Composites Science and Technology 62 (2002) 1011–1031

www.elsevier.com/locate/compscitech

Analysis of adhesive bonded joints: a unified approach

F. Mortensen, O.T. Thomsen*
Institute of Mechanical Engineering, Aalborg University, Pontoppidanstraede 101, DK-9220 Aalborg East, Denmark

Received 2 October 2001; accepted 28 January 2002

Abstract
This paper presents a newly developed unified approach for the analysis and design of adhesive bonded joints. The adherends are
modelled as beams or wide plates in cylindrical bending, and are considered as generally orthotropic laminates using classical
laminate theory. Consequently, adherends made as asymmetric and unbalanced composite laminates can be included in the analy-
sis. The adhesive layer is modelled in two ways. The first approach assumes the adhesive layer to be a linear elastic material and the
second approach takes into account the inelastic behaviour of many adhesives. The governing equations are formulated in terms of
sets of first order ordinary differential equations. The multiple-point boundary value problem constituted by the differential equa-
tions together with the imposed boundary conditions is solved numerically by direct integration using the ‘multi-segment method’
of integration. The approach is validated by comparison with finite element models and a high-order theory approach. # 2002
Elsevier Science Ltd. All rights reserved.
Keywords: Adhesive bonded joints; C. Laminates

1. Introduction ment method, is an elaborate and computational

The use of polymeric fibre reinforced composite
materials has gained widespread acceptance as an
excellent way to obtain stiff, strong and very lightweight
structural elements. However, load introduction into
composite structural elements through joints, inserts
and mechanical fasteners is associated with considerable
difficulties. The primary reason for this is the layered
structure of composite laminates, which results in poor
strength properties with respect to loading by inter-
laminar shear and transverse normal stresses. Thus, the
interaction between composite elements and adjoining
parts often proves to be among the most critical areas of
a structural assembly.
Joining of composite structures can be achieved
through the use of bolted, riveted or adhesive bonded
joints. The performances of the mentioned joint types
are severely influenced by the characteristics of the
layered composite materials, but adhesive bonded joints
provide a much more efficient load transfer than
mechanically fastened joint types. Accurate analysis of
adhesive bonded joints, for instance using the finite ele-
* Corresponding author. Tel.: +45-9635-9319; Fax :+45-9815-
1411.
E-mail address: fm@ime.auc.dk or ott@ime.auc.dk (O.T. Thomsen).
0266-3538/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved.
PII: S0266-3538(02)00030-1
demanding task, see Crocome and Adams [1] and Harris
and Adams [2], and there is a specific need for analysis
and design tools that can provide accurate results with
little computational efforts involved. Such tools would
be very useful for preliminary design purposes, i.e. in
the stages of design where fast estimates of stress and
strain distributions as well as joint strengths and margin
of failure are needed.
The first attempt of analyzing adhesive bonded joints
was carried out by Volkersen [3] who modelled the
adhesive layer as continuously distributed shear springs.
The modelling was later refined by adopting a two-
parameter elastic foundation approach, in which the
adhesive layer is described in terms of uniformly dis-
tributed transverse normal and shear springs. This type
of model was originally suggested by Goland and
Reissner [4] for the analysis of single lap joints, but has
later formed the basis for many investigations of the
structural response of various types of adhesive bonded
joints such as Hart-Smith [5–7], Tong [8], Yuceoglu and
Updike [9–12], Renton and Vinson [13,14], Pickett [15,
16] and Thomsen [17,18].
A different approach that overcomes the limitations
of the elastic foundation formulation has been sug-
gested by Frostig et al. [19] following a high-order
theory approach.
1012 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031
Most of the references reviewed, except for Hart-
Smith [6], Pickett [15] and Thomsen [17,18] assume the
adhesive layer to behave as a linear elastic material.
Analyses based on this assumption do provide useful
information about the intensities of stress concentra-
tions and their location, but do not include the fact that
most polymeric structural adhesives behave non-lin-
early, and that many of them show pronounced viscous
and plastic effects even at temperatures below their glass
transition temperature. Thus, the results do not reflect
the true stress distribution at appreciable levels of load-
ing because of the non-linear behaviour of the poly-
meric structural adhesives.
A full treatment of the visco-elastic–plastic behaviour
of the polymeric structural adhesives, characterized by
their sensitivity towards strain rate and environmental
effects such as temperature and moisture, is a very difficult
task, and not many investigations have been carried out on
this topic. Instead, several investigations using an elasto-
plastic approach which neglects strain rate and tempera-
ture effects and includes plastic residual strains have been
carried out, see Hart-Smith [5–7], Pickett [15,16], Adams et
al. [20], Gali and Ishi [21] and Thomsen [17,18].
The main objective of the present paper is to consider
a newly developed unified approach for the analysis of
adhesive bonded joints. The approach can be used for the
analysis of most types of adhesive bonded joints. The
approach has been used for the analysis of the following
types of joints (see Mortensen and Thomsen [22–24] and
Mortensen [25]): single lap joints, scarfed single lap joints,
bonded doubler lap joints, double lap joints, stepped lap
joints (single and double sided) and scarfed lap joints
(single and double sided). The different structural bonded
joints that have been modelled using the proposed unified
approach are shown schematically in Figs. 1 and 2.
In the analysis the adherends are modelled as beams
or wide plates in cylindrical bending and are considered
as generally orthotropic laminates using classical lami-
nate theory. Consequently, the effect of having adher-
ends made as asymmetric and unbalanced composite
laminates is included in the analysis. The adhesive layer
is modelled as continuously distributed linear tension/
compression and shear springs. As non-linear effects in
the form of adhesive plasticity play an important role in
the load transfer, the analysis allows inclusion of non-
linear adhesive properties. The load and boundary con-
ditions can be chosen arbitrarily.
For simplicity, the description of the unified approach
is restricted to the case of a single lap joint. For a
detailed description of the approach used on all the joint
types mentioned above see Mortensen [25].
Analysis procedures for all the joint types mentioned
have been developed and implemented in the new com-
posite analysis and design software package ESAComp
2.0. ESAComp, which is being developed for the Euro-
pean Space Agency, provides an easy-to-use environment
for preliminary evaluation and analysis of plies, lami-
nates as well as structural composite components. The
developed analysis tools for adhesive bonded joints are
completely integrated in the ESAComp environment,
which offers complete access to the ESAComp design
system, see Saarela et al. [26]. Thus, the system provides
access to for instance linear/non-linear adhesive load
response analysis, adhesive/adherend failure analysis etc.
2. Structural modelling
The structural modelling is carried out by adopting a
set of basic restrictive assumptions for the behaviour of
bonded joints. Based on those from which the constitutive
and kinematic relations for the adherends are derived, and
constitutive relations for the adhesive layers are adopted.
Finally, the equilibrium equations for the joints are
derived, and by combination of these equations and
relations, the set of governing equations is obtained.
This results in a set of first order ordinary differential
equations, called the governing system equations,
describing the system behaviour. The governing system
equations are solved numerically using the ‘multi-seg-
ment method of integration’.
As an example of one of the different bonded joint
types, Fig. 3 shows a single lap composed of similar or
dissimilar generally orthotropic laminates subjected to
general loading conditions.
2.1. Basic assumptions for the structural modelling
The basic restrictive assumptions adopted for the
structural modelling are the following:

The adherends:
 Beams or plates in cylindrical bending, which
are described by use of ordinary ‘Kirchhoff’;
plate theory (‘Love–Kirchhoff’; assumptions).
 Generally orthotropic laminates using classi-
cal lamination theory (e.g. asymmetric and
unbalanced composite laminates can be
included in the analysis).
 The laminates are assumed to obey linear
elastic constitutive laws.
 The strains are small, and the rotations are
very small.

