Fracture of Polymers, Composites and Adhesives II

B.R.K. Blackman, A. Pavan and J.G. Williams (Eds)

© 2003 Elsevier Ltd. and ESIS. All rights reserved. 279

FRACTURE MECHANICS TESTS TO CHARACTERIZE BONDED

GLASS/EPOXY COMPOSITES:
APPLICATION TO STRENGTH PREDICTION IN STRUCTURAL ASSEMBLIES

PETER DAVIES, JEFFREY SARGENT

Materials & Structures group, IFREMER, 29280 Plouzane, France

BAE Systems, Advanced Technology-Centre Sowerby, Bristol, UK

ABSTRACT
This paper presents results from a study of assemblies composed of glass fibre reinforced
epoxy composites. First, tests performed to produce mixed mode fracture envelopes are
presented. Then results from tests on lap shear and L-stiffener specimens are given. These
enabled failure mechanisms to be examined in more detail using an image analysis technique
to quantify local strain fields. Finally the application of a fracture-mechanics-based analysis
to predict the failure loads of top-hat stiffeners with and without implanted bond-line defects
is described. Correlation between test results and predictions is reasonable, but special
attention is needed to account for size effects and micro-structural variations induced by the
assembly process.

Keywords. Fracture envelope, Stiffener, Top-hat, Debonding, Image analysis.

INTRODUCTION
Adhesive bonding is finding increasing applications in marine and aerospace composite
structures. The weight savings associated with this form of assembly can be considerable, but
designing to optimise strength requires great care. The overall aim of the present work is to
develop reliable methods to predict the failure strength of composite assemblies. There have
traditionally been two approaches available for strength analysis, either stress-based or
fracture mechanics based, though several variants exist according to the adherend and
adhesive properties [1-3].
The stress-based approach involves performing a stress analysis of the structure, either
analytically, for simple assemblies, or numerically by finite elements for more complex
structures, and then comparing stresses or strains to failure criteria for the different
constituents (adherend, adhesive and if possible interface). The difficulties with this approach
in the past have been the analysis of thin bond-line regions and tackling the stress singularities
at joint extremities. With the increasing computer power available today the former is now
less of a problem, but the latter remains a major limitation. It should be noted also that
sources of reliable strength data to apply in failure criteria have always been scarce,
particularly when the adherends are composites. At first sight a fracture mechanics approach
which enables crack tip fields to be treated looks very attractive for this type of problem, and
indeed it has been employed in the description of delaminations in orthotropic materials with
some success (refs). However its application to composite assemblies has been limited so far,
partly due to a severe shortage of fracture data to use in a failure criterion but also because it
280 P. DA VIES AND J. SARGENT

is necessary to assume the presence of a crack. The question is then raised, where should this
crack be introduced and how long should it be ? Some work has been performed to address
this question, notably by Femlund and Spelt [4,5] and encouraging results have been
presented for a range of bonded metal joints. It should be noted that these two approaches are
not the only ones attracting interest at present. Other recent developments have included the
use of damage mechanics models [6] and the development of cohesive zone models [7,81.
A first aim in the present study was to establish whether the approach proposed by
Femlund & Spelt, which will be described in more detail below, could be applied to bonded
composite joints. A second objective was to evaluate the benefits to be gained from a fracture
mechanics characterisation when industrial structural assemblies are considered, both
qualitatively in material selection, and quantitatively when failure predictions are required.
The paper is presented in three parts. First, the tests employed to determine the mixed
mode fracture envelope of a glass fibre reinforced epoxy composite adhesively bonded with
either a brittle or a ductile adhesive are briefly described. These include mode I (DCB), and
mixed mode (MMB) with various mixed mode (I/II) ratios. In the second part of the paper
different structural joints will be discussed. These include single and double lap shear and L-
specimens. In a recent European thematic network lap shear and double lap shear composite
joints were tested, and predictions of failure load were made by different academic and
industrial partners [9,10]. It was apparent that considerable differences existed between
different analytical predictions and FE analyses, and correlation with tests proved complex. In
particular, the progressive damage development in assemblies bonded with a ductile adhesive
was not treated adequately. A more detailed study of damage mechanisms was therefore
undertaken, using image analysis combined with microscopy to examine the crack tip strain
fields and measure adherend displacements. This is described below and correlation is made
between predicted displacements and failure loads, based on the mixed mode envelope
determined previously, and measured values.
In the final part of the paper the extension of the fracture mechanics approach to the
prediction of the pull-off of top hat stiffeners on sandwich panels is presented. Composite top-
hat stiffeners have been studied by several workers as they are often used in ship structures.
Smith [11] gives an overview of structural connections in marine structures and describes
experimental results. Shenoi and colleagues have described failure mechanisms under pull-off
and flexural loading and compared failure loads with predictions from finite element analysis
[12,13]. For their geometry and materials they predict delamination of the overlaminate under
tensile pull-off loads, using a failure criterion based on a critical value of through-thickness
stress. They also examined strain energy release rates, but for the overlaminate not the bonded
interface [13], In the present work we have concentrated on the latter. Experimental details
and initial results are given and the influence of implanted bond-line defects is discussed.

