See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256972858

Failure criteria for adhesively bonded joints

Article  in  International Journal of Adhesion and Adhesives · September 2012

DOI: 10.1016/j.ijadhadh.2012.01.009

CITATIONS READS

31 3,111

5 authors, including:

René Quispe Rodríguez William Portilho Paiva

Universidade Federal de Santa Maria Federal Institute of Education, Science and Technology of São Paulo
14 PUBLICATIONS   64 CITATIONS    8 PUBLICATIONS   114 CITATIONS   

SEE PROFILE SEE PROFILE

Paulo Sollero Marcelo Bertoni

University of Campinas Embraer
72 PUBLICATIONS   720 CITATIONS    3 PUBLICATIONS   34 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Bonded joint View project

Experimental analysis and numerical modeling of surfaces under degeneration View project

All content following this page was uploaded by René Quispe Rodríguez on 11 October 2017.

The user has requested enhancement of the downloaded file.

 International Journal of Adhesion & Adhesives 37 (2012) 26–36

Contents lists available at SciVerse ScienceDirect

International Journal of Adhesion & Adhesives

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

Failure criteria for adhesively bonded joints

René Quispe Rodrı́guez a, William Portilho de Paiva a, Paulo Sollero a,n,
Marcelo Ricardo Bertoni Rodrigues a, Éder Lima de Albuquerque b
a
Departamento de Mecânica Computacional, Faculdade de Engenharia Mecânica, Universidade Estadual de Campinas, 13083-970 Campinas, SP, Brazil
b
Faculdade de Tecnologia, Universidade de Brası́lia, 70910-900 Brası́lia, DF, Brazil

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

Available online 21 January 2012 Adhesively bonded joints are an alternative structural technique to welded, bolted or riveted joints,
Keywords: presenting better load transfer and stress distributions. Most analytical formulations for stress–strain
Stress distribution analysis consider maximum values as failure criteria. Several failure criteria based on fracture
Finite element stress analysis mechanics and extensive evaluation of experimental data are also available in literature. The main
Lap-shear difficulty with these two criteria is the need to perform numerous tests, resulting in expensive methods
Peel and usually applicable to a particular material combination. This paper reviews several linear and
non-linear analytical models of stress distributions in adhesively bonded joints. These models were
implemented in a user-friendly software. The developed software allows evaluation of stress distribu-
tions for each analytical solution and also features the comparison among these solutions and
numerical results. Comparison among analytical and numerical analyses was achieved by the
interaction between the MATLAB code and ABAQUS models using Python script. Different failure
criteria were also reviewed and implemented. The bond strain energy density criteria combine shear
and peel effects. Results obtained by these criteria are compared with numerical results and
experimental data.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction criteria implemented, several experimental tests in accordance

with the ASTM standard were carried out.
The demand of adhesively bonded joints has increased con-
siderably in recent years. When compared with traditional joining
techniques, bonded joints present many advantages: more 2. Analytical methods
efficient load transfer, possibility to conform light weight struc-
tures, enhanced fatigue properties, improved corrosion resistance, 2.1. Volkersen model
smoother surfaces, among others. Computer programs for the
design of adhesively bonded joints were implemented by da Silva The first analytical method known in literature for the stress
et al. [1] and Dragoni et al. [2]. This paper reviews several linear analysis of bonded joints was developed by Volkersen [3]. The
and non-linear analytical models of stress distributions in adhe- Volkersen’s method, also known as the shear-lag model, intro-
sively bonded joints. These models were implemented in a user- duced the concept of differential shear. The bending effect caused
friendly software. The developed software was named ‘‘KISPEO’’ by the eccentric load path is not considered. The adhesive shear
and was mainly programmed with the GUI (Graphical User stress distribution t is given by
  
Interface) applicative of MATLAB. This software allows numerical Po coshðoxÞ t t t b ol sinhðoxÞ
t¼ ol þ   ð1Þ
analysis of each analytical model and a comparison among them. 2b sinh 2 tt þ tb 2 cosh o2l
Defining some extra variables as geometry or mesh data it is
possible to create a numerical model through an interaction with where
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
the commercial software ABAQUS. This interaction is possible Ga tt
with the use of Python scripts. Different failure criteria were also o¼ 1þ
Ett t a tb
reviewed and implemented. For the validation of the failure
The reciprocal of o has units of length and is the characteristic
shear-lag distance [4]; tt is the top adherend thickness; tb is the
n
Corresponding author. Tel.: þ55 19 3521 3390; fax: þ 55 19 3289 3722. bottom adherend thickness; ta is the adhesive thickness; b is the
E-mail address: sollero@fem.unicamp.br (P. Sollero). bonded area width; l is the bonded area length; E is the adherend