The adhesive layer:
 Modelled as continuously distributed linear
tension/compression and shear springs.
 Inclusion of non-linear adhesive properties,
by using a secant modulus approach for the
nonlinear tensile stress–strain relationship in
conjunction with a modified von Mises yield
criterion.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1013

Load and boundary conditions: longitudinal coordinate only. As a consequence of this,

 Can be chosen arbitrarily. the displacement field in the width directions will be
uniform. Thus, the displacement field can be described
The system of governing equations is set up for two as:
different cases, i.e. the adherends are modelled as plates
in cylindrical bending or as wide beams. ui0 ¼ ui0 ðxÞ; vi0 ¼ vi0 ðxÞ; wi ¼ wi ðxÞ ð1Þ

2.2. Modelling of adherends as plates in cylindrical
bending
For the purposes of the present investigation, and
with references to Fig. 3, cylindrical bending can be
defined as a wide plate (in the y-direction), where the
displacement field can be described as a function of the
where u0 is the midplane displacement in the long-
itudinal direction (x-direction), v0 is the midplane dis-
placement in the width direction (y-direction), and w is the
displacement in the transverse direction (z-direction). The
displacement components u0, v0, w are all defined relative
to the middle surfaces of the laminates, and i corresponds
to the laminate/adherend number (see Fig. 3).

Fig. 1. Schematic illustration of the different standard bonded joints analysed using the developed unified approach.
1014 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

Fig. 2. Schematic illustration of the different advanced bonded joints analysed using the developed unified approach.

As a consequence of this, the following holds true:

ui0;y ¼ vi0;y ¼ w;iy ¼ w;iyy ¼ 0 ð2Þ

The boundary conditions at the boundaries in the

width direction are not well defined within the concept
of ‘cylindrical bending’’. However, it is assumed that
there are some restrictive constraints on the boundaries,
such that they not are capable of moving and rotating
freely. This is illustrated conceptually in Fig. 4. It
should be noted that the concept of ‘cylindrical bend-
ing’; is not unique, and that other definitions than the
one used in the present formulation can be adopted, see
Whitney [27].
Substitution of the quantities in Eq. (2) into the con- Fig. 3. Schematic illustration of adhesive single lap joint with straight
stitutive relations for a laminated composite material adherends in the overlap zone subjected to general loading conditions.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1015

2.3. Modelling of adherends as beams

Modelling of the adherends as wide beams can be

considered as a special case of cylindrical bending.
When the adherends are modelled as beams the width
direction displacements are not considered, and only the
longitudinal and vertical displacements are included.
Thus, the displacement field in Eq. (1) is reduced to:

u0 ¼ u0 ðxÞ; w ¼ wðxÞ ð5Þ

For this case the constitutive relations for a composite

beam are reduced to:

i i i i
Nxx ¼ A11 u0;x  B11 w;ixx
i i i i ð6Þ
Mxx ¼ B11 u0;x  D11 w;ixx
Fig. 4. Schematic illustration of adhesive single lap joint ‘clamped’
between two vertical laminates, which prevent the adherends of the
single lap joint from moving and rotating freely in the width direction.
This represents the conceptual interpretation of cylindrical bending as The kinematic relations [(Eq. (4)] are the same as for
defined in the present formulation. the cylindrical bending case except that all variables
associated with the width direction are nil.
(Whitney [27]) gives the constitutive relations for a
laminate (i) in cylindrical bending: 2.4. Constitutive relations for the adhesive layer

i i i i i i The coupling between the adherends is established

Nxx ¼ A11 u0;x þ A16 v0;x  B11 w;ixx
through the constitutive relations for the adhesive layer,
i i i i i i
Nyy ¼ A12 u0;x þ A26 v0;x  B12 w;ixx which as a first approximation is assumed to be homo-
i
Nxy i i
¼ A16 i i
u0;x þ A66 i
v0;x  B16 w;ixx geneous, isotropic and linear elastic. The constitutive
i i i i i i
ð3Þ relations for the adhesive layer are established by use of
Mxx ¼ B11 u0;x þ B16 v0;x  D12 w;ixx
a two-parameter elastic foundation approach, where the
i i i i i i
Myy ¼ B12 u0;x þ B26 v0;x  D12 w;ixx adhesive layer is assumed to be composed of con-
i
Mxy i i
¼ B16 i i
u0;x þ B66 i
v0;x  D12 w;ixx tinuously distributed shear and tension/compression
springs. The constitutive relations of the adhesive layer
are suggested in accordance with Thomsen [17,18] and
where Ajki , Bjki and Djki ðj; k ¼ 1; 2; 6Þ are the extensional, Tong [8]:
coupling and the flexural rigidities as defined by classi-
i i i
cal lamination theory, see Jones [28]. Nxx , Nyy and Nxy   9
i i
are the in-plane stress resultants and Mxx , Myy and Mxy i Ga
i j
 Ga i ti ðxÞ i j tj ðxÞ j >>
ax ¼ u u ¼ u0    u0   >
are the moment resultants. ta ta 2 x 2 x > >
>
>
=
For the advanced joint types such as a scarfed or Ga
i j
 G a
i j 
stepped lap the rigidities Ajki , Bjki and Djki ðj; k ¼ 1; 2; 6Þ ay ¼ v v ¼ v0  v0
ta ta >
>
>
>
are determined as functions of the longitudinal direction Ea
i  >
>
a ¼ w w j >
;
of the joint within the overlap zone, since the adherend ta
thicknesses are variable within the overlap.
ði; j ¼ 1; 2; 3Þ;
From the ‘Love–Kirchhoff’; assumptions, the following ð7Þ
kinematic relations for the laminates are derived: ði 6¼ jÞ

ui ¼ ui0 þ zix; ix ¼ w;ix; iy ¼ 0 ð4Þ

where ui is the longitudinal displacement, ui0 is the
longitudinal displacement of the midplane, and wi is the
vertical displacement of the ith laminate.
where i and j are the numbers of the adherends, Ga is the
shear modulus, and Ea is the elastic modulus of the
adhesive layer.
The consequence of using the simple spring model
approach for the modelling of the adhesive layers is,
1016 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

1 2 9
that it is not possible to satisfy the equilibrium conditions Nxx;x ¼ ax; Nxx;x ¼ ax >
>
>
>
at the edges (free) of the adhesive, where the shear >
>
1
Nxy;x ¼ ay ; 2
Nxy;x ¼ ay >
>
stresses must equal zero, see Fig. 3. However, in real >
>
>
>
adhesive joints no free edges are present at the ends of 1 2 =
Qx;x ¼ a ; Qx;x ¼ a
the overlap, since a fillet of surplus adhesive, a so-called >
t1 ðxÞ þ ta t2 ðxÞ þ ta >
>
spew-fillet, is formed at the ends of the overlap zone. 1 >
>
Mxx;x ¼ Qx1  ax ; 2
Mxx;x ¼ Qx2  ax >
>
This spew fillet allows for the transfer of shear stresses 2 2 >
>
>
>
at the overlap ends. A detailed discussion of the validity 1 t1 ðxÞ þ ta t2 ðxÞ þ ta >
;
Mxy;x ¼ Qy1  ay ; 2
Mxy;x 2
¼ Qy  ay
of the adhesive layer spring model approach is presented 2 2
later in this paper. 0 4 x 4 L: ð9Þ