MATERIALS

The results described in this paper were all obtained from tests on E-glass reinforced
composite materials produced by hand lay-up. This is the manufacturing route most
frequently used for marine structures. For the majority of the tests reported here the E-glass
fibres were either quasi-unidirectional (250 g/m^ with 1 g/m^ of polyester fibres bonded in the
90° direction to keep the UD fibres in place) or stitched quadriaxial (0/45/907-45° 1034 g/m^)
cloths. The same uniaxial ply is used in both cloths. The resin is based on DGEBA epoxy
(SRI 500) with an amine hardener (2505) from Sicomin, France. All epoxy specimens were
post-cured at 90°C for 6 hours. Some results are also shown for a woven glass (0/90° 500
g/m^) reinforced isophthalic polyester for comparison, as this is the traditional marine
 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 281

material. Fibre fraction of the quasi-unidirectional panels was measured by bum-off to be

57±1% by weight, that of the quadriaxial panels was 58 ±1% (this is around 37% by volume).
Starter films for crack initiation were 8 micron thick polypropylene.
The bonded assemblies were produced using two adhesives. The first, which will be
discussed in more detail here, is the matrix epoxy resin employed either to bond the
composite adherends (in the case of lap shear and L specimens) or to overlaminate onto a
balsa sandwich or monolithic composite base in the case of top hat specimens. The second is
a more ductile epoxy resin (Redux 420), used to bond the composite adherends. Figure 1
gives an indication of the tensile behaviour and fracture toughness of these resins.

— Brittle epoxy

Gc 300 J/m^ —Tough epoxy

Gc1400J/m2

0 2 4 6 8 10
Tensile Strain, %

Figure 1. Tensile stress-strain behaviour. Indicative Gc values from tests on cast brittle resin,
and on tough epoxy as adhesive on steel substrates.

FRACTURE MECHANICS APPROACH TO PREDICT FAILURE LOAD

In order to determine the failure load of a structure under a given loading the approach is
essentially as follows: First determine the maximum strain energy release rate in the structure
for the load of interest. This may be achieved either by applying analytical expressions, when
the geometry is simple, or by finite element methods using virtual crack closure for more
complex structures. This gives values of Gi and Gn (or Jj and Jn). to be compared to a fracture
criterion, which has been determined by a separate series of tests on the same materials. If the
loads of interest are well defined only a small part of the mixed mode failure envelope may be
needed, but more generally the complete mixed mode envelope is required. Figure 2 shows
this procedure schematically for a double lap shear specimen.

