You are on page 1of 12

Composite Structures 116 (2014) 441452

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

A study of intra-laminar damage in double-lap, multi-bolt, composite


joints with variable clearance using continuum damage mechanics
Yinhua Zhou a,b, Hamed Yazdani-Nezhad b, M.A. McCarthy b,, Xiaopeng Wan a, Conor McCarthy b
a

Northwestern Polytechnical University, Xian, China


Department of Mechanical, Aeronautical and Biomedical Engineering, Irish Centre for Composites Research, Materials and Surface Science Institute, University of Limerick,
Limerick, Ireland
b

a r t i c l e

i n f o

Article history:
Available online 10 June 2014
Keywords:
Composites
Bolted joints
Clearance
Continuum damage mechanics
Stiffness
In-plane damage

a b s t r a c t
A damage modelling approach, based on a continuum damage model (CDM) formulation, is proposed and
applied to the problem of double-lap, multi-bolt, bre-reinforced composite joints with variable clearances, subjected to quasi-static tensile loading. A new method of dealing with bre failure is included
in the CDM model, which is implemented in a commercial implicit nite element analysis code. The
model is validated at element and structural levels by comparing with experimental data. It has been
found that, for the joints examined in this paper, our formulation is capable of modelling development
of damage from bearing failure onset all the way to ultimate catastrophic net-tension failure without
numerical problems, which is an advance over previous work. The predictions from the CDM model of
net-tension failure modes and ultimate loads are in good agreement with those from the experiments.
Furthermore the model is capable of explaining some non-intuitive experimental ndings, such as the
larger energy absorption obtained in joints with higher clearances.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Composite bolted joints (CBJs) are widely used in critical structures in aerospace and energy applications. Though competing
with rapidly growing application of adhesive bonding and bonded
repair techniques [1], CBJs are still of major interest, especially
when thick composite parts are to be joined or where parts need
to be disassembled for inspection or repair during the life time of
the structure. The joints are the critical parts in the structure due
to the stress concentrations and bre discontinuities they
introduce. Hence, various methodologies have been developed to
investigate the structural response of the CBJs subjected to various
loading conditions, by means of numerical analysis, e.g. nite
element (FE) analysis, and/or experimental testing [24].
The two most well-known FE approaches for modelling brereinforced polymer (FRP) composite material are Progressive Damage Analysis (PDA) [5] and continuum damage mechanics (CDM).
PDA has been widely used for modelling CBJs, whereas the use of
CDM is much less prevalent for CBJs in the open literature. In
general, PDA is governed by material damage laws when an
undamaged element material reaches a critical stress and/or strain

Corresponding author. Tel.: +353 61 202222; fax: +353 61 202944.


E-mail address: michael.mccarthy@ul.ie (M.A. McCarthy).
http://dx.doi.org/10.1016/j.compstruct.2014.05.051
0263-8223/ 2014 Elsevier Ltd. All rights reserved.

state. Each element is assumed to be either in an intact state or


rupture state. However, the CDM approach assumes that the element material transforms smoothly and gradually from the intact
state to the rupture state. Ladeveze and Le Dantec [6] developed a
CDM model based on a plane stress assumption for unidirectional
bre-reinforced composite plies to describe matrix micro-cracking
and bre/matrix debonding. OHiggins et al. [7,8] and Frizzell et al.
[9,10] then employed the model, enhanced it to account for bre
breakage, and extended it to a three-dimensional (3D) meso-scale
to predict the damage response of notched CFRP composites under
quasi-static and dynamic loading. This same CDM model was
recently enhanced by combining a delamination model with the
in-plane damage model in Nezhad et al. [11], and successfully
applied to model low velocity impact of at, unnotched CFRP panels manufactured from a high strain composite material used for
fuselage barrels.
Recently, 3D FE modelling has been extensively used to study
the damage response of CBJs [4,5,1217], sometimes in combination with experimental work. Various types of joint have been
investigated, such as single-lap, double-lap, single-bolt and
multi-bolt joints, as well as joints with countersunk and protruding
head bolts. Several of these studies have noted that bolt clearance
plays a key role in the performance and damage tolerance of CBJs.
McCarthy et al. [17] experimentally investigated the effect of bolthole clearance (from 0 lm to 240 lm) on the stiffness and strength