0143-7496/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijadhadh.2012.01.009
 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36 27

     
modulus; Ga is the adhesive shear modulus; P is the applied force. 2k 0 lx lx
B ¼ R1 l þ lk sinhðlÞsinðlÞ sinh sin , and
The origin of x is the middle of the overlap and is shown in Fig. 1. 2 c c
Eq. (1) shows that for a joint with different adherends the sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
adhesive stress is maximum (and thus failure most likely) at the 0 kc P
k ¼ 3ð1n2 Þ
overlap end of the thinner loaded adherend. Further, the lowest t tE
adhesive stresses are obtained when the adherends are identical, is the transverse force factor, also
tt ¼tb ¼t. Assuming that the joint is sufficiently long such that
c Ea t 1
sinh(ol)¼cosh(ol), as mentioned in [5], Eq. (1) will result in l ¼ g , g4 ¼ 6 , D ¼ ðsinð2lÞ þ sinhð2lÞÞ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi t E ta 2
 ffi
oP Ga tt P R1 ¼ coshðlÞsinðlÞ þ sinhðlÞcosðlÞ
t¼ ¼ 1þ ð2Þ
2 Ett t a tb 2 R2 ¼ coshðlÞsinðlÞ þ sinhðlÞcosðlÞ

This is a useful formula which shows a number of important

features about the size of the peak adhesive stress in a single
overlap joint [5] as follows:
2.3. Hart-smith model

 For long joints, it is independent of the joint length.
 It increases with increasing adhesive shear modulus.
 It increases with decreasing adherend modulus and thickness
and adhesive thickness.
2.2. Goland and Reissner model
Goland and Reissner [6] were the first to consider the effects
due to rotation of the adherends, Fig. 2. They divided the problem
In contrast to Volkersen [3] and Goland and Reissner [6],
Hart-Smith [7] considered adhesive plasticity. In the report pre-
sented for the NASA he analyzed both the Single Lap Joint (SLJ) and
the Double Lap Joint (DLJ). For both analyses he combined elastic
peel stress with plastic shear stresses. According to Hart-Smith, the
adhesive elastic shear stress distribution tðxÞ is given by
tðxÞ ¼ A2 coshð2l0 xÞ þ C 2 ð5Þ
where
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s

into two parts: (a) determination of the loads at the edges of the 1 þ 3ð1n2 Þ 2Ga Ga 6ð1n2 ÞM 1
l0 ¼ , A2 ¼ Pþ 0 0 ,
joints, using the finite deflection theory of cylindrically bent 4 t a Et t a Et t 2l sinhð2l cÞ
plates and (b) determination of joint stresses due to the
 
applied loads. 1 A2 t þ ta 1 P
C2 ¼
0
P 0 sinhð2l cÞ , M¼P  2 2 , x2 ¼
The adhesive shear stress distribution t found by Goland and 2c l 2 1þ xc þ x c D
6
Reissner is given by
  and D is the adherend bending stiffness given by D ¼ Et3 =12 ð1n2 Þ.
1 P bc coshððbc=tÞðx=cÞÞ
t¼ ð1 þ 3kÞ þ 3ð1kÞ ð3Þ Variables P, Ga, ta, E, Ea, v, t, c have the same meaning as
8c t sinhðbc=tÞ
presented by Volkersen and Goland and Reissner models. The
where, P is the applied tensile load per unit width, c is half of the adhesive peel stress distribution sðxÞ is given by
overlap length, t is the adherend thickness, n is Poisson’s ratio and
sðxÞ ¼ AcoshðwxÞcosðwxÞ þ BsinhðwxÞsinðwxÞ ð6Þ
k is the bending moment factor
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffi The shear plastic stress was modeled using a bi-linear elastic-
coshðu2 cÞ 3ð1n2 Þ 1 P 2 Ga t perfectly plastic approximation. The overlap was divided into
k¼ pffiffiffi , u2 ¼ , b ¼8
coshðu2 cÞ þ 2 2sinhðu2 cÞ 2 t tE E ta three regions, a central elastic region of length l and two outer
plastic regions. Coordinates x and x0 were defined as shown in
The adhesive peel stress distribution s is given by
Fig. 3.
1 Pt The problem is solved in the elastic region in terms of the
s¼ ½A þ B ð4Þ
D c2 shear stress according to
where tðxÞ ¼ A2 coshð2l0 xÞ þ tp ð1KÞ ð7Þ
     
2k 0 lx lx
A ¼ R2 l þ lk coshðlÞcosðlÞ cosh cos , and the shear strain in the plastic region according to
2 c c
gðx0 Þ ¼ ge f1 þ2K½ðl0 x0 Þ2 þ l0 x0 tanhðl0 dÞg ð8Þ

where tp is the plastic adhesive shear stress and A2 ¼

0
K tp =coshðl dÞ. K and d are solved by an iterative approach using
the following equations:
 
P 0 0 ld 0 0
ðl lÞ ¼ 2l þ ð1KÞðl dÞ þ K tanhðl dÞ ð9Þ
ltp 2
Fig. 1. Volkersen model.

Fig. 2. Goland and Reissner model. Fig. 3. Regions considered by Hart-Smith.