1. Analysis of specimen, Gi:Gii

Figure 2. Schematic illustration of procedure to determine failure load,

Gc=Gi + Gii, (non-linearity (NL) values shown).
282 P. DA VIES AND J. SARGENT

DETERMINATION OF MIXED MODE FRACTURE ENVELOPE

The use of fracture mechanics tests to characterise structural adhesives was proposed over 30
years ago by RipHng, Mostovoy and colleagues, e.g. [14]. Interlaminar fracture testing of
composites and adhesively bonded composite assemblies has been reviewed recently [15].
The loading of interest for the structures described in the present paper are in the mode I
region so the test configurations used were the double cantilever beam (DCB) for pure mode I
and the mixed mode MMB for mode I/mode II. The determination of the mixed mode fracture
envelope for the composite and adhesively bonded specimens has been described elsewhere
[16,17]: the envelope shovm in Figure 2 comes from that work. Several initiation criteria can
be defined, based on values taken from the load-displacement plots (e.g. non-linearity,
maximum load), acoustic emission parameters or visual crack observations. It seems
reasonable to apply these values to predict the behaviour of the same materials in the form of
small assemblies such as lap shear specimens, produced by bonding composite adherends
with the matrix epoxy resin, provided the adhesive bond-line thickness is similar to the
interlaminar layer thickness and that the same criterion is applied. This exercise will be
described in the following section.

ANALYSIS OF TEST SPECIMENS

A first step in the validation of this approach is to test simple specimens under controlled
conditions and to compare predictions with measured failure load values. First lap shear
geometries were examined, then an L-geometry was studied in more detail. The bond-line in
these small specimens was very similar to that in the quasi-unidirectional fracture specimens
as the small dimensions allow panels to be pressed uniformly after assembly (which is not the
case for industrial top-hat stiffeners).

Lap shear specimens

Femlund and Spelt [5] used large deformation beam theory (permitting modelling of non-
linear geometric behaviour) together with the J formulation of the critical strain energy release
rate to derive fracture load predictions for cracked lap-shear joints. This had the merit that
they were accessible, and could be applied relatively easily to a range of different geometries
based on different adherend lengths and thickness. The solutions assumed that the adhesive
layer was sufficiently thin such that the global deformation of the joint was determined
entirely by adherend material. Assuming no influence of adhesive is probably reasonable
given long free adherend lengths. However, if the adherends were short, giving rise to a short
effective crack length, or if the adhesive layer was thick, then the adhesive would make a
significant contribution to the global deformation, and the analysis would be approximate. In
addition, it should be noted that the analysis did not deal explicitly with the consequences of
any non-linear plasfic behaviour of the adherends (e.g. as a resuh of damage or yielding) or
adhesive. In spite of these limitations, however, the authors noted good agreement between
theory and experiment for a rubber toughened adhesive system with very thick (12.7 mm)
aluminium adherends and long free adherend lengths (-140 - 260 mm).
In order to examine the application of this approach to composite assemblies, tests
were performed on single and double lap shear specimens as shown in Table 1. Test results
were compared with predictions based on reference [5] ^ Figure 3 shows representative
results for both adhesives.

NB Equation 25 in reference [5] is incorrect. The correct equation, used in tlie analysis here, is:
-DiC3>.i^cosh(>.iLi) = -D2>.2^{C7COsh(?.2Li) + C8sinh(>.2Li)}
 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 283

Single lap shear SLS Double lap shear, DLS

Specimen

Note two acoustic emission

transducers on each
specimen ^B^lT^P^^
Thickness (mm) 3,6 3/3/3 & 3/6/3
Overlaps (mm) 10,20,30 10,20,30
Adhesive SR1500, Redux 420 SR1500, Redux 420 j
Joint end geometry With & without fillet No fillet

Table 1. Lap shear tests performed

Redux 420

z
"S 12000 SR1500

i
Test no
fillet
Model Test no
fillet
Test with
fillet
SI

Model

Figure 3. Test-prediction comparisons for 3mm adherend single lap shear specimens.

The correlation is quite good for the SRI 500 resin, while for the more ductile adhesive resin
the predictions overestimate the measured failure loads. However, in the latter case an
extensive damage zone develops before final failure and the non-linear elastic fracture model
is no longer appropriate. It is interesting to note however, that when a fillet is left at the end of
the overlap the test values are much closer to the predictions.

Strain mapping and deformation analysis of "L " type joints.

The "L" type joint used here and shown in Figure 4 represent a common generic
geometric element of structures used in modem aircraft designs.

Figure 4. Schematic diagram showing the loading and geometry for the "L" type specimen.