442

Y. Zhou et al. / Composite Structures 116 (2014) 441452

of single-bolt, single-lap CBJs. The joint stiffness, 2% offset bearing


strength, and ultimate bearing strength and strain were obtained
in [17] according to ASTM [19]. The results indicated that clearances in the range used in aircraft applications can affect the 2%
offset strength and ultimate bearing strain in protruding-head
joints, but have little effect in countersunk joints. Based on this
work, Egan et al. [20,21] carried out further assessment, using FE
analysis, on the mechanical behaviour of single-bolt, single-lap
countersunk joints. By providing a highly-detailed representation
of stress state surrounding the countersunk hole, they were able
to elucidate the failure mechanisms in countersunk joints.
Concerning multi-bolt composite joints, McCarthy et al. [15,22]
carried out a series of experimental tests to investigate the effect of
variable bolt-hole clearance on the quasi-static strength, fatigue
life and failure modes in double-lap and single-lap CBJs containing
three protruding head bolts. The effects on quasi-static ultimate
strength were found to be relatively small, chiey affecting the failure mode, and failure strain, while the failure load was only mildly
affected. In contrast, the failure onset load, dened in [15,22] as the
load corresponding to 30% loss in joint stiffness, had a strong correlation with clearance. The fatigue tests also showed that joints
with one loose-t hole had a signicantly shorter fatigue life than
joints with all neat-t holes. McCarthy et al. [16] also developed a
3D FE PDA model of the multi-bolt, double-lap joints in [15,22],
and surface strains and loaddisplacement response correlated
well with the quasi-static experimental results. Bolt load plots
showed that minor changes in clearance caused major changes in
load distribution. The effects of clearance on damage progression
were also studied. However, convergence problems meant that
the numerical model could only follow the response up to just after
initial bearing failure, which was only a small fraction of the damage and failure response observed in the experiments. The nal
failure mode in the experiments was net-tension failure, which
the PDA model could not capture. Consequently, the ultimate load
level was not captured either.
From the above, it can be seen that studies on bolted joints with
CDM models are few in number compared to those using a PDA
model. PDA models have exhibited convergence problems in previous investigations of joints, meaning that ultimate load cannot be

predicted. In the present paper, a CDM model is suggested to alleviate these convergence difculties. An important feature which
has been found to improve the accuracy of predictions is the introduction of a new approach to derive the initial bre failure strain.
The model is capable of properly capturing the net-tension failure
mode and ultimate load of multi-bolt, double-lap, protruding-head
composite joints with different clearances, which the PDA model in
[16] failed to do. The proposed CDM model is used to correlate the
damage evolution with the joint stiffness variations during failure.
2. Overview of specimens
Brief details of the CBJ specimens studied are given here; for full
details of the experimental set-up see [15,22]. Specimen dimensions are shown in Fig. 1. The ratios of the width (w), edge distance
(e), and bolt pitch (p) to bolt diameter (d) were w/d = 6, e/d = 3, and
p/d = 4.5 respectively. The laminates were manufactured from carbonepoxy HTA/6376, and the quasi-isotropic lay-ups for the skin
(centre) plate and splice (outer) plates were respectively [45/0/
45/90]4s and [45/0/45/90]2s, with nominal ply thickness of
0.13 mm. All tests were carried out with titanium alloy protruding-head bolts, secured with a steel nut and washer. The elastic
material constants for the joint components are given in Table 1.
In the tests, an axial displacement was applied quasi-statically to
the splice plates while the skin plate was held stationary.
The four levels of clearance studied, labelled C1C4, are shown
in Table 2, which shows nominal clearance values, possible ranges
on the reamer and bolt diameters according to the tolerances used
on both, and consequent possible range on each clearance. The
nominal clearances are 0 lm, 80 lm, 160 lm and 240 lm, representing, respectively, percentage clearances of 0%, 1%, 2% and 3%
of the hole nominal diameter, 8 mm. The relatively large clearances
C3 and C4 are slightly outside of recommended tolerance for
aerospace applications, but may exist in practice if improper drilling or fastening procedures are followed. All the bolts had nominal
diameter 8 mm, manufactured with f7 ISO tolerance. For the multibolt joints, six different clearance combinations are studied, as
listed in Table 3. For example, the C1_C1_C1 case had (nominally)
zero clearance in all holes, while the C4_C1_C1 case had a large

Fig. 1. Double-lap, multi-bolt, specimen geometry (all dimensions in mm).

Table 1
Material properties of HTA/6376 carbonepoxy, titanium alloy and steel [15,22].
E011 (GPa)
Lamina

Titanium alloy
Steel

E022 (GPa)

E033 (GPa)

G012 (GPa)

G013 (GPa)

G023 (GPa)

m12

m13

m23

10

5.2

5.2

3.9

0.3

0.3

0.5

140

10

E (GPa)

110
210

0.29
0.3

443

Y. Zhou et al. / Composite Structures 116 (2014) 441452


Table 2
Bolt-hole clearance details.
Clearance
code

C1
C2
C3
C4

Nominal
clearance
(lm)

0
80
160
240

Reamer
diameter

Bolt diameter

Possible
clearance

Min
(mm)

Max
(mm)

Min
(mm)

Max
(mm)

Min
(lm)

Max
(lm)

7.985
8.065
8.145
8.225

7.994
8.074
8.154
8.234

7.972
7.972
7.972
7.972

7.987
7.987
7.987
7.987

2
78
158
238

22
102
182
262

Table 3
Joint clearance cases.
Case code

C1_C1_C1
C2_C1_C1
C4_C1_C1
C1_C3_C1
C3_C3_C1
C1_C1_C4

Nominal clearances (lm)


