (FEM). The maximum difference observed was 2.82% corre-
sponding to the brittle adhesive and the shortest overlap. Correspondingly, differences
between the stress distributions determined by the FEM and NNRPIM were evaluated,
the maximum difference observed correspond to the peak shear stress, being 6.56%. In
this application, the NNRPIM method provided similar results, both in magnitude and
pattern, to those obtained with the well-accepted FEM. Therefore, it can be concluded
that the NNRPIM method is suitable for this application leading to further research
applications such as DLJ and industrial applied joint geometries. Moreover, this
research leads to further exploration of the use of elastic-plastic material models in
order to achieve more accurate strength predictions regardless, overlap length or adhe-
sive toughness. Currently, application of RPIM and NNRPIM methods in the elastic-
plastic analysis of SLJ is being carried out; good results had been achieved with overlaps
lower than 37.5 mm, but for the 50 mm strength prediction is still low. Further research
work is being performed in order to improve material models and yield criteria, which
will allow to enhance strength predictions.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
The authors truly acknowledge the funding provided by the Ministerio da Ci^encia, Tecnologia
e Ensino Superior - Fundac~ao para a Ci^encia e a Tecnologia (Portugal), under project funding
[MIT-EXPL/ISF/0084/2017 and ‘POCI-01-0145-FEDER-028351’]. Additionally, the authors
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 25

gratefully acknowledge the funding of Project ‘NORTE-01-0145-FEDER-000022’ - SciTech -


Science and Technology for Competitive and Sustainable Industries, co-financed by Programa
Operacional Regional do Norte [NORTE2020], through Fundo Europeu de Desenvolvimento
Regional (FEDER).

ORCID
I.J. Sanchez-Arce http://orcid.org/0000-0001-7463-6608
L.D.C. Ramalho http://orcid.org/0000-0003-2579-4057
R.D.S.G. Campilho http://orcid.org/0000-0003-4167-4434
J. Belinha http://orcid.org/0000-0002-0539-7057

References
[1] Adams RD, Wake WD. Structural adhesive joints in engineering. Essex, England:
Elsevier Applied Science Publishers LTD; 1984.
[2] Fay PA. History of adhesive bonding. In: Adams RD, editor. Adhesive bonding science,
technology and applications. Cambridge, England: Woodhead Publishing Limited; 2005.
p. 3–22.
[3] €
da Silva LFM, Ochsner A, Adams RD. Handbook of adhesion technology. In: da Silva

LFM, Ochsner A, Adams RD, editors. Berlin; Heidelberg: Springer-Verlag; 2011.
[4] Gui C, Bai J, Zuo W. Simplified crashworthiness method of automotive frame for con-
ceptual design. Thin-Walled Struct. 2018;131:324–335.
[5] da Silva LFM, Dillard DA, Blackman B, Adams RD, editors. Testing adhesive joints.
Weinheim: Wiley-VCH Verlag & Co. KGaA; 2012.
[6] da Silva LFM, Rodrigues T, Figueiredo MV, et al. Effect of adhesive type and thickness
on the lap shear strength. J Adhes. 2006;82(11):1091–1115.
[7] Banea MD, da Silva LFM, Campilho R. The effect of adhesive thickness on the mechan-
ical behavior of a structural polyurethane adhesive. J Adhes. 2015;91(5):331–346.
[8] Volkersen O. Die nietkraftoerteilung in zubeanspruchten nietverbindungen konstanten
loschonquerschnitten. Luftfahrtforschung. 1938;15:41–47.
[9] da Silva LFM, das Neves PJC, Adams RD, et al. Analytical models of adhesively bonded
joints-part II: comparative study. Int J Adhes Adhes. 2009;29(3):331–341.
[10] Goland M, Reissner E. The stresses in cemented joints. J Appl Mech. 1944;66:A17–A27.
[11] Hart-Smith LJ. Adhesive-bonded single-lap joints. Hampton, VA: Langley Research
Center, NASA; 1973.
[12] Cooper PA, Sawyer JW. A critical examination of stresses in an elastic single lap joint.
Hampton, VA: Langley Research Center, NASA; 1979.
[13] Ojalvo IU, Eidinoff HL. Bond thickness effects upon stresses in single lap adhesive
joints. 1977. [cited 2019 Jun 13]. Available from: https://www.sae.org/content/770090/
[14] Crocombe AD. Global yielding as a failure criterion for bonded joints. Int J Adhes
Adhes. 1989;9(3):145–153.
[15] Her S-C. Stress analysis of adhesively-bonded lap joints. Compos Struct. 1999;47(1–4):
673–678.
[16] Adams RD, Peppiatt NA. Stress analysis of adhesive-bonded lap joints. J Strain Anal.
1974;9(3):185–196.
[17] Roberts TM. Shear and normal stresses in adhesive joints. J Eng Mech. 1989;115(11):
2460–2479.
[18] Adams RD, Peppiatt NA. Effect of poisson’s ratio strains in adherends on stresses of an
idealized lap joint. J Strain Anal. 1973;8(2):134–139.
26 I.J. SÁNCHEZ-ARCE ET AL.