These could be found, for example, in "zed" section stringers or parts of rib-elements in
aircraft wings. In-situ testing of small, approximately 2mm wide sections, of such structures
was performed on an optical microscope using a "Minimaf miniature materials tester [18]
284 P. DAVIESANDJ. SARGENT

with the loading arrangement shown schematically in Figure 4. Image analysis [19] was then
used to spatially correlate surface features from images taken in the initial unloaded state,
with the same features in the loaded state, to derive strain field maps within the adhesive
bond-line and to make measurements of adherend displacements. Figure 5 shows an example
of a typical vector displacement map for the detailed inset region of the above specimen.

Figure 5. Vector displacement map for the inset region of the specimen from Figure 4. Heavy
dashed white lines show adherend outline. Lower adherend approximately 1.6 mm thick.

It was estimated that displacements could be obtained with an accuracy of better than 1/20^^ of
a pixel, giving an equivalent accuracy of approximately 0.2JLI when using a 2.5x microscope
objective, and strains could be measured with an accuracy of approximately 0.1%. Figure 6(a)
and (b) shows, respectively, maps of the tensile (^Uy/5y) and shear {dwjdy +dnyldx) strain
components recorded for the specimen from Figure 5 with an intact fillet and 1.6 mm thick
SRI 500 composite adherends under a load P3 of approximately 20 N/mm.

0.02

•g O.OIH

0.0 I

(a) {dviyldy) (b) {dwjdy +auy/ax)

Figure 6a, b. Strain maps for the specimen with a fillet from Figure 5. Load P3 = 20N/inm.

Examination of these images showed significant strain developed throughout the whole fillet
with maximum strain values of approximately 1% for the tensile (Syy ) and shear (Cxy =
{dwxidy +d\Jiyldx)l2 ) components at a region located within the fillet and adjacent to its free
surface.
Specimens without a fillet were also tested in which starter cracks had been
deliberately introduced into the bond line. Figure 7 (a) and (b) shows, respectively, the tensile
and shear components of strain for an example of such a specimen with a crack of length
0.5mm.
 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 285

0.02

c I
•g 0.01
00

0.0 I

(a) {dwyldy) (b) {d\xjdy +d\Xyldx)

Figure 7. Strain maps for specimen without a fillet and with a 0.5 mm crack (effective crack
length "a" = 4.8 mm). Load P3 = lON/mm.

This specimen, which failed at a load P3 of 11.7N/mm, is shown at a load before failure of
approximately lON/mm. Inspection of Figure 7 shows that whilst the bulk of the adhesive
shows tensile strains of approximately 0.1%, regions of tensile and shear maxima apparently
followed the outline of the interfacial region between adherend and adhesive, with some
regions indicating strains in excess of 2-3%. It should also be noted however, that a careful
inspection of the vector displacement map showed that these large apparent strains were in
fact due to a discontinuity in the displacement vectors as a result of sub-critical damage
accumulation ahead of the nominal crack front. Figure 8 shows a plot of the adherend
displacements (Uy) as a function of "x" for the specimen from Figure 7.
1....LJ i L
5r
'^ 1
X r xifeiais Fflajc bnXejnt
"^"^x;^ ^ i „ 1
^ ToV'<
idhen
c
^ X-J....
ocd ^
i"^ .440W eri.^ 4dh<j renldttt ±:x :t:: :±:
'^
U-l(|) 1400
"x"iDixel s

Figure 8. Transverse adherend displacements Uy for the specimen from Figure 7.

The nominal crack front was located at an "x" pixel position of 450, equivalent to an effective
crack length "a" of approximately 4.8 mm from the loading point P. The dotted line shows the
net w(x) displacement difference between the adherends.

Beam theory modelling of "L " type adherend deflection

Beam theory methods offer a potentially simple method for modelling the deformation
of the adherends and also for the prediction of failure loads using linear elastic fracture
mechanics. These methods usually make use of beam on elastic foundation models in order to
describe the deformation of the adherends, from which the strain energy release rates can then
be calculated. An analysis of the transverse beam deflection w(x) was undertaken here by
approximating the lower adherend as a single beam resting on an elastic foundation of
modulus "k" comprising a variable adhesive bond thickness and fixed rigid upper adherend of
constant thickness. The accuracy of such an approximation was determined by comparing the
predicted beam displacement with that measured experimentally (shown in Figure 9).
286 P. DAVIESANDJ. SARGENT

10.0
Approximate crack
Effective crack length "a" (mm)
location
Figure 9. Comparison between predicted and experimental measurements of net beam
displacement (w(x)) for the specimen from Figure 5.