Hole 1

Hole 2

Hole 3

0
80
240
0
160
0

0
0
0
160
160
0

0
0
0
0
0
240

clearance in hole 1 only. As described in [15,22], special jigs, xtures and reamers were made for the tests, and the bolt loads were
measured with instrumented bolts.
3. Finite element model development
3.1. Parts and assembly of the joint
Finite element modelling was performed here using ABAQUS
[23]. A one-quarter symmetric model was used, as shown in
Fig. 2, taking advantage of the symmetry of the joint shown
in Fig. 1. The gripped areas were not modelled, i.e. perfect
gripping was assumed. The washers and bolts were modelled
and meshed separately. One element per ply was used through
the thickness. Three-dimensional 8-node reduced-integration
(C3D8R) elements with hourglass control were used to discretise
the model. The smallest element size in the laminate was
0.32 mm  0.4 mm  0.13 mm, and the model contained 131,616
elements. The boundary constraints are also shown in Fig. 2. The
nodes at one end of the skin plate model were xed in X, Y and Z

directions. Displacement was applied at the end nodes of the splice


plates along the Z direction. The nodes of faces labelled as Y-symmetric and Z-symmetric in Fig. 2 were constrained, respectively, in
Y and Z direction.
Contact pairs were dened between interacting surfaces of the
parts. Contact was dened by implementing tangential and normal
behaviour with hard pressure-overclosure enforcement [23].
According to McCarthy et al. [24], suitable frictional coefcients
are 0.7, 0.3, 0.3 and 0.1 for laminate-to-laminate, bolt-to-washer,
washer-to-laminate and bolt-to-laminate interfaces, respectively,
and these were used here. A nger-tight torque of 0.5 Nm
was applied to simulate the actual assembly torque, using PRETENSION SECTION in ABAQUS [23]. To specify the clearances
tabulated in Tables 2 and 3, the diameter of the bolts and their
corresponding holes were varied in the model. This approach
proved more reliable than using the built-in CLEARANCE option
in ABAQUS which led to penetration problems. To represent the
neat-t (C1) case, rather than choosing zero clearance and running
into initial contact problems, very small values, viz. 10 lm and
15 lm, were respectively chosen as the bolt-hole clearance and
the washer-hole clearance.
3.2. CDM model with new bre damage method
The basic formulation of the CDM model for predicting damage
initiation and growth used in this study was developed originally
by Ladevze and Le Dantec [6]. They successfully implemented this
meso-scale model for the prediction of transverse and shear damage in composite plies. A generalised (3D) version of this model has
been adopted here [7,9]. The strain energy density ED for a damaged ply is dened as:

"
hr11 i2
hr22 i2
1
hr11 i2
hr22 i2 r2
ED

0  033
0
0
2 E011 1  d1
E11
E22 1  d2
E22
E33
2

m12 r11 r22

2
13
G013

E011

2
23
G023

2

m13 r11 r33


E011

2

m23 r22 r33

with

hai a if a P 0; otherwise hai 0


hai aif a 6 0; otherwise hai 0

Fig. 2. FE assembly model of the joint.

E022

s212
G012 1

 d12
1

444

Y. Zhou et al. / Composite Structures 116 (2014) 441452

where E011 , E022 , E033 , G012 , G013 , G023 and m12, m13, m23 are respectively the
Youngs moduli, shear moduli and Poissons ratios of an undamaged
ply, and d1, d2 and d12 are the bre, transverse and shear scalar
damage variables, respectively, that govern the degradation of the
moduli. These variables take on values from 0 to 1 (=dmax). It is
noted that the transverse tension energy and compression energy
are split since transverse micro-cracks close up under compression.
Thermodynamic forces are derived from Eq. (1) in the form of
the energy dissipation rates, analogous to energy release rates for
crack propagation, which govern development of shear and transverse tensile damage for the matrix. The thermodynamic forces Y2,
Y12 are associated with damage variables d2 and d12 for dissipation,
and are dened as:

Y2


@ED 
r2 J
0 22 s2 2

@d2 r;d1 ;d12 2E22 1  d2

where Js2 = 1 if r22 > 0 else Js2 = 0

Y 12

d1n1


@ED 
r212

@d12 r;d1 ;d2 2G012 1  d12 2

For describing the evolution of (i) matrix micro-cracking under


tensile loading and (ii) bre/matrix debonding, two additional
quantities are introduced:

Yt maxs6t

q
Y 12 s bY 2 s

Y 2 t maxs6t

p
Y 2 s

where Y(t) and Y 2 (t) are set to their maximum values attained over
any previous time s up to the current time t for preventing the
material healing, and b is the shear-transverse damage coupling
factor.
OHiggins et al. [25] carried out numerous statistically robust
experimental tests to describe the damage development laws and
extract the material properties for the HTA/6376 CFRP material
modelled here. For this CFRP material system, OHiggins et al.
[25] wrote d12 and d2 as follows:

d12

8
0
>
>
<
Y 12a ln Yt  Y 12b
>
>
:
dmax

if

Yt 6 Y 12

if

d12 < dmax ; Y 2 t 6 Y S ; Yt 6 Y R

otherwise
6

where Y12a, Y12b are the constants dening the logarithmic function,
YS is the transverse tension brittle damage threshold, YR is the shear
brittle damage threshold and Y 12 0 is the initial shear damage point.