28 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36

 
     
2
ta P 2 ld gp 0 ld
strain at a distance or over a zone, limit state criteria, fracture
1 þ3kð1n2 Þ 1 þ l ¼2 þ K 2l mechanics criteria and damage mechanics criteria [9]. This paper
t tp 2 ge 2
ð10Þ focuses on the first three criteria.
It is worth to point out that the adhesive thickness effect was not
  (  
2 )
gp ld directly analyzed in this work. It is well known that the experi-
0 0 2 0
2 ¼K 2l þ tanhðl dÞ tanh ðl dÞ ð11Þ mental joint strength decreases as the adhesive thickness increases.
ge 2
This is justified in various manners (through thickness variation of
where ge and gp are the elastic and plastic adhesive shear strain, the stresses which is more severe in the case of thick adhesives,
respectively. better polymerization if the adhesive is thin, higher probability of
defects in a thick adhesive layer, among others). Elastic models do
2.4. Ojalvo and Eidinoff model not predict this behavior and show exactly the opposite trend.
Among all models presented in our work, the only one that may
Ojalvo and Eidinoff [8] model is based on Goland and Reissner reproduce the trend of the joint strength is the Elastoplastic model
model. They modified some coefficients in the shear stress of Hart-Smith. However, the study of the adhesive thickness effect
equations by adding new terms to the differential equation and was out of the scope of our work. For more details, the authors
considering new boundary conditions for bond peel stress calcu- recommend works presented by da Silva et al. [10,11].
lation. Their leading work was the first in predicting the variation
of shear stress through the bond thickness. The adhesive non- 3.1. Maximum stress or strain criteria (maximum value criteria)
dimensional shear stress tn distribution found by Ojalvo and
Eidinoff is given by This type of failure criteria is considered to be the most
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  popular and intuitive category for bonded joints. Failure criteria
tn ¼ Acosh l 2 þ6ð1 þ bÞ2 xn þ B ð12Þ have evolved naturally as analytical methods evolved too. For
example, in the Volkersen model, the adhesive was assumed to
where deform just in shear. It becomes natural to consider the maximum
2lð1 þ 3ð1
ffi  þ Þ kÞ2
pbffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi  shear stress as the failure criteria. This approach was also used by
A ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 þ 6ð1 þ bÞ2 sinh l 2 þ 6ð1 þ bÞ2 xn Greenwood et al. [12]. They used the formulation developed by
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Asinh l 2 þ 6ð1 þ bÞ2 Goland and Reissner [6] and found that the maximum shear stress
B ¼ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi
l 2 þ 6ð1 þ bÞ occurs at about 451 across the adhesive layer.
l2 ¼ Ga c2
and b ¼ h Likewise, various quantities have been used to predict bonded
En th t
joint strength:
En ¼E for adherends in plane stress and En ¼ E=1n2 for
adherends in plane strain. Ga, c, E, t are the same variables as  Maximum peel stresses [13,14]. Peel stresses should be mini-
defined previously and h is the adhesive thickness. k is the mized by design rather than used as a design limit on the
bending moment factor as seen in Hart-Smith model. The strength of bonded joints [15]. Hart-Smith [7] considered the
maximum non-dimensional stress at the bond/adherend interfaces maximum peel stress as a failure criterion. However, his
is given by approach is more sophisticated. He distinguished the cases in
which failure occurs in the adhesive (short overlap) or in the
tnn ¼ tn 7 Dtn ð13Þ
adherend (long overlap).
where  The maximum principal tensile stress and strain criteria were
Ga h n0 used in [16], predicting the strength of SLJs to about 10%
Dtn ¼ s accuracy using elastoplastic finite element analysis. These
2Ea
same criteria were used on cleavage and compressive shear
The solution for the non-dimensional peel stress sn is given by test by Crocombe et al. [17].
sn ¼ Csinhða1 xn Þsinða2 xn Þ þD coshða1 xn Þcosða2 xn Þ ð14Þ  The maximum von Mises stress was used by Ikegami et al. [18]
as a failure criterion for bonded scarf joints. This criterion
where
was found ineffective on double lap joints as described by
2 2
3bl r 3bl r 24Ea c4 Charalambides et al. [19]. This criteria is ineffective because
a1 2 ¼ þ , a22 ¼  þ , and r2 ¼ n 3
2 2 2 2 E ht the von Mises criteria do not consider the hydrostatic stress,
which significantly affects the yield and deformation behavior
Constants C and D are obtained upon substitution of the
of polymers.
derivatives of Eq. (14) into Eqs. (15) and (16):
 The maximum shear strain criteria were used by Lee and Lee
sn ð 71Þ6bl2 sn ð 7 1Þ ¼ 8kgr2 ð1 þ bÞ
000 0
ð15Þ [20] and the effective uniaxial plastic strain was used by
00
Crocombe and Adams [21].
sn ð 7 1Þ ¼ kgr2 ð1þ bÞ ð16Þ
where g ¼ t=2c In this work, the maximum shear and peel stresses are
All analyses were done in a non-dimensional basis. Equiva- considered as failure criteria. It is well known that an applied load
lence is given by: tn ¼ t=t, sn ¼ s=t, xn ¼ x=c, where t ¼ P=2c. and some geometry conditions generate a stress distribution
attaining its maximum value at the overlap edge. Then, the applied
load that produces a maximum value of stress equivalent to the
3. Failure criteria maximum admitted in the adhesive will be called failure load.