Formally, the differential equation relating the transverse deflection of a simple beam of
width "b", thickness "h" is given by:
,4

-^p-H4l^w(x) := 0 where: X^ = 3.k/Ex.b.h^

dx'
Ex = modulus of adherend in "x" direction (25.6GPa), k = foundation modulus, with a
general solution of the form:
w(x) := C.cos(Xx)cosh(^x) + C.cos(lx)sinh(^x) + C.sin(\x)cosh(lx) + C.'sinh(^x)sin(^x)
This follows analysis given by Sargent and Wilson [20] whereby the adherend was
approximated as a partly free beam and a beam partly supported on the adhesive. The partly
supported beam was divided into five discrete sections, each of which rested on a Winkler
elastic foundation with a foundation modulus which was representative of the adhesive
thickness under that particular part of the beam, together with a contribution from the
deforming beam and fixed beam. A Winkler type foundation can be imagined as a continuous
array of independent linear springs which simulates the transverse elasticity of the uncracked
region. By matching boundary conditions at each beam element junction this resulted in a set
of six simultaneous differential equations which were solved numerically. Figure 10 shows
the resulting stress distribution calculated using this approach. It may be noted that the
stepwise changes in stress reflect the 5 discrete foundation modulus sections used in the
analysis.
80-

5.8 6.8 7. 9.8

Effective crack length "a" (mm)
Figure 10. Approximate stress distribution (ay) for "L" type specimen predicted for load P3 of
lON/mm. Predictions based on 5 stepwise adhesive thicknesses from 0.8 mm to 0.08 mm
at a = 4.8 mm, 5.6 mm, 6 mm and 6.4 mm. See text for details.
 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 287

Reference to Figure 9 shows that the predicted displacement is underestimated using

this procedure. Using a modulus of 2.8 GPa for the adhesive in combination with the stress
derived from Figure 10 implied predicted tensile strain levels adjacent to the crack front (a =
4.8 mm) of approximately 1%. These may be compared with the measured values of
approximately 0.1% for the bulk adhesive strains shown in Figure 8. It is likely that these
discrepancies between theoretical and experimental strain values arise from the exclusion of
the damaged regions, which were noted in Figure 7.

Failure load prediction of'L" type joints

If the "L" type specimen is considered to act as a single deforming beam which can be
modelled using simple linear elastic beam theory, and that both damage accumulation and
adhesive shear deformation may be ignored, then a simple first approximation to estimating
failure loads may be derived by considering that the failure load is controlled simply by a
critical mode I strain energy release rate. Making the further assumption that a single average
adhesive bond line thickness may be used, Sargent and Wilson [20] shows that the failure
load prediction (P) for the single beam arm of the specimen with the current geometry may be
found from:
6 P -a Ex
1-H
r» uO-25
where: B:= 3
b^-h^-Ex B^.h^-^ 8Gxy Exb/
Bh -a
where: Gxy = shear modulus of adherend (2.5GPa), a = effective crack length (distance
between load point "P" and crack front shown in Figure 4).
Noting that the specimen is statically determinate, then the failure load P3 is related to
P from Figure 4 via P = P3/ (1 + L1/L2). Figure 11 shows a comparison between the predicted
(P3) and experimental failure loads as a function of the effective crack length. In general,
failure loads are underestimated using this procedure, though there is some scatter in results.
20
18 • Experimental pointsj
16 -»- Predictions
14
12
o 10-1
8-1
6
4-]
2-|
0
3.5 4.0 4.5 5.0 5.5 6.0
Effective crack length "a" (mm)
Figure 11. Comparison between predicted and experimental failure loads (P3) for "L" type
specimen as a function of effective crack length. Using Gic = 240 J/m^.