2.5. Equilibrium equations where t1(x) and t2(x) are the adherend thicknesses and ta
is the adhesive layer thickness. For the single lap joint
The equilibrium equations are derived based on equi- and for the bonded doubler the adherend thicknesses
librium elements in- and outside the overlap zone for will remain the same in the entire overlap zone. For the
each of the considered joint types. The equilibrium single sided stepped lap joint, the adherend thicknesses
equations are derived for plates in cylindrical bending, will change inside the overlap zone between each step.
since the equilibrium equations for the beam modelling
can be considered as a reduced case of this. The general 2.6. The complete set of system equations
equilibrium equations outside the overlap zone for each
of the adherends, i.e. in the regions  L1 4 x 4 0 and From the equations derived, it is possible to form the
L 4 x 4 L þ L2 , are all the same (see Fig. 3) and are complete set of system equations for each of the bonded
derived based on Fig. 5: joint configurations. Thus, combination of the con-
stitutive and kinematic relations, i.e. Eqs. (3) and (4),
i 9 together with the constitutive relations for the adhesive
Nxx;x ¼0 >
>
> layers, i.e. Eq. (7), and the equilibrium equations lead to
i
Nxy;x ¼0 >
>
>
= a set of 8 linear coupled first-order ordinary differential
i  L1 4 x 4 0 and L 4 x 4 L þ L2 : ð8Þ
Qx;x ¼0 equations describing the system behaviour of each of
>
>
i
Mxx;x ¼ Qxi >
> the adherends. The total set of coupled first-order
>
>
; ordinary differential equations within the overlap zone
Mxy;x ¼ Qyi
i
is therefore 16 for joints with 2 adherends inside the
overlap zone, and 24 for the joints with 3 adherends
where i corresponds to the adherends i=1, 2, 3. inside the overlap zone. The actual derivations of the
In cylindrical bending the stress and moment resul- governing equations for the bonded joint types included
tants are only a function of the longitudinal coordinate in this study are quite lengthy and tedious (although
x, and the derivatives with respect to the width direction straightforward). Therefore the derivations are not
y are all equal to zero. included in this paper, but the resulting sets of govern-
The equilibrium equations derived inside the overlap ing equations for the single lap joint are shown in
zones can be divided into the following two groups: Appendix A. The governing equations can be expressed
in the following general form within each region, ie

Joints with one adhesive layer inside the overlap insite and outside the overlap zone:
zone.

Joints with two adhesive layers inside the overlap 2 r
  3
A11 ðxr Þ ... r
A1n ðxr Þ
zone. 6 7
6 7

6 7
These two groups are further divided into joints with y ðx Þ ; x ¼ 6
n i r
6
7
7
6 7
straight or scarfed adherends within the overlap. How- 4 5
   r r ð10Þ
ever, in the following only the equilibrium equations for r
An1 ðxr Þ . . . Ann ðx Þ
joints with two straight adherends within the over lap

will be shown, i.e. single lap joint (see Fig. 3); bonded n yi ðxr Þ þ  n Bir ðxr Þ

doubler and single sided stepped lap joint. For a full n yi ðxr Þ ¼ y1 ðxr Þ ; y2 ðxr Þ ; . . . ; yn ðxr Þ

description of the derivation of the equilibrium equa-
tions for the rest of the joint types, see Mortensen [25].
The equilibrium equations for joints with two straight
adherends within the over lap are derived based on
Fig. 6:
 n Bir ðxr Þ ¼ B1r ðxr Þ ; B2r ðxr Þ ; . . . ; Bnr ðxr Þ
where n is the number of adherends within the region
considered. The values of n are between 1 and 3 (both
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1017

Fig. 5. Equilibrium elements of adherend outside the overlap zone;  L1 4 x 4 0 or L 4 x 4 L þ L2 .

Fig. 6. Equilibrium element of adherends inside the overlap zone for joints with one adhesive layer and straight adherends; 0 4 x 4 L.
1018 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

included) depending
  on the type of joint and the region The boundary conditions for adherend 1 at x=L and
considered. Aijr ðxr Þ ði; j ¼ 1; . . . nÞ is a ðm; mÞ sub-coeffi- for adherend 2 at x=0 are derived from the assumption
cient matrix

for the system of governing equations, and that the adherend edge is free, see Fig. 3.
Bir ðxr Þ is a (m,1) sub-matrix of non-homogeneous
load terms, where m is the number of equations for each 2.8. Multi-segment method of integration
adherend, i.e. 8 for the cylindrical bending case and 6
for the beam
case. For the cylindrical bending case the The governing equations, together with the boundary
vector yi is the vector containing the fundamental conditions constitutes a multiple-point boundary value
variables, which are those quantities that appear in the problem to which no general closed-form solution is
natural boundary conditions at an edge x=constant obtainable. The multiple-point boundary value problem
defined by: is therefore solved using the ‘multi-segment method of
i
n i i i i i i i
o integration’. The method is based on a transformation
y ¼ u0 ; w ; x ; v0 ; Nxx; Nxy ; Mxx ; Qxi
of the original ‘multiple-point’ boundary value problem
ð11Þ into a series of initial value problems. The principle
i ¼ 1; 2; 3:
behind the method is to divide the original problem into
a finite number of segments, where the solution within
These variables will be determined through the ana- each segment can be accomplished by means of direct
lysis. In addition, the quantities integration. Fulfilment of the boundary conditions, as
well as fulfilment of continuity requirements across the
i
n i i i
o
segment junctions, is assured by formulating and sol-
yres ¼ Nyy ; Myy ; Mxy ; Qyi i ¼ 1; 2; 3 ð12Þ
ving a set of linear algebraic equations. For a detailed
description of the method, see [29] and Mortensen [25].
can be determined from the equilibrium equations and
the constitutive relations. These quantities can be con-
sidered as the stress and moment resultants necessary to 3. Examples and discussion
keep the structure in a state of cylindrical bending.
For the beam case the problem is reduced to a set of 6 To show the applicability of the developed linear
coupled first-order ordinary differential equations for solution procedures an example of a single lap joint is
each adherend, and the solution vector containing the presented.
fundamental variables for each adherend for this pro- The adherends used in the examples are chosen to be
blem is defined by: laminates, and the lay-ups are made such that asym-
i
i i i
metric laminates, and thereby coupling effects appear in
i i
y ¼ u0 ; w ; x ; Nxx ; Mxx ; Qxi i ¼ 1; 2; 3: ð13Þ the example presented. This is chosen to show the cap-
abilities of the approach, i.e. the modelling of the
In this case, no additional quantities need to be adherends as plates in cylindrical bending including
determined. coupling effects.
The basic adherend and adhesive properties assumed
2.7. The boundary conditions in the examples are shown in Table 1.
The adhesive used, AY103 from Ciba-Geigy, is a two-
To solve the adhesive bonded joint problems (see component plasticized epoxy, which is considered to be
Fig. 1) the boundary conditions and continuity condi- a general structural adhesive forming strong semi flex-
tions have to be stated. In the following the boundary ible bonds.
conditions and continuity conditions are stated for a The lay-up of the adherends, the dimensions as well as
single lap joint, see Fig. 3: the boundary conditions used in the example, are shown
) in Table 2.
x ¼ L1 ; L þ L2 : prescribed : ui0 or Nxx i
; wi or Qxi ;
ix or Mxx
i
; vi0 or Nxy
i

i ¼ 1; 2
Table 1
Specification of ply and adhesive material properties used for the
x ¼ 0 : adherend 1 : continuity across junction single lap joint example
2 2 2
adherend 2 : Nxx ¼ Nxy ¼ Mxx ¼ Qx2 ¼ 0 Plies Graphite/epoxy E1=164.0 GPa, E2=E3=8.3 GPa,
G12=G31=G23=2.1 GPa, v12=v13=v23=0.34,
1 1 1
x ¼ L : adherend 1 : Nxx ¼ Nxy ¼ Mxx ¼ Qx1 ¼ 0 t=0.125 mm

adherend 2 : Continuity across junction ð14Þ Adhesive Epoxy AY103 (Ciba Geigy), Ea=2800 MPa, va=0.4
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1019