Alike analytical methods, literature for failure criteria is
extensive. In this paper a small review of the main failure criteria
known in literature for the case of static loads is shown.
Generally, failure criteria can be grouped into the following
categories: maximum stress or strain criteria, critical stress or
3.2. Critical stress or strain at a distance over a zone (finite zone
criteria)
This type of criteria has been naturally adopted due to the high
dependency on the mesh with the previous category (maximum
 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36 29

value criteria), as a result of the inevitable requirement of dealing

with singularities. Zhao et al. [22] used a weighted averaged
maximum stress criterion, where the adhesive thickness is used
as the distance over which the stresses are averaged and then
compared to the adhesive yield strength. Charalambides et al.
[19] lately showed that for double-lap joints, the location of the
maximum stress occurs further down the fillet edge, outside the
averaged zone.
Clarke and Mcgregor [23] stated that for failure to occur, the
maximum principal stress must exceed the ultimate tensile stress of
the adhesive over a finite zone. They used three kinds of geometries
(single-lap, double strap and T-peel joints) to demonstrate that such
zone was independent of the joint geometry. Lately Charalambides
et al. [19] found that predictions for DLJs overestimated experi-
mental data by approximately 68% for long overlaps.
Critical strain at a distance were used by Towse et al. [24] on
DLJs. They used a nonlinear analysis which included the effect of
residual thermal stresses. The joint was demonstrated to fail
when the strain near the singularity reached the adhesive
ultimate strain. Same criteria were used by Towse et al. [25] in
Fig. 4. Adhesive plastification process in the elastic–plastic model of Hart-Smith.
their study of a novel combjoint. For both studies, the applied (For interpretation of the references to color in this figure legend, the reader is
distance had to be determined experimentally. Other authors referred to the web version of this article.)
[26,27] also used experimental data to determine that character-
istic distance. The problem with this approach is that it becomes
unviable to predict the strength of any joint that does not use Making d¼0, (‘‘global yielding’’ state):
same adherends and adhesives characterized previously.
P
P¼ ¼ ltp ð18Þ
b
3.3. Limit state criteria
It means that the failure load depends only on the yield
This type of criteria was first introduced by Crocombe [28]. It strength of the adhesive (tp), the overlap length (l) and the joint
is based on a concept termed global yielding, which applies when width (b). From Eq. (18), we have
a path of adhesive along the overlap region reaches the state in PFAILURE ¼ bltp ð19Þ
which it can sustain no further significant increase in applied
load. To demonstrate the applicability of this concept, three In this model is also considered the maximum shear strain as
different geometries were tested experimentally: SLJ, DLJ and failure criteria. Failure will occur when one of these two failure
compressive shear. Non-linear finite element analyses were criteria is reached.
carried out and the load at which the adhesive layer was
completely yielded was determined. Zhao et al. [22] used this
approach in a CTBN (carboxyl terminated butadiene nitrile) 3.4. Failure envelopes
toughened adhesive lap joint. Schmit and Fraisse [29] include
global yielding as one of the possible failure mechanisms in their This failure criterion was developed by Tong [31]. The author
analysis of stepped double lap joints. This criterion is only established that failure occurs when the maximum bond strain
applicable to a limited range of adhesive joints. The majority of energy rate in the adhesive attains its ultimate value UC for a
structural epoxy adhesives do not have sufficient ductility for the combined shear and peel strain. A small extract of this theory is
entire layer to yield prior to joint failure [30]. presented herewith. An alternative failure criterion was proposed
As described above, in the analytical methods, the elastic– by Chen et al. [32] based on an specific strain energy criterion.
plastic model of Hart-Smith divides the overlap into three regions, Bond strain energy density is defined as the sum of the area
a central elastic region of length l and two outer plastic regions of under the shear stress–strain curve for a given shear strain and
length d. The plastic zone increases as the applied load increases the area under the peel stress–strain curve for a given peel strain.
too. This pattern can be observed in Fig. 4. Red lines indicate the Similarly, the bond strain energy rate is defined as the product of
elastic behavior of the stress distribution, while the blue line adhesive thickness and the bond strain energy density. For a more
indicates the plastification process at the yield stress of the general case, it can be defined as the integration of the bond strain
adhesive. energy density over the adhesive thickness. For a given shear and
When the adhesive has plastified totally (d ¼0), the joint peel strain, the bond strain energy rate is given by [31]
reaches a level where no more load can be applied. This state is U ðCÞ ¼ U II þ U I ð20Þ
called ‘‘global yielding’’ and means that the adhesive has yielded
completely [28]. As a consequence, failure will take place as soon where C is the strain energy ratio defined by
as the state of ‘‘global yielding’’ is reached. The formulation sffiffiffiffiffiffi!
proposed by Hart-Smith [7] is complex and it requires an iterative U II
C ¼ arctan ð21Þ
process to be solved. However, it is possible to find a failure load, UI
by applying some simplifications and without solving the pro-
blem completely. This process is shown below. UI and UII are the bond peel and shear strain energy rates,
Re-writing Eq. (9): respectively. The strain energy ratio is 01 when the adhesive is
  subjected to pure peel strain and 901 when subjected to pure
P 0 0 ld 0 0 shear strain, and varies from 01 to 901 when subjected to
ðl lÞ ¼ 2l þ ð1KÞðl dÞ þ K tanhðl dÞ ð17Þ
ltp 2 combined shear and peel strains.
30 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36