8
0
>
>
<
d2 Y 2a ln Yt  Y 2b
>
>
:
dmax

if

Yt 6 Y 2 0

if

d2 < dmax ; Y 2 t 6 Y S ; Y t 6 Y R

otherwise
7

where Y 2 0 is the start point of transverse damage.


For the case of compressive stress, the matrix material is
assumed to be linear elastic to failure, under the assumption that
matrix micro-cracks close under compressive loading. Hence,
transverse compressive failure only occurs when r22 + r33 < 0,
and [26]:

1
Sc22

Sc22

2

2S23

!
 1 r22 r33

r12 r13 2
S212

P1

where rij and Sij are the stress and strength components, respectively, and superscript c refers to compression.
In general, damage in the bre direction governs the nal fracture mode of the structure. It is common knowledge that carbon
bre, though tough, exhibits brittle behaviour. Although the process from aw initiation to fracture is quite short, it is still necessary to simulate the damage accumulation not only for the
accuracy of the CDM model, but also for the sake of simulation convergence, as instantaneous (sudden) development of bre failure
may cause numerical uctuation and hence instability. In [25],
bre damage evolution is governed by a simple criterion based
on the strain in the bre direction only. Fibre damage initiates at
a user-dened initial failure strain ei11 . It then develops according
to Eq. (9).

r22 r33
4S223

2
23

 r22 r33
S223
8

8
0
>
>
>
>
< u e11 ei11
n1
d1 eu11 ei11
>
>
u
>
>
: 1  1  du1 e e11
11

if

e11n1 < ei11

if

ei11 6 e11n1 6 eu11

if

n1

u
11

e11n1 > e

where eu11 is the ultimate failure strain that is obtained from


uniaxial tensile testing of unidirectional bre-reinforced laminates
u
loaded in bre direction and d1 is the bre damage value at eu11 .
Both compression and tension failure evolution use the same
methodology.
Using Eq. (9) good agreement with the experimental data has
been obtained for notched and unnotched composite plates in
[25]. However, we have found that for complex structures such
as CBJs, the maximum strain criterion signicantly overestimates
the joints mechanical capability, i.e. ultimate load. This is believed
to occur because the criterion does not take into account the complexity of the stress state in tearing/shearing behaviour of the joint.
In contrast, the Hashin [26,27] criterion which is widely used in
PDA methods includes the shearing mechanism, but such PDA
methods lack continuity of damage evolution and have convergence difculties as mentioned before. Consequently, these two
methodologies are combined in this work, i.e. Hashins criterion
is initially called within the subroutine to calculate the bre failure
initiation in each element as a function of its stressstrain state. Eq.
(10) represents the criterion:

8  2  2  2
>
< rX11 rS 12 rS 13 P 1
T
12
13
 2
>
r
: 11 P 1
XC

r11 > 0
10

r11 < 0

where XT, XC, S12, and S13 are strengths for bre tension, bre compression, shear in 12 plane and shear in 13 plane respectively.
Once the criterion is satised, the corresponding strain is stored
as the initial failure strain, ei11 for use in Eq. (9), and then passed
to the next increment of the simulation to continue damage accumulation according to Eq. (9), which gives the bre damage variable, d1. Then, the damage variables (d1, d12, d22) are used to
update the stiffness matrix, and determine stresses in the next
increment.
Data for the CDM model are given in Table 4. The strength constants are common properties that are widely used in PDA models,
e.g. in [16], while the other parameters required in Eqs. (4)(10)
are tted from the experimental data in [25].
3.3. Validation of damage model
To validate the proposed CDM model with the new approach to
evaluate the initial failure strain, comparisons are made at the element scale and the structural scale. Stressstrain curves from single-element and four-element FE models are compared with

445

Y. Zhou et al. / Composite Structures 116 (2014) 441452


Table 4
Data for CDM model.
ST11 MPa

SC11 MPa

2200
p
Y 2b MPa
1.623

1.55

p
MPa

SC22 MPa

S12 MPa

S23 MPa

S13 MPa

Y 12a

1600

250

eu11 %

eu11c %

300
b

120
dmax

120
u
d1

Ys

0.3087
p
Pa

YR

1.48

-0.296

0.9

0.99

2700

565

experiments from [25], and loaddisplacement characteristics of


joints (at the structural scale) are compared with the experiments
in [22].
The damage model is implemented in ABAQUS, using the userdened material subroutine, UMAT. As in-plane damage mechanisms comprising bre fracture, transverse matrix damage and
shear damage are modelled here, the comparisons at the element
scale are made in bre, transverse and shear directions. Single solid
elements were used for modelling behaviour in the bre and transverse directions with plies oriented in the 0 and 90 directions,
respectively. Boundary conditions and results are shown in
Fig. 3(a) and (b). For modelling shear behaviour, four elements
each containing a single ply were stacked in a (45/45)s lay-up
and pulled axially as shown in Fig. 3(c). As seen, the CDM model
provided accurate prediction of strainstress behaviour for these
models over the whole loading duration until fracture. The predicted shear damage response is non-linear as in the experiments.
It is well known that bre-reinforced composites display signicant non-linear shear stressstrain behaviour [28]. Note that this
non-linear shear behaviour plays an important role in bre damage
evolution and hence ultimate failure of the material, which is
captured in our new bre damage model in Eq. (10).