APPLICATION TO A MARINE TOP-HAT STRUCTURE

The results above suggest that it may be possible to apply fracture mechanics data to
determine failure loads of more complex structures, provided that (i) the adhesives used are
not too ductile, (ii) bondline thickness is known and controlled, (iii) non-linear behaviour due
to adherend and interface damage is limited, and (iv) the specimens employed to determine
288 P. DAVIESANDJ. SARGENT

the fracture envelope are representative of the real structure. In order to examine this two
series of top hat stiffeners were manufactured. These are typical of real stiffeners in naval
superstructures. The first series used a woven glass fibre reinforced polyester representing
current shipbuilding practice. The second series was based on the glass/epoxy material. For
both materials, in addition to the standard stiffener additional specimens were prepared with
starter cracks (8 micron thick polypropylene film) of different lengths (20, 40 and 60 mm)
placed at the stiffener/base interface during manufacture. The test set-ups used are shown in
Figure 12. When the base is a stiff balsa sandwich material the loading is pure mode I, Figure
12(a). As the base becomes thinner there is a mode II component. Figure 12(b), so a simple
tensile pull-off allows the I/II mode mix to be varied. In the limited space available here only
results from the former will be given.

Figure 12. Top hat stiffener pull-off test set-up (a) mode I, (b) mixed mode I/II.

Two digital cameras interfaced with a PC and a load input were used to record the crack
length and opening displacement continuously. An image analysis programme then enabled
crack length versus applied force and crack opening displacement plots to be generated. Two
acoustic emission transducers were placed on the stiffeners and interfaced to a second PC.
The load and machine cross-head displacement were recorded on a third PC. The specimens
without implanted defects will be considered first, it is apparent from the load-displacement
plots that replacing the woven glass/polyester by a QX/epoxy material results in a very
significant increase in pull-offload. Figure 13.
Pull-off tests, no defect
8000

QX/Epoxy

5 10 15 20
displacement, mm

Figure 13. Load-displacement plots, top-hats without defects.

 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 289

However it should be emphasised that many factors affect pull-off load including geometry,
fillet material, and surface preparation. The top-hats employed here were produced to allow
the existing fracture data to be applied, they are certainly not optimised.
When specimens with implanted defects are tested there is a very large drop in pull-off
load for both materials. Examples of load and crack length versus time are shown in Figure
14. For the shortest defect the crack propagation is unstable and asymmetric but for the longer
defects stable, symmetrical propagation was observed.

QX/Epoxy pull-off Nominal ao = 40mm

QX/epoxy pull-off, Nominal ao = 20mm

Figure 14. Load and crack length versus time plots top-hats with defects.

Transverse pull-off tests induce mainly mode I loading, provided the base panel is sufficiently
rigid. Finite element analyses have been performed to look at this geometry in more detail,
and will be reported elsewhere, but here a simple analytical beam theory expression is used to
predict the pull-off failure load [21]:
4B'h'EGjc
F=2 [
21a' ]
The results from 12 pull-off tests on QX/epoxy specimens with implanted defects are shown
in Figure 15. Both measured and predicted values are shown. Different criteria may be used to
compare top hat pull-off and fracture test values. These include various acoustic emission
parameters (first acoustic events, first events above a certain amplitude), visual or image
analysis parameters or values on the load-displacement plots. Several criteria have been
examined, here non-linear values are shown (Gic = 240 J/m^, the lower, dashed line).
QX/Epoxy top hat pull-off (E=15 GPa, t=5mm)
4000 1 M"

3000
A • NL experimental points
1 \\
\\ - - Gc = 240 J/m^
\\
\\
I 2000 Gc = 330J/m2
c 1 ^%."
o
c
w 1000
o

04
H 1

0 0.02 0.04 0.06 0.08

Defect length, m

Figure 15. Test results and predictions, QX/epoxy top-hats on balsa sandwich.
290 P. DA VIES AND J. SARGENT

FURTHER CONSIDERATIONS

While the results from this first attempt to use fracture data from standard specimens to
predict failure loads are quite encouraging there are a number of aspects which have been
neglected and these must be examined in more detail if the work is to be extended in the
future.

Size effects
One point which should be addressed is size effects. This is important, not only because
dimensions of marine structures can be very large and the hand lay-up fabrication can lead to
rather heterogeneous materials, but also because the reinforcement repeat unit size may be
tens of millimetres. Larger specimens were therefore produced, 50 mm wide and 350 mm
long rather than the standard 20 mm by 140 mm dimensions tested previously. These large
specimens were manufactured from exactly the same materials as the top hat stiffeners, i.e.
four layers of quadriaxial glass cloth each side of the starter film for a total specimen
thickness of 8 mm. The layers were arranged symmetrically about the specimen mid-plane,
with the starter film placed between two 0° layers, i.e. exactly the same interface as that in the
specimens used previously. A special scaled-up MMB fixture was developed to test them
under mixed mode loading. Here only the mode I results will be presented, in the following
section.