Then, failure will occur when Re-writing Eqs. (26) and (27) for a single lap configuration and
considering equal adherends (t1 ¼t2 ¼t), we have
U ðCÞ ¼ U C ðCÞ ð22Þ
 þ 2
A1 A2 t N1 tM 1þ
where UC(C) is the adhesive failure envelope that is defined as the U II ¼ þ ð28Þ
8t a ðA1 þ A2 Þ A1 2D1
critical strain energy rate of the adhesive corresponding to
various combinations of shear and peel strains. It is desirable to D2 t
UI ¼ ðM1þ Þ2 ð29Þ
measure the whole in situ failure envelope for a specific material 2t a ðD1 þD2 ÞD1
combination. However, measurement of such in situ failure
For the implemented models analyzed previously, it is known
envelope requires enormous effort and is only applicable to the
that the bending moment (M) is dependent on the applied load
particular material combination, thus can be very expensive. For
per unit width ðPÞ, the adherend thickness (t) and on the bending
this reason, the following interactive failure criterion was used:
moment factor (k). This bending moment factor varies from
    method to method. The analytical model of Volkersen does not
U II a UI b
þ ¼1 ð23Þ consider this factor, because the rotation of the adherends is not
U IIC U IC
considered. The model of Goland and Reissner [6] was the first to
where a and b are real constants, and UIC and UIIC are the critical take into consideration this effect, in this model the bending
bond strain energy rates of pure peel and pure shear, respectively. moment factor (k) is calculated as follows:
Tong [31] defined two simple failure criteria: coshðu2 cÞ
Linear failure criterion (a ¼ b ¼1): kG ¼ pffiffiffi ð30Þ
coshðu2 cÞ þ 2 2sinhðu2 cÞ
U II
þ
UI
¼1 ð24Þ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffi
U IIC U IC 3ð1n2 Þ 1 P
u2 ¼ ð31Þ
2 t tE
Quadratic failure criterion (a ¼ b ¼2):
The formulation developed by Hart-Smith [7] considers the
 2  2 effect of big deformations and presents an alternative procedure
U II UI
þ ¼1 ð25Þ to calculate the factor k
U IIC U IC
 
ta 1
Tong [31] deduced analytical formulations for the shear strain kH ¼ 1 þ ð32Þ
t 1 þ xc þðx2 c2 =6Þ
energy rate UII and peel strain energy rate UI for single lap joints
and lap-shear configurations. He also demonstrated that these where
energies depend only on the longitudinal membrane forces
P
Ni(i¼ 1,2), bending moments Mi, elasticity modulus Ei and adher- x2 ¼ ð33Þ
D
end thickness ti. Fig. 5 shows the two configurations considered
in [31]. Once k is defined, it is possible to determine the bending
moment, and with this, strain energy densities could be defined
 þ 2
A1 A2 N1 t 1 M 1þ for each analytical method. A simplified formulation for the evalua-
U II ¼ þ ð26Þ tion of the bending moments in SLJs is presented by Zhao et al. [33].
8ðA1 þ A2 Þ A1 2D1
For the Goland and Reissner model:
" !#2
D2 ðt 1 D2 t 2 D1 Þ ðM 1þ Þ2 A1 A2 P t2 P coshðu2 cÞ
UI ¼ ðM1þ Þ2  ð27Þ uGII ¼ þ pffiffiffi ð34Þ
2ðD1 þD2 ÞD1 2ðD1 þD2 Þðt 1 þ t 2 Þ D1 8t a ðA1 þ A2 Þ A1 4D1 coshðu2 cÞ þ 2 2sinhðu2 cÞ
where Ai ¼Eiti and Di ¼ Ei t 3i =12ði ¼ 1,2Þ are the membrane and !2
bending stiffness of the adherends. D2 coshðu2 cÞ Pt
uGI ¼ pffiffiffi ð35Þ
2t a ðD1 þ D2 ÞD1 coshðu2 cÞ þ2 2sinhðu2 cÞ 2

For the Hart-Smith model:

8 2 392
<P   =
A1 A2 t2 P 4 ta 1
uHII ¼ þ 1þ  2 2 5 ð36Þ
8t a ðA1 þ A2 Þ :A1 4D1 t 1 þ xc þ x c ;
6

"  #2
D2 ta 1 Pt
uHI ¼ 1þ ð37Þ
2t a ðD1 þD2 ÞD1 t 1 þ xc þðx2 c2 =6Þ 2

Keeping in mind failure criteria the following are included:

Linear failure criterion:
uII uI
þ ¼1 ð38Þ
uIIC uIC
Quadratic failure criterion:
   2
uII 2 uI
þ ¼1 ð39Þ
uIIC uIC
Total strain energy density criterion:

Fig. 5. (a) Stress resultants acting on both ends of an overlap in a SLJ. (b) Stress
uI þuII
¼1 ð40Þ
resultants acting on both ends of an overlap in a lap-shear joint [31]. uIC
 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36 31

where uIC and uIIC are experimentally determined, being the area make a comparison between two or more analytical methods.
under the peel and shear stress–strain curve, respectively. The Here, options such as numerical analysis and a comparison
strain energy densities uI and uII may be evaluated analytically between analytical and numerical models are also available.
from Eqs. (34–37). Alternatively, uI and uII may be evaluated Fig. 10 shows an example of an individual analysis [6]. For this
using ABAQUS. analysis, only the number of analysis points is required, and all
remaining data are obtained from data introduced in the data
input window. In this step, it is possible to evaluate the stress
4. Software distribution at a particular point and to print the corresponding
result in a n.txt file.
The implementation and understanding of analytical methods
and failure criteria in adhesively bonded joints is not trivial. The
complexity of the formulation turns the implementation difficult.
The main objective of the developed software was to offer a
friendly way to find stress distributions and failure loads without
the neccesity of going into details in analytical or numerical
formulations. The software was denomined ‘‘KISPEO’’ and was
mainly programmed with the GUI (Graphical User Interface)
applicative of MATLAB. With the use of the software it is possible
to carry out an individual analysis of each analytical method
implemented as well as a comparison between these models.
Defining some extra variables as geometry or mesh data it is
possible to create a numerical model through an interaction with
the commercial software ABAQUS. This interaction is possible
with the use of Python scripts. Results from ABAQUS are returned
to the software and then compared with analytical models. Fig. 6
shows a flow chart of the software.
Fig. 7 shows the presentation of the software, while Fig. 8
shows the data input window. In this window, data such as
material properties, geometry and the applied load are intro-
duced. In this window an interaction with a data base is also
included.
Fig. 9 shows the main window. In this window, it is possible to
choose an analytical model for stress distribution analysis, or to Fig. 7. Software presentation.

Fig. 6. Flow chart of the developed software.

32 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36

Fig. 8. Data input window.

Fig. 9. Main software window.

Fig. 11 shows the numerical analysis window. Here, data such
as the geometry or mesh parameters are introduced. Boundary
conditions are considered as the same in a real ASTM D1002 test.
All these data are saved in a n.txt file. This file is imported using a
Python script, allowing numerical analysis using the commercial
software ABAQUS.
The software uses a regular structured mesh. The mesh is
generated by the ABAQUS pre-processor with the mesh density
introduced by the user. Convergence may be achieved by the user,
increasing the number of elements and comparing the behavior of
peak stresses.
Then, results obtained from the numerical analysis are again
returned to the software for the comparison between analytical
and numerical analysis, as shown in Fig. 12.
Fig. 13 shows the failure criteria window. Here, data input for
the failure criteria implemented are introduced. It is possible to
analyze the bonded joint using just the analytical or the numer-
ical formulation. If numerical analysis is chosen, a n.txt file is
generated with the numerical input data for ABAQUS analysis. If
analytical analysis is chosen, a n.txt file is created with results
from this analysis.
5. Results
5.1. Experimental tests
For the validation of the implemented failure criteria, several
experimental tests were carried on in accordance with ASTM
standards D1002, D5656 and D2095. Materials in these tests were
Aluminum 2024-T3 for adherends and the AF163.2K by 3M for
the adhesive.
The ASTM D1002 is the most common shear test. The main
objective of this test is to obtain the experimental failure load of a
SLJ. This experimental failure load is then compared to analytical
and numerical results. Six specimens were tested and the failure
load was 11.3 71.5% kN.
The ASTM D5656 is a pure shear test, unlike the ASTM D1002,
where the peel stresses are almost eliminated due to the high
thickness of the adherends and the small length of the overlap.
The main objective is to find the shear strain energy density (area
under the shear–stress curve), which is an input data in the
energy-based failure criteria. Six specimens were tested and the
mean value of the shear strain energy density was evaluated,
 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36 33

Fig. 10. Goland and Reissner analysis.

Fig. 11. Numerical analysis window.

obtaining 64.5 710% MJ/m3. Some other parameters can also be
obtained from this test, such as the plastic adhesive shear stress
tp, the elastic adhesive shear strain ge and the plastic adhesive
shear strain gp. These parameters were obtained from the equiva-
lence between the real strain–stress curve and an elastic-
perfectly-plastic idealized curve, as can be seen in Fig. 14, where
A is the shaded area.
The ASTM D2095 test, best known as ‘‘Butt joint test’’, was
assumed here as a pure traction test. A more precise evaluation of
the tensile stress distribution is shown in [34]. The objective of
this test is to obtain the experimental peel strain energy density,
which is the area under the stress–strain curve. Five specimens
were tested, and the mean value of the peel strain energy was
evaluated, obtaining 38.2 715% MJ/m3.
34 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36

Fig. 12. Comparison between analytical and numerical models.

Fig. 13. Failure criteria window.

Table 1 shows all parameters that were taken from normalized
tests. Some values were directly provided by the adhesive
manufacturer and are identified with the symbol n.
5.2. Maximum value criterion
Table 2 shows failure loads. The Error percentage (% Error) is
defined as
PFAILURE Pexperimental
%Error ¼  100%
Pexperimental
where PFAILURE is the failure load and Pexperimental is the experi-
mental failure load. The failure loads were evaluated according to
the maximum shear stress criterion (Pt) or the maximum peel
stress (Ps), whenever it was possible. The lower load of these two
criteria was considered as the failure load (PFAILURE). Fig. 15 shows
graphically failure loads for each analytical method.
5.3. Global yielding criterion
This failure criterion, developed by Crocombe [28], takes into
consideration the material non-linearity in the adhesive. The
failure load was equivalent to 13.8 kN, presenting an error of
 22.3%, when it is compared with the experimental failure load.
This value is high, however, it shows a better approximation
when it is compared with failure loads found in the
previous group.
ð41Þ
5.4. Energy-based criteria
Table 3 shows failure loads and errors (%) for the three energy-
based failure criteria. The analytical models of the energy based
failure criteria are presented in Section 3.4. Strain energy
densities, according to each failure criterion are evaluated
numerically from Eqs. (34)–(37). The maximum total strain
 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36 35

Table 3
Failure loads considering energy-based criteria.