Y 12b

p
MPa

1.7125
p
Pa

Y 2a

p
MPa

0.2961

Validations at the structural scale focus on loaddisplacement


curves for the CBJs, as presented in Fig. 4(a)(f). Detailed analysis
of the tests in [22] showed that the joints failed initially in bearing
at one or more holes at loads varying from 50 kN for the all neat-t
case (C1_C1_C1) to 37 kN for the C3_C3_C1 case. All the joints
failed ultimately by net-section failure with only minor variations
in ultimate load. As seen in Fig. 4, good agreement between the FE
solution and the experimental data is obtained for all combinations
of clearances. The nal part of the FE curves shows a non-linear
and almost smooth degradation due to damage accumulation,
not in keeping with the catastrophic net-section failures in the
tests; this is a necessary outcome of maintaining convergence all
the way to failure. Nevertheless, the disparities with experiment
shown in Fig. 4 are small.
One interesting point is that the stiffest joint, i.e. C1_C1_C1,
failed in net-tension quite quickly after the initial bearing failure,
with a maximum joint displacement of only 3 mm. In contrast,
joints with clearance displayed a higher level of non-linear (damaging) behaviour before nal net-tension failure, e.g. the
C3_C3_C1 joint lasted until a displacement of 4.8 mm (which is
not explainable solely by the addition of a total of 260 lm
clearance in the joint). This behaviour has been noted before for

Fig. 3. Validation of the model at the element scale: (a) bre damage response of HTA/6376; (b) matrix damage response of HTA/6376; (c) shear damage response of HTA/
6376.

446

Y. Zhou et al. / Composite Structures 116 (2014) 441452

Fig. 4. Validation of CDM model at the structural scale. Loaddisplacement curves of (a) C1_C1_C1; (b) C2_C1_C1; (c) C4_C1_C1; (d) C1_C1_C4; (e) C1_C3_C1 (f) C3_C3_C1.

single-bolt joints [17], where more volume of the laminate material exhibited damage at nal failure at large clearances, leading
to the suggestion that clearance could be advantageous for joints
designed to absorb energy. The interesting point is that the model
has captured this, i.e. in Fig. 4(f) we note the load stabilises again
even after the sharp drop at 3.8 mm displacement, indicating there
is still more load-carrying capacity available through further damage evolution before the nal net-tension failure at 4.8 mm. This
point is returned to in the next section.
With the CDM model shown to produce good results at the element and structural level, the next section discusses the FE prediction of the model for the CBJ in detail.

4. Results and discussion


4.1. Stiffness progression and damage evolution
As damage accumulates, the load in the joint re-distributes
between bolts and the joint stiffness changes. In this section, the
change in joint stiffness (slope of the loaddeection curve) with
damage evolution is studied. Fig. 5(a)(f) compares the FE solution
for stiffness vs. load with the experimental data in [22]. Good
agreement is provided for all clearance combinations by the model,
providing reasonable magnitude and trend of the joint stiffness
degradation. Initial oscillations in load occur due to intermittent

Y. Zhou et al. / Composite Structures 116 (2014) 441452

sticking and slipping, and growth in contact area, as the bolts settle
into the holes. Notice this initial oscillatory region is largest for
the joint with the largest total clearance (C3_C3_C1) as expected.
The highest level the stiffness reaches before it drops due to the
damage onset also matches the experiments quite well, as does
the load at which damage onset occurs (e.g. around 50 kN for the
C1_C1_C1 joint and 40 kN for the C3_C3_C1 joint).
After the bolts settle into the holes, all the stiffness curves have
three main stages: a relatively constant stiffness phase, a transition
phase where the stiffness oscillates again, and then a sharp decline
at the end. The stiffness oscillates in the transition phase (between
around 40 kN and 60 kN) because of the re-distribution of load
from the hole(s) that fail rst in bearing (the neat-t holes) to

447

the larger clearance holes which had initially taken much lower
load.
The process is illustrated in Fig. 6, which shows the bre damage level in the C1_C1_C4 joint at loads of 48 kN, 54 kN and 58 kN.
From Fig. 5(d) we see these loads correspond respectively to just
before the rst major drop in stiffness (rst bearing failure), just
at the point when the stiffness has dropped to a low value, and just
after it has rebounded somewhat to a higher value. At 48 kN
(Fig. 6(a)) we observe bre compressive (i.e. bearing) damage at
the two neat-t holes (holes 1 and 2), but none at the loose t hole
(hole 3). This is to be expected since the neat t holes will initially
take almost all the load, with hole 3 assuming only a small portion
of it (see [22] for unequivocal experimental evidence of this). At

Fig. 5. Joint stiffness comparisons between experiments and FE models: (a) C1_C1_C1; (b) C2_C1_C1; (c) C4_C1_C1; (d) C1_C1_C4; (e) C1_C3_C1; (f) C3_C3_C1.