[19] Liew KM, Peng LX, Kitipornchai S. Analysis of symmetrically laminated folded plate
structures using the meshfree galerkin method. Mech Adv Mater Struct. 2009;16(1):
69–81.
[20] Mubashar A, Ashcroft IA. Comparison of cohesive zone elements and smoothed par-
ticle hydrodynamics for failure prediction of single lap adhesive joints. J Adhes. 2017;
93(6):444–460.
[21] Bodjona K, Lessard L. Nonlinear static analysis of a composite bonded/bolted single-lap
joint using the meshfree radial point interpolation method. Compos Struct. 2015;134:
1024–1035.
[22] Farahani BV, Belinha J, Amaral R, et al. Extending radial point interpolating meshless
methods to the elasto-plastic analysis of aluminium alloys. Eng Anal Bound Elem. 2019;
100:101–117.
[23] Dinis L, Natal Jorge RM, Belinha J. The radial natural neighbours interpolators
extended to elastoplasticity. In: Ferreira AJM, Kansa EJ, Fasshauer GE, Leit~ao VMA,
editors. Progress on meshless methods. New York, NY: Springer Science & Business
Media B.V.; 2009. p. 175–198.
[24] Nguyen VP, Rabczuk T, Bordas S, et al. Meshless methods: a review and computer
implementation aspects. Math Comput Simul. 2008;79(3):763–813.
[25] Belinha J. Meshless methods in biomechanics: bone tissue remodelling analysis. Cham:
Springer International Publishing; 2015.
[26] Tsai CL, Guan YL, Ohanehi DC, et al. Analysis of cohesive failure in adhesively bonded
joints with the SSPH meshless method. Int J Adhes Adhes. 2014;51:67–80.
[27] ASTM International Standard. Standard test methods for tensile testing of metallic mate-
rials [Internet]. Vol. E M-03, Annual book of ASTM standards. ASTM Internationa
Standard; 2004. Available from: http://scholar.google.com/scholar?hl=en&btnG=Search&
q=intitle:Standard+Test+Methods+for+Tension+Testing+of+Metallic+Materials#0.
[28] Campilho R, Pinto AMG, Banea MD, et al. Strength improvement of adhesively-bonded
joints using a reverse-bent geometry. J Adhes Sci Technol. 2011;25(18):2351–2368.
[29] da Silva LFM, Giannis S, Adams RD, et al. Manufacture of quality specimens. In: da
Silva LFM, Dillard DA, Blackman B, Adams RD, editors. Testing adhesive joints: best
practices. Weinheim: Wiley-VCH Verlag & Co. KGaA; 2012. p. 1–78.
[30] Campilho R, Banea MD, Pinto AMG, et al. Strength prediction of single- and double-
lap joints by standard and extended finite element modelling. Int J Adhes Adhes. 2011;
31(5):363–372.
[31] Neto JABP, Campilho RDSG, da Silva LFM. Parametric study of adhesive joints with
composites. Int J Adhes Adhes. 2012;37:96–101.
[32] Campilho R, Banea MD, Neto J, et al. Modelling adhesive joints with cohesive zone
models: effect of the cohesive law shape of the adhesive layer. Int J Adhes Adhes. 2013;
44:48–56.
[33] Campilho R, Moura DC, Gonçalves DJS, et al. Fracture toughness determination of
adhesive and co-cured joints in natural fibre composites. Compos Eng. 2013;50:
120–126.
[34] de Sousa C, Campilho R, Marques EAS, et al. Overview of different strength prediction
techniques for single-lap bonded joints. Proc Inst Mech Eng Part L J Mater Des Appl.
2017;231(1–2):210–223.
[35] Faneco TMS, Campilho R, Silva FJG, et al. Strength and fracture characterization of a
novel polyurethane adhesive for the automotive industry. J Test Eval. 2017;45(2):
398–407.
[36] Crocombe AD. Stress analysis. In: Adams RD, editor. Adhesive bonding science, tech-
nology and applications. Cambridge, England: Woodhead Publishing Limited; 2005.
[37] da Silva LFM, Costa M, Viana G, et al. Analytical modelling of the single-lap joint. In:
Campilho RDSG, editor. Strength prediction of adhesively-bonded joints. Boca Raton:
CRC Press; 2017. p. 8–46.
JOURNAL OF ADHESION SCIENCE AND TECHNOLOGY 27

[38] da Silva LFM, das Neves PJC, Adams RD, et al. Analytical models of adhesively bonded
joints-part I: literature survey. Int J Adhes Adhes. 2009;29(3):319–330.
[39] Belinha J, Ara ujo AL, Ferreira AJM, et al. The analysis of laminated plates using dis-
tinct advanced discretization meshless techniques. Compos Struct. 2016;143:165–179.
[40] Nunes SLS, Campilho R, da Silva FJG, et al. Comparative failure assessment of single
and double lap joints with varying adhesive systems. J Adhes. 2016;92(7–9):610–634.
[41] Adams RD, editor. Adhesive bonding: science, technology and applications. Oxford:
Woodhead Publishing Limited; 2005.
[42] Liu Z, Huang Y, Yin Z, et al. A general solution for the two-dimensional stress analysis
of balanced and unbalanced adhesively bonded joints. Int J Adhes Adhes. 2014;54:
112–123.
[43] Davis M, Bond D. Principles and practices of adhesive bonded structural joints and
repairs. Int J Adhes Adhes. 1999;19(2–3):91–105.
[44] Tuzi I, Shimada H. Photoelastic investigation of the stresses in cemented joints. Bull
JSME. 1964;7(26):263–267.
[45] Liu GR, Gu YT. An introduction to meshfree methods and their programming.
Dordrecht, The Netherlands: Springer; 2005. p. 497.

You might also like