Method Pru (% Error) Prl (% Error) Prq (% Error)

Goland and Reissner 15.39 kN 10.85 kN 13.37 kN

(  36.4) (3.8) ( 18.4)
Hart-Smith (Elastic) 15.12 kN 14.37 kN 18.29 kN
(  33.9) ( 27.3) ( 62.1)
Numerical (ABAQUS) 9.84 kN 11.33 kN 13.46 kN
(12.8) ( 0.4) ( 19.3)

Fig. 14. Equivalence between real and idealized curve for the ASTM D5656
standard.

Table 1
Parameters taken from normalized tests.
Fig. 16. Failure loads for energy-based criteria.
Variable Description Value Standard test

PFAILURE Experimental failure load 11.28 kN ASTM D1002

zMAX Maximum shear stress 48.51 MPa ASTM D5656
According to Table 3, the failure criterion that presented the
gUL Maximum shear strain 1.55 ASTM D5656
ge Elastic adhesive shear strain 0.08 ASTM D5656 best fit to experimental failure load of the ASTM D1002 test is the
gp Plastic adhesive shear strain 1.46 ASTM D5656 linear failure envelope. This failure criterion predicted failure
zp Plastic adhesive shear stress 42.78 MPa ASTM D5656 loads with absolute value errors in the range of 0.4–27.3%. The
sMAX Maximum peel stress 48.26 MPa ASTM D638n quadratic failure envelope had the worst approximation to
uII Shear strain energy density 64.49 MJ/m3 ASTM D5656
uI Peel strain energy density 38.18 MJ/m3 ASTM D2095
experimental data, with errors in the range of 18.4–62.1%.
E Adhesive elasticity modulus 1110 MPa ASTM D3039n
v Adhesive Poisson’s ratio 0.34 ASTM D3039n
6. Conclusions