Manufacturing effects
If the production assembly process influences the material structure, compared to that found
in the quasi-unidirectional specimens used to obtain the fracture envelope, then there is no
reason for predictions to correlate with measurements. Two manufacturing procedures were
studied, the first was a continuous lay-up in which a panel was produced with no pause during
manufacture. This corresponds to the composite adherend fabrication. The second procedure
was in two steps. Half the panel thickness (4 layers) was impregnated. Fabrication was then
stopped, a peel ply layer was applied and the panel was left overnight. The following day the
peel ply was removed and the four remaining layers were laminated. This second procedure
corresponds more closely to real manufacturing of top hat stiffeners by over-lamination,
where large panels are produced first and stiffeners are added later. It can result in poor
bonding and/or a resin rich layer at the interface between the stiffener and base plate. When
specimens of both materials were tested there was no significant difference in initiation \alues
for the epoxy composite but the polyester material showed lower values for the material
manufactured with a delay. Thus while tests on continuously-produced specimens indicate
very similar values for the polyester and the epoxy when the real fabrication process is used
the former are only half those of the latter and this is reflected in the pull-offloads.

Woven/Polyester QX/Epoxy UD/Epoxy

50 X 350mm2 50 X 350 mm^ 20 X 140 mm^
Continuous 247 (9%) 292 (28%) 242(11%)
With delay 177(6%) 331 (19%) -

Table 2. Initiation values of Gic at non-linearity, J/m^ (coefficients of variation in brackets)

The values measured on the QX/epoxy with delay are also higher than those for the UD
material. Predictions with this higher value are shown on Figure 15, but are not sufficient to
explain the under-estimation of the pull-offloads in the top hats with longer defects.
 Fracture Mechanics Tests to Characterize Bonded Glass/Epoxy Composites 291

Another aspect to be considered is the difficulty in producing curved structures with

the same fibre content as flat laboratory panels. This effect is shown in Figure 16, at the
comer the laminate thickness is larger than at the flat section and fibre content is rather lower.
This will affect the bending stiffness of the arm and the predicted failure load. This figure also
shows the fillet, which is critical to initiation in the specimens without implanted defects. It is
well known that fillets can significantly alter the load path in lap shear joints and increase the
failure loads (see [1] and Figure 3 for example). If a fracture mechanics approach is to be
applied this effect must be considered. Some recent studies on stress intensity factors for such
cases may allow this to be addressed [22].

Figure 16. Detail of top hat stiffener crack tip

Ductile adhesives
While the approach described above works reasonably well for the polyester and epoxy
matrix resins which do not show extensive ductility there is clearly a need for an alternative
approach when very ductile structural adhesives are applied. The development of extensive
damage zones can be treated more efficiently using damage mechanics models and these are
now being examined.

3-D structures
Finally, it should be emphasised that the objective of this work is to optimise design of real
three-dimensional structures such as stiffened panels. While the top hat section is part of this
structure debonding is generally seen to initiate at the stiffener ends under complex (mixed
mode) loading and stiffener/panel debonding is not the unique failure mechanism. The top hat
pull-off test is still of some interest, as mixed mode loading (I/II) can be studied by varying
the plate thickness of the lower panel (Figure 12b). However, tests on real structural elements
are important if the correct failure mechanisms are to be modelled and some of these are
being examined in a current project (EUCLID RTF 3.21, [23]).

CONCLUSION

Fracture mechanics characterisation tests have been performed to determine the mixed mode
fracture envelope of an epoxy bonded glass/epoxy composite. Analysis of lap shear, and L-
stiffener geometries has shown that for this relatively brittle adhesive reasonable first
estimations of failure loads can be obtained for both cracked and uncracked specimens. An
image analysis technique has been developed which enables failure mechanisms to be
292 P. DA VIES AND 1 SARGENT

analysed in more detail. Damage development and accumulation have been observed before
final fracture, and must be taken into account if better predictions are required. The extension
of the predictions to an industrial top hat stiffener shows that while the fracture mechanics
properties do mirror the structural behaviour many geometrical and manufacturing details
must be taken into account if reliable predictions are to be made even for this simple
geometry. Further work is concentrating on more detailed damage modelling and more
complex loading cases.