448

Y. Zhou et al. / Composite Structures 116 (2014) 441452

Fig. 6. Fibre damage evolution of C1_C1_C4 joint: (a) at 48 kN; (b) at 54 kN; (c) at 58 kN.

Fig. 7. Failure modes of quasi-static tests: (a) C1_C1_C1, C1_C1_C4 and C1_C3_C1: net tension of splice (outer) plates at H3; (b) C2_C1_C1, C4_C1_C1 and C3_C3_C1: net
tension of skin (inner) plate at H1.

Y. Zhou et al. / Composite Structures 116 (2014) 441452

449

Fig. 8. Fibre damage at ultimate failure for C1_C1_C1: (a) splice plate and skin plate; (b) hole 1 of skin plate.

Fig. 9. Fibre damage at ultimate failure for C1_C3_C1: (a) splice plate and skin plate; (b) hole 1 of skin plate.

54 kN (Fig. 6(b)), the bearing damage at holes 1 and 2 has progressed sharply causing the sharp drop in stiffness of the joint,
but hole 3 is still not taking enough load to cause any bearing damage at that location. In fact, the take up of load by hole 3 is being
slightly delayed due to the onset of some tensile bre damage at
the net-tension section of hole 3, which allows the hole to stretch
(this occurs due to high by-pass loads as discussed below). Finally
at 58 kN (Fig. 6(c)), we nd signicant bearing damage at hole 3,
indicating all three holes are now fully sharing the burden of the
load, which leads to the temporary increase in stiffness of the joint
in Fig. 5(d) due to the increase in bearing area. This transition from
an unequal sharing of the load up to rst bearing failure, to an
equal sharing of the load by the time ultimate failure occurs,
explains why clearance affects 2% offset bearing strength but does
not signicantly affect ultimate strength.

4.2. Ultimate failure mode


As noted above, all the joints in the experimental study failed
ultimately in net tension. However they did not all fail at the same
location. To understand this behaviour, we need to consider the
bypass stress at each hole. Consider Fig. 1 again. Letting w = the
width of the plates, t = the thickness of the splice plate, and
2t = the thickness of the skin plate, then if P1, P2 and P3 are the
loads taken by bolts 1, 2 and 3 respectively, the gross-section
bypass stress in the splice plate is highest at hole 3 and equals
(P1 + P2)/(2wt), while in the skin it is highest at hole 1 and is
(P2 + P3)/(2wt). In conventional CBJ design, it is assumed that
clearances are the same at all holes, that the load distribution does
not change during loading, and that in this three-bolt conguration, P1 = P3, and P2 < P1. Thus conventional theory would say a

450

Y. Zhou et al. / Composite Structures 116 (2014) 441452

net tension failure is equally likely at hole 3 in the splice plates or


hole 1 in the skin plate since the bypass stresses are the same in
both locations, and which happens rst depends on random factors
like internal defects or slight manufacturing variations. However,
in our study the clearances are not necessarily all the same. Since
P1 will be smaller than P2 and P3 when a clearance exists in hole
1, the bypass stress in the splice plate at hole 3 is less than the
bypass stress in the skin at hole 1, so we would expect a net tension failure in the skin at hole 1 in the C2_C1_C1, C4_C1_C1, and
C3_C3_C1 cases. This is precisely what happened (see Fig. 7(a)).
In the C1_C1_C1 and C1_C3_C1 cases either failure location should
be equally likely, while in the C1_C1_C4 case, the failure should be
in the splice plate at hole 3; in fact all these cases failed in the
splice plate at hole 3 (see g. 7(b)).
This simple theory considering loading in the elastic region is
thus capable of qualitatively explaining the likely effects of clearance on ultimate failure mode. However, it could not make quantitative predictions such as what level of clearance is needed in

hole 1 to guarantee a net tension failure in the skin at hole 1. This


is because, as noted in the previous section, once the initial bearing
failures start to occur, the load shifts around in a very complex way
as the damage at the various holes progresses. Thus the damage
model provides a much more detailed analysis, and we now
address some additional insights it can provide and check them
against experimental ndings.
Fibre damage at ultimate failure of four cases (C1_C1_C1,
C1_C3_C1, C2_C1_C1, and C3_C3_C1) is shown in Figs. 811. The
upper plate in the gures is the splice plate and the lower one is
the skin plate (only the top half of the skin plate is modelled due
to symmetry). Note that damage up to ultimate failure was not
possible to predict in our previous study using PDA, which could
not obtain a converged solution this far into the process [16].
As can be seen from Figs. 8(a) and 9(a), for the C1_C1_C1 and
C1_C3_C1 joints, compressive bre damage (bearing failure if there
is enough of it) occurs at all hole surfaces, but the ultimate cause
of failure in both joints is tensile bre damage across the entire

Fig. 10. Fibre damage at ultimate failure for C2_C1_C1: (a) splice plate and skin plate; (b) hole 1 of skin plate.