Table 2 The vertical and the width direction displacements of

Laminate lay-ups, thicknesses, lengths and boundary conditions used
for the single lap joint example (see Fig. 3)
Laminate 1 Graphite/epoxy [0 , 30 , 60 ]4, t1=1.5 mm
Laminate 2 Graphite/epoxy [60 , 30 , 0 ]4, t2=1.5 mm
Lengths L1=L2=30.0 mm, L=20.0 mm
Adhesive ta=0.05 mm
Modelling Wide plates in cylindrical bending
Load x=L1: u10 ¼ w1 ¼ v10 ¼ Mxx 1
¼ 0;
and B.C.’’s x=L+L2: w2 ¼ v20 ¼ Mxx 2
¼ 0; Nxx 2
¼ 100 N=mm
The overlap length is L=20 mm, see Fig. 3 for refer-
ence, and the adherends are modelled as plates in
cylindrical bending.
The lay-up of the adherend laminates is made such
that the joint is symmetric seen from the adhesive layer
midplane, and such that the plies facing the adhesive
layers are 0 plies.
The joint is assumed to be simply supported at both
ends, and, additionally, prevented from moving in the
width direction, i.e. vi0 ¼ 0 for (i=1,2) at x=-L1 and
x=L+L2.
In the examples only in-plane normal loading has
been applied, and the adhesive layer stresses are there-
fore normalized with respect to the prescribed load on
the loaded edge, which is given by:
Ni ¼ Ni =ti ; ði ¼ 1; 2Þ
where Ni is in-plane normal stress resultant applied.
In the example the results for a selected number of the
fundamental variables for the single lap joint problem
are presented to show the output from the analysis,
where the adherends are asymmetric and unbalanced
laminates modelled as plates in cylindrical bending.
The single lap joint configuration assumed in the
example is shown in Fig. 7.
Fig. 7. Single lap joint simply supported at both ends (clamped in the
width direction), t1=t2=1.5 mm, ta=0.05 mm, L1=L2=30 mm,
2
L=20 mm, Nxx =100 N/mm $ N ¼ 67 MPa.
the adherends are shown in Figs. 8 and 9, respectively.
From Fig. 8 it is seen that relatively large deflections
occur due to the eccentricity in the load path, and that
the joint rotates around the center of the overlap zone,
i.e. at x=10 mm. As described previously the joint
configuration is symmetric seen from the adhesive layer
midplane, and this is the reason for the perfect skew-
symmetric deflection pattern displayed.
The width direction displacements v10 and v20 , shown in
Fig. 9, occur because of the coupling effects in the
laminates. Comparison of the lateral deflections in Fig. 8
and the width direction displacements in Fig. 9 shows
that the displacements in the width direction are much
smaller. The in-plane normal stress resultants in the
longitudinal direction (x-direction) for the adherends
are shown in Fig. 10.
From Fig. 10 it is observed that the in-plane normal
stress resultants are constant outside the overlap zone,
and that the load is transferred between the adherends
through the adhesive layer in the overlap zone.
The in-plane shear stress resultants for the adherends
are shown in Fig. 11.
Fig. 11 shows a similar pattern of the in-plane shear
stress resultants as for the in-plane normal stress resul-
tants shown in Fig. 10. The shear stress resultants occur
since the adherends are asymmetric and unbalanced
laminates, and since displacements and rotations are
ð15Þ prevented in the width direction at the boundaries at
both ends (see Fig. 4). The peak values of the shear
stress resultants are about 24% of the in-plane normal
stress resultants applied to the bonded joint.
The bending moment resultants in the adherends in
the longitudinal direction are shown in Fig. 12.
From Fig. 12 it is observed that the maximum bend-
ing moment resultants occur at the ends of the overlap.
The large bending moment resultants occur due to the
eccentricity of the load path.
Fig. 13 shows the out-of-plane shear stress resultants.
It is observed from Fig. 13 that highly localized shear
stress resultants are present at the ends of the overlap
zone, and this demonstrates the presence of strong local
bending effects. The peak values of the out-of-plane
shear stress resultants are about 17% of the in-plane
normal stress resultants applied to the structure.
The results shown in Figs. 8–13 are the fundamental
variables determined through the analysis according to
Eq. (11).
In Fig. 14 the distributions of the normalized adhesive
layer stresses are shown.
From Fig. 14 it is seen that the peek stresses are
located at the ends of the overlap. As a consequence of
the coupling effects in the laminates, it is seen that width
direction shear stresses (ay ) are non-zero (although
of minor magnitude). Considering Fig. 8 it is difficult
to distinguish any noticeable differences between the
1020 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

Fig. 8. Vertical displacements w1, w2 of the adherends.

Fig. 9. Width direction displacements v10 , v20 of the adherends.

deflections of the adherends, but from the adhesive layer
stress distribution displayed in Fig. 14, it is seen that
significant transverse normal stresses do occur. These
stresses are induced from differences in the lateral
deflections of the adherends.
If the adherends of a bonded joint fulfil the assump-
tions for cylindrical bending, i.e. if the width of the
bonded joint is considerable or if the movement of the
sides of the bonded joint is subjected to restraints, the
analysis should be carried out following this approach.
The ‘width direction’; variables under such conditions
can be of considerable magnitude.
4. Non-linear adhesive formulation
The structural modelling described is based on the
assumption that the adhesive layer behaves as a linear
elastic material. This is a good approximation for most
brittle adhesives, especially at low load levels, and the
approach is useful to predict the stress distribution and
the location of the peak stress values. However, most
polymeric structural adhesives exhibit inelastic beha-
vior, in the sense that plastic residual strains are induced
even at low levels of external loading.
4.1. Non-linear formulation and solution procedure
The concept of effective stress/strain is one way of
approaching this problem, and it assumes, for a ductile
material, that the plastic residual strains are large com-
pared with the creep strains at normal loading rates.
Therefore, a plastic yield hypothesis can be applied, and
the multidirectional state of stress can be related to a
simple unidirectional stress state through a function
similar to that of von Mises.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1021

1 2
Fig. 10. In-plane normal stress resultants Nxx , Nxx in the adherends in the longitudinal direction.

1 2
Fig. 11. In-plane shear stress resultants Nxy , Nxy in the adherends.

1 2
Fig. 12. Bending moment stress resultants Mxx , Mxx in the adherends.
1022 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

Fig. 13. Out-of-plane shear stress resultants Qx1 , Qx2 in the adherends.

Fig. 14. Normalized adhesive layer stresses, ax =N ; ay =N ; a =N .

However, it is widely accepted that the yield beha-
viour of polymeric structural adhesives is dependent on
both deviatoric and hydrostatic stress components. A
consequence of this phenomenon is a difference between
the yield stresses in uniaxial tension and compression,
see Adams et al. [20], Gali [30], Adams [2] and Thomsen
[17,18].
This behaviour has been incorporated into the analy-
sis by the application of a modified von Mises criterion
suggested by Gali et al. [30]:
where s is the effective stress, J2D is the second invariant
of the deviatoric stress tensor, J1 is the first invariant of
the general stress tensor and l is the ratio between the
compressive and tensile yield stresses. For l ¼ 1, Eq.
(16) is reduced to the ordinary von Mises criterion. At
the failure load level, the first of Eq. (16) is transformed
into the expression:
sult ¼ CS;ult ðJ2D Þ1=2
ult þCV;ult ðJ1 Þult ð17Þ

pffiffiffi where the subscript ‘ult’; denotes ‘ultimate’’. Eq. (17)