Table 2 There are many analytical methods and failure criteria avail-
Failure loads considering maximum stresses as failure criteria.
able in literature for bonded joint analysis. In this paper four
Analytical method Pz (% Error) Ps (% Error) PFAILURE (% Error) analytical methods were implemented: Volkersen [3], Goland and
Reissner [6], Hart-Smith [7] and Ojalvo and Eidinoff [8]. Then,
Volkersen 10.77 kN (4.5) – 10.78 kN (4.5) each analytical method was associated with the best failure
Goland and Reissner 7.65 kN (32.2) 6.13 kN (45.7) 6.13 kN (45.7) criteria. Methods implemented were considered sufficient to
Hart-Smith (Elastic) 7.72 kN (31.6) 7.24 kN (35.8) 7.24 kN (35.8)
Ojalvo and Eidinoff 7.70 kN (31.7) 5.09 kN (54.9) 5.09 kN (54.9)
achieve a consistent result, which would be useful for prelimi-
nary design purposes and as a consequence would reduce
costly tests.
One of the contributions of this work was the development of
a software. This software, that was named ‘‘KISPEO’’, allows the
calculation of stress distributions and failure loads for a SLJ
configuration. The programming was accomplished using the
GUI applicative of the commercial software MATLAB. This soft-
ware allows the use of friendly interfaces. Python scripts were
used for the interaction between analytical and numerical
models.
Several standard tests, such as, the ASTM D1002, ASTM D5656
and ASTM D2095 were carried out. The main objective of these
tests was to characterize adhesive parameters. These parameters
were used as data input for failure criteria implemented over this
work. Failure loads obtained from the failure criteria implemen-
ted were then compared with the experimental failure load
Fig. 15. Failure loads for the maximum value criterion.
obtained from the standard test ASTM D1002.
The analytical failure criterion that best predicted the experi-
energy density load (Pru) was evaluated from Eq. (40). The linear mental failure load was the linear failure envelope that consid-
failure envelope load (Prl) was evaluated from Eq. (38) and the ered the Goland and Reissner [6] formulation, according to
quadratic failure envelope load (Prq) was evaluated from Eq. (39). Eqs. (34), (35) and (38), with 3.8% of error. Numerical failure
These failure loads are also shown graphically in Fig. 16. criteria were on average more accurate than analytical failure
36 R. Quispe Rodrı́guez et al. / International Journal of Adhesion & Adhesives 37 (2012) 26–36
criteria. Among these criteria, the most accurate was the linear
failure envelope, with  0.4% of error.
References
[1] da Silva LFM, Lima RFT, Teixeira RMS. Development of a computer program
for the design of adhesive joints. J Adhes 2009;85:889–918.
[2] Dragoni E, Goglio L, Kleiner F. Designing bonded joints by means of the
JointCalc software. Int J Adhes Adhes 2010;30:267–80.
[3] Volkersen O. Luftfahrtforschung 1938:15–41.
[4] Shigley JE, Mischke CR. Adhesive bonding and design considerations. In:
Mechanical engineering design, 6th ed. McGraw-Hill; 2003. p. 562–78
[Chapters 9–11].
[5] Crocombe AD, Ashcroft I. Simple lap joint geometry. In: da Silva LFM, Öchsner
A, editors. Modeling of adhesively bonded joints. Springer; 2010.
p. 3–23. [Chapter 1].
[6] Goland M, Reissner E. J Appl Mech 1944;66:A17.
[7] Hart-Smith LJ. NASA contract report, NASA TR-112236; 1973.
[8] Ojalvo IU, Eidinoff HL. Bond thickness effects upon stresses in single-lap
adhesive joints. AIAA J 1978;16:204–11.
[9] Randolph A, Clifford M. Improved 2D model for bonded composite joints. Int J
Adhes Adhes 2004;24:389–405.
[10] da Silva LFM, Rodrigues TNSS, Figueiredo MAV, de Moura MFSF, Chousal JAG.
Effect of adhesive type and thickness on the lap shear strength. J Adhes
2006;82:1091–115.
[11] da Silva LFM, das Neves PJC, Adams RD, Wang A, Spelt JK. Analytical models
of adhesively bonded joints—part II: comparative study. Int J Adhes Adhes
2009;29:331–41.
[12] Greenwood L, Boag T, McLaren A. Stress distribution in lap joints. In: Adhesion:
fundamentals and practice. London: McLaren and Sons Ltd.; 1969. p. 273–9.
[13] Crocombe A, Tatarek A. A unified approach to adhesive joint analysis. In:
Proceedings of adhesives, sealants and encapsulants 85. London: Plastic and
Rubber Institute; 1985.
[14] Adams RD, Gregory DA, Panes GA. Effect of three dimensional stresses on the
failure of single lap joints. In: Proceedings of Euradh; 1994. p. 228–31.
[15] Hart-Smith LJ. Designing to minimize peel stresses in adhesive bonded joints.
In: Johnson WS, editor. Delamination and debonding of materials. Pittsburgh;
1985. p. 238–66.
[16] Harris J, Adams RD. Strength prediction of bonded single lap joints by non-
linear finite element methods. Int J Adhes Adhes 1984;4:65–78.
[17] Crocombe A, Bigwood D, Richardson G. Analysing structural adhesive joints
for failure. Int J Adhes Adhes 1990;10:167–78.
View publication stats
[18] Ikegami K, Takeshita T, Matsuo K, Sugibayashi T. Strength of adhesively
bonded scarf joints between glass fibre reinforced plastics and metals. Int J
Adhes Adhes 1990;103:199–206.
[19] Charalambides M, Kinloch A, Matthews F. Strength prediction of bonded
joints. In: Proceedings of AGARD conference on bolted/bonded joints in
polymeric composites; 1997.
[20] Lee S, Lee G. Development of a failure model for the adhesively bonded
tubular single lap joint. J Adhes 1992;40:1–14.
[21] Crocombe A, Adams RD. An elastoplastic investigation of the peel test.
J Adhes 1982;13:241–67.
[22] Zhao X, Adams RD, da Silva LFM. Single lap joints with rounded adherend
corners: experimental results and strength prediction. J Adhes Sci Technol
2011;25:837–56.
[23] Clarke J, Mcgregor I. Ultimate tensile stress over a zone: a new failure
criterion for adhesive joints. J Adhes 1993;42:227–45.
[24] Towse A, Davies R, Clarke A, Wisnom M, Adams RD, Potter K. The design and
analysis of high load intensity adhesively bonded double lap joints. In:
Proceedings of the fourth international conference on deformation and
fracture of composites, Manchester; 1997. p. 479–88.
[25] Towse A, Potter K, Wisnom M, Adams RD. A novel comb joint concept for
high strength unidirectional carbon fibre bonded joints. In: Proceedings of
the 11th international conference on composite materials, Goald Coast; 1996.
p. 95–101.
[26] Trantina G. Fracture mechanics approach to adhesive joints. J Compos Mater
1972;6:192–207.
[27] Chow C, Lu T. Analysis of failure properties and strength of structural
adhesive joints with damage mechanics. Int J Damage Mech 1992;1:
404–34.
[28] Crocombe A. Global yielding as a failure criteria for bonded joints. Int J Adhes
Adhes 1989;9:145–53.
[29] Schmit F, Fraisse P. Fracture mechanics analysis of the strength of bonded
joints. In: Materiaux et techniques, vol. 80; 1992. p. 55–63.
[30] Crocombe A, Kinloch A. MTS adhesives project 2: failure modes and criteria.
Review of adhesive bond failure criteria. Report 1; August 1994.
[31] Tong L. Strength of adhesively bonded single-lap and lap-shear joints. Int J
Solids Struct 1998;35:2601–16.
[32] Chen Z, Adams RD, da Silva LFM. Prediction of crack initiation and propaga-
tion of adhesive lap joints using an energy failure criterion. Eng Fract Mech
2011;78:990–1007.
[33] Zhao X, Adams RD, da Silva LFM. A new method for the determination of
bending moments in single lap joints. Int J Adhes Adhes 2010;30:63–71.
[34] da Silva LFM. Failure strength tests. In: da Silva LFM, Öchsner A, Adams RD,
editors. A handbook of adhesion technology. Springer; 2011.