Fig. 11. Fibre damage at ultimate failure for C3_C3_C1: (a) splice plate and skin plate; (b) hole 1 of skin plate.

Y. Zhou et al. / Composite Structures 116 (2014) 441452

451

Fig. 12. Fibre damage in 45/0/45/90 plies at ultimate failure: (a) hole 3 of splice plate for C1_C1_C1; (b) hole 1 of skin plate for C3_C3_C1. Shown also is the length of the
region in which bre damage occurs at the specimen edge (mm); this length is the original length of the damaged elements.

cross-section of the splice plate at hole 3, i.e. net-tension failure at


this location, as in the experiments. However, notice in Figs. 8(b)
and 9(b) that substantial bre damage also exists across the net
tension section of hole 1 in the skin plate, i.e. there was almost
an equal likelihood that failure would occur at hole 3 in the splice
plate or hole 1 in the skin plate, and which failure occurred rst
depended on very small differences in the way the damage progressed at each hole, which is in line with the discussion above.
In contrast for joints with a non-zero clearance in hole 1, viz.
C2_C1_C1 and C3_C3_C1, shown in Figs. 10 and 11 respectively,
net tension failure occurs at hole 1 in the skin plate, as shown in
Figs. 10(b) and 11(b). This is in agreement with the experiments
and the simple theory. Once again bearing damage occurs at all
hole surfaces prior to this, and tensile bre damage also exists at
hole 3 of the splice plate. However, this time the tensile bre damage is much more substantial at hole 1 of the skin plate than at hole
3 of the splice plate, so there is no uncertainty about which failure
will occur rst, which is also in line with the discussion above.
The above shows that the model is reproducing the main failure
mechanisms in the joints in a correct way, although it could be
argued that the simple bearingbypass analysis could predict the
same global, if not detailed, behaviour. Something the simple
methods could not give insight into, however, is the issue raised
at the end of Section 3.3, viz. why the C3_C3_C1 joint exhibits a
much higher joint deformation (and hence more energy absorption) at failure than the C1_C1_C1 joint. The damage model gives
an interesting insight into this. Comparing Fig. 11 with Fig. 8, we
see that the elongation of hole 1 in the skin plate of the
C3_C3_C1 joint is much higher than that of hole 3 in the splice
plate of the C1_C1_C1 joint. In addition, the amount of material
that exhibits bre damage is much higher in the C3_C3_C1 joint.
Thus, the C3_C3_C1 joint deforms much more and absorbs more
energy than the C1_C1_C1 joint. Why does this happen? Well, it
is well known [14] that increased clearance shifts the peak tangential tensile stress at the hole away from the net tension plane and

towards the bearing plane (i.e. the loading direction) by as much as


20. This shifts the tensile load somewhat from the 0 plies to the
45 and 45 plies, thus spreading the load and leading to more bre
damage in the off-axis plies, which can be seen clearly in Fig. 12.
Thus, a larger region of damageable material is created so the hole
can stay intact longer, and more energy is ultimately absorbed.
5. Conclusions
A detailed three-dimensional FE model of double-lap, multibolt composite joints, has been developed. Material damage and
failure has been modelled using a CDM model, incorporating a
new method of determining bre initiation strain, which takes into
account the interaction of tensile and shear stresses on the bre.
The effects of variable clearance on stiffness, damage evolution,
and failure mode have been studied numerically.
In single and four-element tests, the CDM model has achieved
stressstrain responses in the longitudinal, transverse and shear
directions which match well with experimental materials tests.
The shear response exhibited an obvious non-linear stressstrain
behaviour, which actually signicantly affects the failure of the
bres, due to the new bre damage criterion. When used in the
multi-bolt joint models, the CDM model provided stable and sensible results all the way to nal failure, which is a considerable
improvement over previous PDA models.
Correlation of the non-linear behaviour in the loaddeection
curves with the bre damage evolution showed that the model
was capturing all the important damage and failure phenomena
in the joint. The effects of variable clearance, which is not generally
incorporated into standard joint design, could be predicted with a
high degree of delity all the way to nal failure. This suggests it
could be used to predict the effects of many other joint parameters,
and is therefore a useful design tool. The model could also give signicant additional insight into the detailed progression of damage,
which can aid design.