1=2 3ð 1 þ lÞ
s ¼ CS ðJ2D Þ þCV J1 ; CS ¼ ; describes the failure envelope for the general case of a
2l ductile material, and in three-dimensional stress space
ð16Þ
l1 c Eq. (17) represents a paraboloid with its axis coincident
CV ¼ ; l¼
2l t with the line 1 ¼ 2 ¼ 3 .
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031
The effective strain e is given by Gali [30]:
1 1
e ¼ CS ðI2D Þ1=2 þCV ð I1 Þ ð18Þ
1þv 1  2v
where v is Poisson’s ratio, I2D is the second invariant of
the deviatoric strain tensor and I1 is the first invariant of
the general strain tensor.
The non-linear adhesive properties are included by
implementing an effective stress–strain relationship
derived experimentally from tests on adhesive bulk spe-
cimens, see Thomsen [17,18] and Tong [8]. Thus it is
assumed that the bulk and ‘in situ’; mechanical proper-
ties of the structural adhesive are closely correlated as
discussed by Gali et al. [30] and shown experimentally
by Lilleheden [32].
Based on a secant modulus approach for the non-lin-
ear effective stress–strain relationship for the adhesive,
the solution procedure for determining the stress dis-
tribution in the adhesive layer can be described by the
following steps:
1023
been used for the analysis of non-linear adhesive beha-
viour in tubular lap joints by Thomsen [17,18].
4.2. Examples and discussion
The effects of non-linear adhesive behaviour is illu-
strated by the following example. The basic differences
between the linear and non-linear analysis is that the
adhesive stresses, at higher levels of loading, are reduced
and smoothed out in the regions adjacent to the ends of
the overlap zone.
To illustrate the effects of non-linear adhesive beha-
viour the single lap joint example shown in Section 3
will be used. Therefore, the results obtained can be
compared with the results shown in Section 3. The
dimensions, laminates, loads etc. are all shown in
Table 2 in Section 3.
The adhesive used in the example is AY103 from Ciba-
Geigy, which is a two-component plasticized epoxy
adhesive as described in Section 3. The tensile stress–
strain curve for this adhesive has been obtained from bulk
specimens (see Adams [20,31] and Harris and Adams [2]),
which have been subjected to the cure cycle usually used
1. Calculate the effective strains e1 and stresses s1
[Eqs. (16) and 18)] for each point of the adhesive
layer, using the linear elastic solution procedure,
assuming a uniform elastic modulus E1 for the
adhesive.
for joints bonded with this adhesive. The stress–strain
relation for the used adhesive material is shown in
Fig. 15 and the material properties are given in Table 3.
Analysis of the single lap joint example shown in
Section 3 by use of the non-linear solution procedure,
2. If the calculated effective stresses s1 are above the
proportional limit denoted by sprop, determine
the effective stresses s1 for each point of the
adhesive layer according to the corresponding
effective strains e1 calculated in step (1).
based on the adhesive properties given in Table 1 and
the non-linear adhesive stress–strain relation shown in
Fig. 15, results in the adhesive layer stress distribution
shown in Fig. 16.
Comparison of the adhesive layer stress distribution
3. Calculate the difference %s1 ¼ s1  s1 between
the ‘calculated’; and the ‘experimental’; effective
stresses, and determine the specific secant-mod-
ulus E2t defined by:
non-linear effects are only influential very close to the
E2t ¼ 1  ð%s1 =s1 Þ E1
 is a weight-factor, which determines the change
of the modulus in each iteration.
4. Rerun the procedure [steps (1) and (2)] with the
elastic modulus E1 for each adhesive point mod-
ified as per step (3).
obtained for the linear case, shown in Fig. 14, and for
the non-linear case, shown in Fig. 16, shows that inclu-
sion of the non-linear effects reduce the maximum pre-
dicted stresses with about 25%. It is also seen that the
ends of the overlap zone. The differences between the
linear and non-linear solutions are strongly dependent
on the load level. Failure has been predicted to occur at
a load level corresponding to N ¼ 80 MPa by use of
the ultimate stress criterion.
Fig. 17 displays the differences between the linear and
non-linear solutions, by showing the maximum adhesive
5. Compare the ‘calculated’; effective stresses s for
each adhesive point with the ‘experimental’;
values s obtained from the effective stress–strain
curve.
6. Repeat steps (4) and (5) until the difference
between the ‘calculated’; and ‘experimental’;
stresses (%s) drops below a specified fraction
(2%) of the ‘experimental’ stress value.
Convergence is usually achieved within a few itera-
tions. The procedure described above has previously
layer stresses as a function of the load level obtained
using the same single lap joint configuration.
It is observed from Fig. 17, that the adhesive non-lin-
earity starts to affect the adhesive layer stresses at very
low load levels, and the non-linear effects become
increasingly important as the external loading is
increased. Thus, prediction of the adhesive stresses
using the linear solution procedure will underestimate
the load bearing capability of the adhesive joint, except
for very brittle adhesives where the non-linear effects
are of minor importance. This observation is in close
1024 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

Fig. 15. Tensile stress/strain curve for the adhesive AY103 obtained from bulk specimen, load controlled: 20.0 [MPa/min].

Table 3 solution, shown in Section 4.1, it can be concluded that

Specification adhesive material properties
Adhesive Epoxy AY103

(Ciba Geigy), Ea=2800 MPa, va=0.4,
l ¼ 1:3 l ¼ ct sprop ¼ 27:0 MPa; sult ¼ 71:5 MPa,
eult ¼ 0:049 MPa
agreement with experimental results, see Hart-Smith
[5–7], Adams [20,31] and Harris and Adams [2].
4.3. Summary
From the comparison between the linear elastic solu-
tion, shown in Section 3, and the non-linear adhesive
the non-linear behaviour shown by many polymeric
adhesives exert a strong influence on the adhesive layer
stress distribution. The severe stress concentrations,
predicted by the linear solution procedure, tend to
smooth out when a non-linear solution procedure is
applied. The non-linear effects become influential even
at low load levels, and become very influential at higher
load levels. Thus, in most cases non-linear effects are
unavoidable, and a certain degree of plasticity in the
adhesive layer close to the ends of the overlap cannot be
prevented. Linear elastic solution procedures will there-
fore underestimate the strength of adhesive bonded
joints unless very brittle adhesives with approximately
linear elastic properties are considered.

Fig. 16. Normalized adhesive layer stresses, ax =N ; ay =N ; a =N for a single lap joint obtained by use of the non-linear solution procedure.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1025

5. Validation of the adhesive layer model
To investigate the validity of the adhesive layer model
a high-order theory approach developed for the analysis
of sandwich structures has been used, see Frostig et al.
[33] and Frostig [34], together with FE-analyses. The
investigation is partially presented in the paper by
Frostig et al. [19]. In this paper an extended validation
of the adhesive layer model is presented.
As a consequence of modelling the adhesive layer as
continuously distributed springs (tension/compression
and shear springs) it is not possible to fulfil the condi-
tion of zero shear stresses (ax ¼ 0) at the free edges of
the adhesive layers (see Fig. 3) as assumed in the
boundary conditions for all the joints. However, in real
adhesive bonded joints no free edges at the ends of the
overlap zones are present, since a fillet of surplus adhe-
sive, a so-called spew fillet, is formed at the ends of the
overlap. The differences between the model joint and a
real joint is illustrated in Fig. 18.
Consequently, the shear stresses (ax ) in a real joint
will not be zero at the ends of the overlap zone. It can
therefore be stipulated that using the spring model
approach in some sense is equivalent to assuming the
existence of a spew-fillet at the ends of the overlap zone.
The actual spew-fillet size, however, is not addressed by
adopting the spring model approach.
The validity of the adhesive layer spring model
approach has been investigated by comparison with:

A high-order theory approach including the pre-
sence of a spew-fillet;

FE-analyses including a spew-fillet;
In the following, the method described in this paper is
called the ‘spring model approach’ for simplicity.
5.1. Structural modelling using a high-order theory
approach
The high-order theory approach used has been devel-
oped by Frostig et al. [33] and Frostig [34] for the study
of localized effects in sandwich beams and plates. The
theory includes the transverse flexibility of the core
material. Thus, the core thickness is allowed to change
during the deformation of the sandwich panel, and the

Fig. 17. Maximum adhesive layer stresses, tax ; ay ; a for a single lap joint using the linear and the non-linear solution procedure as a function of the
applied in-plane nominal stresses.

Fig. 18. Illustration of a model joint and a real joint.