REFERENCES
[I] Adams RD, Comyn J, Wake WC, Structural adhesives joints in engineering, Chapman & Hall, 2"^^
edition, 1997.
[2] Tong L, Steven GP, Analysis and design of structural bonded joints, Kluwer Academic, 1999.
[3] Curley A J, Hadavinia H, Kinloch AJ, Taylor AC, Predicting the service life of adhesively bonded
joints. Int. J. Fracture, 103, 2000, 41-69.
[4] Papini M, Femlund G & Spelt JK, Effect of crack growth mechanism on the prediction of fracture
load of adhesive joints, Comp. Sci. & Tech., 52, 1994, 561.
[5] Papini M, Femlund G & Spelt JK, The effect of geometry on the fracture of adhesive joints, Int. J.
Adhesion & Adhesives, Vol 14, 1, 1994, 5.
[6] Allix O, Leveque D, Perret L, Identification and forecast of delamination in composite laminates
by an interlaminar interface model, Comp. Sci & Tech., 58, 1998, 671-678.
[7] Tvergaard V, Hutchinson JW, The relation between crack growth resistance and fracture process
parameters in elastic-plastic solids, J. Mech & Physics of solids, 40, 6, 1992, 1377-97.
[8] Chandra N, Li H, Shet C, Ghonem H, Some issues in the application of cohesive zone models for
metal-ceramic interfaces. Int. J. Solids & Structures 39, 2002, 2827-55.
[9] Davies et al. Failure of bonded glass/epoxy composite joints, Proc. Structural Adhesives in
Engineering (SAE6), Bristol, June 2001.
[10] DOGMA thematic network project website http://www.dogma.org.uk/
[II] Smith CS, Design of Marine structures in composite materials, Elsevier Applied Science, 1990,
section 4.6.
[12] Dodkins AR, Shenoi RA, Hawkins GL, Design of joints and attachments of FRP ships' structures.
Marine Structures, 1994, 7, 365-98.
[13] Phillips HJ, Shenoi RA, Moss CE, Damage mechanics of top-hat stiffeners used in FRP ship
construction. Marine Structures, 12, 1999, 1-19.
[14] Ripling EJ, Mostovoy S, Corten HT, Fracture mechanics: a tool for evaluating structural
adhesives, J. Adhesion, 3, 1971, 107-123.
[15] Moore DR, Pavan A, Williams JG, eds.. Fracture Mechanics Testing methods for Polymers,
Adhesives and Composites, ESIS Publication 28, 2001, Elsevier.
[16] Ducept F, Gamby D, Davies P, A mixed mode failure criterion from tests on symmetric and
asymmetric specimens Comp. Sci & Technology, 59, 1999, p609-619.
[17] Ducept F, Davies P, Gamby D Mixed mode failure criteria for a glass/epoxy composite and an
adhesively bonded composite/composite joint. Int. Journal of Adhesion & Adhesives, 20, 3, 2000
p233-244.
[18] "MINIMAT" miniature materials tester, Polymer Laboratories, UK.
[19] "microDAC", Fraunhofer, IZM, Gustav-Meyer-Allee 25, D-13355 Berlin.
[20] J P Sargent and Q Wilson, "Prediction of "Zed" section stringer pull-offloads". Accepted for
publication in Int. J. Adhesion and Adhesives, 2002.
[21] Hashemi S, Kinloch AJ, Williams G, Mixed mode fracture in fiber-polymer composite laminates,
ASTMSTP 1110, ed O'Brien TK, 1991, 143-168.
[22] Wang CH, Rose LRF, Compact solutions for the comer singularity in bonded lap joints, Int J.
Adhesion & adhesives, 20, 2000, 145-154.
[23] EUCLID RTP 3.21, website http://research.dnv.com/euclid_rtp3.21/