452

Y. Zhou et al. / Composite Structures 116 (2014) 441452

Future work includes combining a delamination model with the


in-plane damage model presented here, as done by our group for
impact on panels in [11], and the addition of a nonlocal regularisation scheme to reduce mesh dependency, such as the one used by
our group [9,10] for modelling open hole specimens and bre
metal laminate joints. Both these developments are currently computationally prohibitive for multi-bolt joints, which is why they
were not included here, but this will change in the near future,
as computing power continues to grow.
Acknowledgements
The authors wish to acknowledge Irish Centre of High-End
Computing (ICHEC) for the provision of computational facilities.
The nancial support from the China Scholarship Council (CSC),
and the provision of research facilities from the Materials and Surface Science Institute (MSSI) at the University of Limerick are also
gratefully appreciated.
References
[1] Katnam KB, Da Silva LFM, Young TM. Bonded repair of composite aircraft
structures: a review of scientic challenges and opportunities. Prog Aerosp Sci
2013;61:2642.
[2] Starikov R, Schon J. Quasi-static behaviour of composite joints with
protruding-head bolts. Compos Struct 2001;51:41125.
[3] Starikov R, Schon J. Quasi-static behaviour of composite joints with
countersunk composite and metal fasteners. Compos B Eng 2001;32:4019.
[4] Ireman T. Three-dimensional stress analysis of bolted single-lap composite
joints. Compos Struct 1998;43:195216.
[5] Perugini P, Riccio A, Scaramuzzino F. Three-dimensional progressive damage
analysis of composite joints. Proceedings of the eighth international conference
on civil and structural engineering computing. Stirling, Scotland: Civil-Comp
Press; 2001.
[6] Ladeveze P, Ledantec E. Damage modeling of the elementary ply for laminated
composites. Compos Sci Technol 1992;43:25767.
[7] OHiggins RM, McCarthy CT, McCarthy MA. Effects of shear-transverse
coupling and plasticity in the formulation of an elementary ply composites
damage model, Part II: material characterisation. Strain 2012;48:5967.
[8] OHiggins RM, McCarthy CT, McCarthy MA. Effects of shear-transverse
coupling and plasticity in the formulation of an elementary ply composites
damage model, Part I: model formulation and validation. Strain 2012;48:
4958.

[9] Frizzell RM, McCarthy CT, McCarthy MA. Predicting the effects of geometry on
the behaviour of bre metal laminate joints. Compos Struct 2011;93:187789.
[10] Frizzell RM, McCarthy CT, McCarthy MA. Simulating damage and delamination
in bre metal laminate joints using a three-dimensional damage model with
cohesive elements and damage regularisation. Compos Sci Technol 2011;71:
122535.
[11] Nezhad HY, McCarthy CT. Deliverable: Low-Velocity, High-energy impacts and
compression after impact. In: MAAXIMUS-WP34-ULIM-DEL-D341-low
velocity/high-energy impact-V01. 2012.
[12] Ireman T, Ranvik T, Eriksson I. On damage development in mechanically
fastened composite laminates. Compos Struct 2000;49:15171.
[13] Padhi GS, McCarthy MA, McCarthy CT. BOLJAT: a tool for designing composite
bolted joints using three-dimensional nite element analysis. Compos A Appl
Sci Manuf 2002;33:157384.
[14] McCarthy CT, McCarthy MA. Three-dimensional nite element analysis of
single-bolt, single-lap composite bolted joints: Part II effects of bolt-hole
clearance. Compos Struct 2005;71:15975.
[15] McCarthy MA. An experimental study of bolt-hole clearance effects in singlelap, multibolt composite joints. J Compos Mater 2005;39:799825.
[16] McCarthy CT, McCarthy MA, Lawlor VP. Progressive damage analysis of multibolt composite joints with variable bolt-hole clearances. Compos B Eng
2005;36:290305.
[17] McCarthy MA, Lawlor VP, Stanley WF, McCarthy CT. Bolt-hole clearance effects
and strength criteria in single-bolt, single-lap, composite bolted joints.
Compos Sci Technol 2002;62:141531.
[19] ASTM standard D 5961M96, Standard test method for bearing response of
polymer matrix composite laminates. 1996.
[20] Egan B, McCarthy CT, McCarthy MA, Frizzell RM. Stress analysis of single-bolt,
single-lap, countersunk composite joints with variable bolt-hole clearance.
Compos Struct 2012;94:103851.
[21] Egan B, McCarthy CT, McCarthy MA, Gray PJ, Frizzell RM. Modelling a singlebolt countersunk composite joint using implicit and explicit nite element
analysis. Comput Mater Sci 2012;64:2038.
[22] Lawlor VP, McCarthy MA, Stanley WF. An experimental study of bolt-hole
clearance effects in double-lap, multi-bolt composite joints. Compos Struct
2005;71:17690.
[23] Abaqus-Inc. ABAQUS v6.10. User Manual, Dassault Systemes Simulia Corp.
2010.
[24] McCarthy CT. Experiences with modeling friction in composite bolted joints. J
Compos Mater 2005;39:1881908.
[25] OHiggins RM. An experimental and numerical study of damage initiation and
growth in high strength glass and carbon bre-reinforced composite materials
[PhD]. University of Limerick, College of Engineering; 2007.
[26] Hashin Z. Failure criteria for unidirectional ber composites. J Appl Mech Trans
ASME 1980;47:32934.
[27] Hashin Z. Fatigue failure criteria for unidirectional ber composites. J Appl
Mech Trans ASME 1981;48:84652.
[28] McCarthy CT, OHiggins RM, Frizzell RM. A cubic spline implementation of
non-linear shear behaviour in three-dimensional progressive damage models
for composite laminates. Compos Struct 2010;92:17381.