1026 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031
face sheets are allowed to deflect differently. In the pre-
sent study the high-order theory approach has been used
for the analysis of a single lap joint, since the bonded joint
in the overlap zone are comparable with a sandwich plate,
i.e. the adhesive layer can be seen as the core material and
the adherends can be seen at the face sheets.
The structural modelling using the high-order theory
approach can be described in the following way (for
more details see Frostig et al. [19]:

The adherends:
- Modelled as beams or plates in cylindrical
bending.

The adhesive:
- Assumed to be a 2-D or 3-D (in case of
cylindrical bending) elastic continuum only
possessing stiffness in the thickness direction,
and therefore only capable of transferring
transverse normal and shear stresses. The
constitutive relations for the adhesive layer
can be written as:
xz ðx; za Þ ¼ Ga xz
ð20Þ
zz ðx; za Þ ¼ Ea "zz
where za is the local z-coordinate in the
adhesive layer. za=0 is at the upper interface
of the adhesive layer, and za=ta is at the
lower interface of the adhesive layer.
The stress field in the adhesive layer is
derived using point equilibrium conditions
together with the constitutive relations in Eq.
(20), and are given by (for more detail see
Frostig et al. [19] and Frostig [32]):
xz ðx; za Þ ¼ xz ðxÞ ¼ ax
ax;x ð2za  ta Þ Ea ðw2  w1 Þ
zz ðx; za Þ ¼  þ
2 ta
ð21Þ
Compared with the stress field for the
adhesive layer obtained using the spring
model approach [Eq. (7)], it is observed, that
the relation for the shear stresses are the same.
In the high-order theory approach, however,
the transverse normal stresses are not constant
across the thickness of the adhesive layer, but
are predicted to vary linearly across the adhe-
sive layer thickness, and are a function of the
derivative of the shear stresses with respect to x.
Thus, comparing the equation for the trans-
verse normal adhesive layer stresses from the
spring model approach, i.e. Eq. (7), with the
same for the high-order theory approach, i.e.
Eq. (21), it is seen that the term:
a;x ð2za  ta Þ
ð22Þ
2
has been added.
Since the adhesive layer is modelled as a
continuum it is possible to prescribe the value
of the shear stresses at the ends of the over-
lap.

The spew-fillet:
 First approach: ax;x ¼ 0 in the adhesive layer
at the ends of the overlap. Usually this
requirement leads to prediction of non-zero
shear stresses at the adhesive edges, thus
resembling the presences of a spew-fillet. The
size of the spew-fillet is not addressed by this
approach, however. By this approach the
predicted stress field for the adhesive layer is
nearly exactly the same as predicted using the
spring model approach.
 Second approach: the spew-fillet is modelled
as an equivalent elastic bar extending between
the two adherends. Thus, the capability of the
spew-fillet to transfer loads directly between
the two adherends is included in the model.
Using the high-order theory approach for the analysis
of adhesive bonded joints it is possible to prescribe the
shear stresses at the ends of the overlap ax to be equal
to zero, thus assuming the boundaries of the adhesive
layer to be free. Doing this the peeling stresses will be
unrealistically large at the ends of the overlap zone, see
Mortensen [25]. Instead the two approaches described
above for the modelling of the spew-fillet can be used.
In the second approach, the spew-fillet is modelled as
an equivalent elastic bar extending between the two
adherends. This concept is based on a practical assump-
tion, indicated by finite element results, see Adams et al.
[20,35,36], Harris and Adams [2], that the stress state in
the spew-fillet consists mainly of unidirectional stresses
that are parallel to the free edge of the spew-fillet, see
Fig. 18. However, using this approach it is necessary to
specify the spew-fillet size, which can be very difficult to
determine in practice, and is usually unknown in the
design phase. Comparison of the two approaches
shows, however, that the first approach provides rea-
sonable results for small spew-fillet sizes, and that the
second approach provides more meaningful results for
larger spew-fillets.
5.2. FE-models used for validation
The aim of this comparison is to investigate the
validity of the spring model approach, and to establish
to which extent the approach provides reasonably
accurate results in predicting the adhesive layer stresses.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031
Therefore, two finite element analyses have been carried
out with different spew-fillet sizes. In the first analysis
the size of the spew-fillet is hsp=2ta and in the second
analysis the size is hsp=1.2ta, where the size of the spew-
fillet is defined according to Fig. 18.
The finite element analysis has been performed
using the finite element code ODESSY developed at
Aalborg University, see Rasmussen et al. [37] and
Rasmussen and Lund [38]. A zoom of the finite element
models at the left end of a single lap joint including a
spew-fillet with the size hsp=1.2ta hsp=2ta is shown in
Fig. 19.
The structure has been modelled using a mixture of 6
and 8 node isoparametric 2D solid elements. The adhe-
sive layer has been divided into six 8 node isoparametric
elements through the thickness. In the near vicinity of
the overlap edges the adhesive layer elements are quad-
ratic, whereas irregular elements are used in the middle
of the overlap regions where low stress gradients are
present. The finite element model for the single lap joint,
including a spew-fillet with the size hsp=2ta, contains
8525 elements with 26208 nodes. The analysis has been
carried out as a plane strain model, since the proposed
simple approach is based on the assumption that the
laminates behave as plates or wide beams. It should be
noted that the prediction of the stresses in the adhesive
layer is affected by the singularities present at the cor-
ners of the adherends facing the adhesive layer at the
ends of the overlap zone.
To support the claim that the presence of a spew-fillet
plays a very important structural role as a ‘stress relie-
ver’ at the ends of the overlap zones, a finite element
analysis of the considered single lap joint configuration
without a spew-fillet has also been performed.
The accuracy of the finite element results obtained
near the ends of the overlap zone is of course again
affected by the singularity present at the corners of the
adherends facing the adhesive layer at the ends of the
overlap zone.
Fig. 19. Zoom of finite element models at the left end of a single lap joint including a spew-fillet with the sizes hsp=1.2ta and hsp=2ta.
1027
5.3. Comparison of results
The comparison has been carried out analysing a sin-
gle lap joint composed of two identical aluminium
adherends with the properties shown in Table 4.
Analysing the single lap joint problem described in
Table 4 and using the three different methods: (1) spring
model approach, (2) high-order theory approach and (3)
the finite element method give the adhesive layer stress
distributions shown in Figs. 20 and 21.
From Figs. 20 and 21 it is seen that the stresses cal-
culated by the three methods compare very well and
display only small differences. The results calculated by
the high-order theory approach, and displayed in the
figures, are determined by the assumption that ax;x ¼ 0
at the ends of the overlap zone, i.e. the actual size of the
spew-fillet is not addressed. In Fig. 20 the transverse
normal stresses in the adhesive layer are displayed in the
top and bottom interfaces of the adhesive layer and are
marked by HOTAT and HOTAB. Only small differ-
encies between the two curves HOTAT and HOTAB are
observed, however. The stresses calculated by the finite
element method and displayed in the figures are deter-
mined from the single lap joint configuration with the
spew-fillet size hsp=2ta. It is seen that the stresses
determined by the finite element method are a bit lower
than the stresses determined by the two other approa-
ches, but if the stresses are determined using the config-
uration with the spew-fillet size hsp=1.2ta, the stresses
will be a bit higher than by the two other approaches.
Thus, it is concluded that the adhesive layer stresses
predicted by the spring model approach essentially
equal the stresses predicted by the high-order theory
approach including a spew-fillet (ax;x ¼ 0 condition at
x=0,L), which again equal the stresses predicted by the
finite element method including a spew-fillet with a size
hsp between 1 and 2 times the adhesive layer thickness ta.
In the manufacturing process of real adhesive bonded
joints, it is unavoidable to have a spew-fillet of at least
1028 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

1–2 times the adhesive layer thicknesses. However, it is usually not known, and, moreover, can be difficult to
possible to have larger spew-fillet sizes, and in this case, determine, it is concluded that the stresses predicted by the
the stresses predicted by the spring model approach will spring model approach represent reasonable and con-
overestimate the stresses. Since the real spew-fillet size is servative estimates of the ‘real’ adhesive layer stresses.
Finally, to emphasize the importance of the presence
Table 4 of a spew-fillet at the ends of the overlap zone, the
Adherend and adhesive properties, thicknesses, lengths and boundary adhesive layer stresses obtained for the single lap joint
conditions assumed for the single lap joint to investigate the validity of configuration without a spew-fillet are displayed in Figs. 22
the spring model approach
and 23. The stress distributions displayed have been deter-
Adherend 1 and 2 Aluminium E1=E2=70 GPa, t1=1.62 mm, mined by use of the spring model approach, the high-order
b1=b2=25.4 mm theory approach and the finite element method.
Adhesive Epoxy Ea=4.82 GPa, v=0.4, ta=0.25 mm From Fig. 23 it is seen that the adhesive layer shear
Lengths L1=L2=50.8 mm, L=12.7 mm stresses display the same pattern and almost same peak
values as obtained with the presence of spew-fillets. The
Load and B.C. Simply supported at both ends,
PN=1.0 kN(N ¼ PN =b)
only exception being that the shear stresses are zero at the
adhesive edges when there is no presence of a spew-fillet.

Fig. 20. Normalized adhesive layer transverse normal stresses, a =N using the spring model approach (SMA), the high-order theory approach
(HOTA, ax;x ¼ 0 at x=0, L) and the finite element method (FEM) for a single lap joint with a spew-fillet of size hsp=2ta.

Fig. 21. Normalized adhesive layer shear stresses, ax =N using the spring model approach (SMA), the high-order theory approach (HOTA, ax;x ¼
0 at x=0,L) and the finite element method (FEM) for single lap joint with a spew-fillet of size hsp=2ta.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031 1029

Fig. 22. Normalized adhesive layer transverse normal stresses, a =N using the spring model approach (SMA), the high-order theory approach
(HOTA, ax ¼ 0 at x ¼ 0; L) and the finite element method (FEM) for a single lap joint without a spew-fillet.

Fig. 23. Normalized adhesive layer shear stresses, a =N using the spring model approach (SMA), the high-order theory approach (HOTA, ax ¼ 0
at x=0, L) and the finite element method (FEM) for single lap joint without a spew-fillet.

Fig. 22 shows, however, that the transverse normal 6. Conclusions

stresses reach extreme values at the adhesive layer free
edges. Furthermore it is seen that the transverse normal
stresses at the upper (HOTAT) and lower (HOTAB)
adhesive interface have different signs, i.e. tremendous
gradients across the adhesive layer thickness are
induced. Thus, from this and by comparison with the
stresses determined for a single lap joint with a spew-
fillet it can be concluded that adhesive bonded joints
without spew-fillets will be structurally very weak and
useless for any practical applications. This conclusion is
well known from engineering pratice.
A general method for the analysis of adhesive bonded
joints between composite laminates has been presented.
The analysis accounts for coupling effects induced by
adherends made as asymmetric and unbalanced lami-
nates. The analysis allows specification of any combi-
nation of boundary conditions and external loading.
The analysis can be carried out with the adherends
modelled as wide beams or as plates in cylindrical
bending. The adhesive layers are as a first approxima-
tion assumed to behave as a linear elastic material. The
1030 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031

thickness of the adhesive layer is assumed to be small i i i i

i B16 B11 i B16 B16
compared with the thickness of the adherends, and the cli ¼ A16  i ; c 2i ¼ A66  i ;
D11 D11
adhesive layers are modelled as continuously distributed i
B16 Bi B i cli i
B16 c3i 1
linear tension/compression and shear springs. The c3i ¼ i
; hli ¼ 11
i
 16
i c
; h2i ¼  i c
 i ;
results obtained using this approach have been com- D11 D11 D11 2i D11 2i D11
i
pared with finite element results and results obtained B16 i i A i cli
h3i ¼ i ; mli ¼ A11  B11 hli  16 ;
using a high-order theory approach (both including D11 c2i c2i
spew-fillets), and the comparison shows that the results i
A16 c3i i
A16
i i
are in very good agreement. m2i ¼   B11 h2i; m3i ¼  B11 h3i ð25Þ
c2i c2i
The linear solution procedure has been used together
with a modified von Mises criterion and a secant mod- Within the overlap zone, i.e. for 0 4 x 4 L (see Fig. 3),
ulus approach for the effective stress–strain relationship combination of Eqs. (3), (4), (7) and (9) yields for lami-
of the adhesive, to perform a non-linear solution for the nates 1 and 2:
adhesive bonded joint problems. Comparison of the
results obtained using the linear and non-linear solution u10;x ¼ k11 Nxx
1 1
þ k21 Nxy 1
þ k31 Mxx
procedures has shown that non-linear adhesive beha- w;1x ¼ 1x
viour influences the adhesive layer stresses even at low
load levels, and that the non-linear adhesive behaviour 1x;x ¼ k41 Nxx
1 1
 k51 Nxy 1
 k61 Mxx
tends to smooth out the severe stress concentrations v10;x ¼ k71 Nxx
1 1
þ k81 Nxy 1
þ k91 Mxx
induced at the ends of the overlap zones. Ga 1 Ga t1 1 Ga 2 Ga t2 2
1
Nxx;x ¼ u þ   u þ 
ta 0 2ta x ta 0 2ta x
1 Ga 1 Ga 2
Appendix A. Governing equations Nxy;x ¼ v  v
ta 0 ta 0
The governing equations shown here are for a single 1 Ga ðt1 þ ta Þ 1 Ga t1 ðt1 þ ta Þ 1
Mxx;x ¼ Qx1 þ u0 þ x
lap joint. From the equations derived, it is possible to 2ta 4ta
form the complete set of system equations for the pro- Ga ðt1 þ ta Þ 2 Ga t2 ðt1 þ ta Þ 2
 u0 þ x
blems. Thus a combination of Eqs. (3), (4) and (8) yields 2ta 4ta
for laminates 1 and 2 in the areas  L1 4 x 4 0 and 1 Ea 1 Ea 2
L 4 x 4 L þ L2 (outside of overlap; see Fig. 3): Qx;x ¼ w  w
ta ta
9 u20;x ¼ k12 Nxx
2
þ k22 Nxy2 2
þ k32 Mxx
ui0;x ¼ kli Nxx
i i
þ k2i Nxy i
þ k3i Mxx >
>
w;ix ¼ ix
>
>
> w;2x ¼ 2x
>
>
ix;x ¼ k4i Nxxi i
 k5i Nxy  k6i Mxxi >
> 2x;x ¼ k42 Nxx
2 2
 k52 Nxy 2
 k62 Mxx
>
>
i i i i =
v0;x ¼ k7i Nxx þ k8i Nxy þ k9i Mxx v20;x ¼ k72 Nxx
2 2
þ k82 Nxy 2
þ k92 Mxx
i
i ¼ 1; 2: ð23Þ
Nxx;x ¼0 >
>
>
> 2 Ga 1 Ga t1 1 Ga 2 Ga t2 2
i
Nxy;x ¼0 >
> Nxx;x ¼ u   þ u  
>
> ta 0 2ta x ta 0 2ta x
Mxx;x ¼ Qxi
i >
>
>
; Ga Ga 2
2
i
Qx;x ¼0 Nxy;x ¼  v10 þ v
ta ta 0
2 Ga ðt2 þ ta Þ 1 Ga t1 ðt2 þ ta Þ 1
Eq. (23) constitutes a set of eight linear coupled first- Mxx;x ¼ Qx2 þ u0 þ x
2ta 4ta
order ordinary differential equations. The coefficients
Ga ðt2 þ ta Þ 2 Ga t2 ðt2 þ ta Þ 2
kli  k9i ði ¼ 1; 2Þ contain laminate stiffness parameters  u0 þ x
and are a result of isolating ui0;x ; vi0;x and w;ixx from 2ta 4ta
Nxxi i
; Nxy and Mxx i
in Eq. (3): 2 Ea Ea
Qx;x ¼  w1 þ w2 ð26Þ
ta ta
1 m3i m2i
kli ¼ ; k2i ¼  ; k3i ¼  Eq. (26) constitutes a set of 16 linear coupled first-
ml i mli mli
order ordinary differential equations.
k4i ¼ hli kli ; k5i ¼ hli k2i þ h3i ; k6i ¼ hli k3i þ h2i
cli 1 cli cli c3i
k7i ¼  kli ; k8i ¼  k2i ; k9i ¼  k3i  References
c2i c2i c2i c2i c2i
ð24Þ [1] Crocombe AD, Adams RD. An effective stress/strain concept in
mechanical characterization of structural adhesive bonding.
where the coefficients cji, hji and mji (j=1,2,3) are: Journal of Adhesion 1981;13(2):141–55.
 F. Mortensen, O.T. Thomsen / Composites Science and Technology 62 (2002) 1011–1031
[2] Harris JA, Adams RD. Strength prediction of bonded single lap
joints by non-linear finite element methods. International Journal
of Adhesion and Adhesives 1984;4:65–78.
[3] Volkersen O. Die Nietkraftverteilung in Zugbeanspruchten Niet-
verbindungen mit Konstanten Laschenquerschnitten. Luftfahrt-
forschung 1938;15:41–7.
[4] Goland M, Reissner E. The stresses in cemented joints. Journal
of Applied Mechanics 1944;1:A17–A27.
[5] Hart-Smith LJ. Adhesive bonded single lap joints. Technical
report NASA CR 112236, Douglas Aircraft Company, USA:
McDonnell Douglas Corporation; 1973.
[6] Hart-Smith LJ. Adhesive bonded double lap joints. Technical
report NASA CR 112237, Douglas Aircraft Company, USA:
McDonnell Douglas Corporation; 1973.
[7] Hart-Smith LJ. Adhesive bonded scarf and stepped-lap joints.
Technical report NASA CR 112235, Douglas Aircraft Company,
USA: McDonnell Douglas Corporation; 1973.
[8] Tong L. Bond strength for adhesive-bonded single-lap joints.
Acta Mechanic 1996;117:101–13 [Springer-Verlag].
[9] Yuceoglu U, Updike DP. Bending and shear deformation effects
in lap joints. Journal of Engineering Mechanics 1981:55–76.
[10] Yuceoglu U, Updike DP. The effect of bending on the stresses
in adhesive joints. Lehigh University, Bethlehem, Pensylvania,
NASA NGR-39–007–011, 1975.
[11] Yuceoglu U, Updike DP. Stress analysis of adhesive bonded
stiffener plates and double joints. Lehigh University, Bethlehem,
Pensylvania, NASA NGR-39–007–011, 1975.
[12] Yuceoglu U, Updike DP. Comparison of continuum and spring
models of adhesive layers in bonded joints. ASME, Advances in
Aerospace Structures and Materials 1981;AD-01:75–83.
[13] Renton WJ, Vinson JR. On the behavior of bonded joints in
composite material structures. Engineering Fracture Mechanics
1975;7:41–60.
[14] Renton WJ, Vinson JR. The efficient design of adhesive bonded
joints. Journal of Adhesion 1975;7:175–93.
[15] Pickett AK. Stress analysis of adhesive bonded lap joints. PhD
thesis. University of Surrey, 1983.
[16] Pickett AK, Hollaway L. The analysis of elasto-plastic adhesive
stress in adhesive bonded lap joints in F.R.P.-structures. Com-
posite Structures 1985;4:135–60.
[17] Thomsen OT. Analysis of adhesive bonded generally orthotropic
circular cylindrical shells. PhD thesis. Institute of Mechanical Engi-
neering. Aalborg University, Denmark, Special Report, 4, 1989.
[18] Thomsen OT. Elasto-static and elasto-plastic stress analysis of
adhesive bonded tubular lap joints. Composite Structures 1992;
21:249–59.
[19] Frostig Y, Thomsen OT, Mortensen F. Analysis of adhesive
bonded joints, square-end and spew-fillet: higher-order theory
approach. Journal of Engineering Mechanics, ASCE Engineering
Mechanics Division 1997;125(11):1298–307.
[20] Adams RD, Coppendale J, Peppiatt NA. Failure analysis of alu-
minium–aluminium bonded joints. Adhesion 1978;2:105–19.
[21] Gali S, Ishai O. Interlaminar stress distribution within an adhe-
sive layer in the nonlinear range. Journal of Adhesion 1978;9:
253–66.
1031
[22] Mortensen F, Thomsen OT. A simple unified approach for the
analysis and design of adhesive bonded composite laminates.
Proc of Eleventh International Conference on Composite Mate-
rials 1997;VI(18):129–39.
[23] Mortensen F, Thomsen OT. Simplified linear and non-linear
analysis of stepped and scarfed lap adhesive joints between com-
posite laminates. Composite Structures 1997;38(1–4):281–94.
[24] Mortensen F, Thomsen OT. Facilities in {ESAComp} for analy-
sis and design of adhesive joints between composite laminates.
Proc. Composite and Sandwich Structures, (NESCOII), ISBN 0
94781794 8. Chameleon Press Ltd., London, UK; 1997. p. 247–
266.
[25] Mortensen F. Development of tools for engineering analysis and
design of high-performance FRP-composite structural elements,
PhD thesis. Institute of Mechanical Engineering, Aalborg
Universitet, Special Report No. 37, ISSN 0905–2305, 1998.
[26] Saarela O, Palantera M, Haberle J, Klein M. ESAComp: a
powerful tool for analysis and design of composite materials.
Proc. of Int. Symp. on Advanced Materials for Lightweight
Structures, ESA-WPP-070, ESTEC. Noordwijk, The Netherlands;
1994. p. 161–9.
[27] Whitney JM. Structural analysis of laminated anisotropic plates.
Lancaster: Technomic Publishing Company, 1987.
[28] Jones RM. Mechanics of composite materials. 2nd ed. Taylor
and Francis; 1999.
[29] Kalnins A. Analysis of shell of revolutions subjected to symme-
trical and non-symmetrical loads. Journal of Applied Mechanics
1964;31:1355–65.
[30] Gali S, Dolev G, Ishai O. An effective stress/strain concept in
mechanical characterization of structural adhesive bonding.
International Journal of Adhesion and Adhesives 1981;1:135–
40.
[31] Adams RD. Stress analysis: a finite element analysis approach.
In: Developments in Adhesives. 2nd ed. London: Applied Science
Publishers; 1981.
[32] Lilleheden L. Properties of adhesive in situ and in bulk. Interna-
tional Journal of Adhesion and Adhesive 1994;14(1):31–7.
[33] Frostig Y, Baruch M, Vilnay O, Sheinman I. Bending of non-
symmetric sandwich beams with transversely flexible core. ASCE
Journal of Engineering Mechanics 1991;117(9):1931–52.
[34] Frostig Y. Behaviour of delaminated sandwich beam with trans-
versely flexible core-high order theory. Composite Structures
1992;20:1–16.
[35] Adams RD, Peppiatt NA. Stress analysis of adhesive-bonded lap
joints. Journal of Strain Analysis 1974;9:185–96.
[36] Adams RD, Wake WC. Structural adhesive joints in engineering.
1st ed. Elsevier Applied Science Publishers, ISBN 0–85334–263–6,
1984.
[37] Rasmussen J, Lund E, Olhoff N. Integration of parametric mod-
eling and structural analysis for optimum design. In: Gilmore, et
al., editors. Proceedings of advances in design automation. The
American Society of Mechanical Engineers. Albequerque, New
Mexico, USA, 1993.
[38] Rasmussen J, Lund E. The issue of generality in design optimi-
zation systems. Engineering Optimization 1997;29:23–37.