Composite Structures 133 (2015) 1198–1208

Contents lists available at ScienceDirect

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

Finite element analysis of catastrophic failure of dynamically-loaded

countersunk composite fuselage joints
B. Egan a, M.A. McCarthy a, R.M. Frizzell b, C.T. McCarthy a,⇑
a
Irish Centre for Composites Research (IComp), Materials and Surface Science Institute (MSSI), Department of Mechanical, Aeronautical and Biomedical Engineering, University
of Limerick, Ireland
b
Bell Labs, Efficient Energy Transfer (get) Dept., Alcatel-Lucent, Blanchardstown, Dublin 15, Ireland1

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

Article history: Accurate models of dynamic structural failure are important for crashworthiness studies. To date, catas-
Available online 8 July 2015 trophic failure of dynamically-loaded composite bolted joints has been studied using global or stacked
shell element models. In this paper, high-fidelity (three-dimensional solid) explicit FE models are
Keywords: used to simulate catastrophic failure of countersunk composite fuselage joints. While current
Finite element analysis state-of-the-art 3D modelling approaches focus almost exclusively on the prediction of composite
Bolted joints damage, this study also investigates the treatment of fastener damage. Fastener fracture is a common
Fastener failure
catastrophic joint failure mode, particularly in joints designed to initially fail in bearing. A Johnson–
Carbon–epoxy
Titanium fasteners
Cook material model and cohesive elements were used to predict plasticity, damage and fracture of
Explicit dynamics the titanium (Ti–6Al–4V) fastener. Although a model calibration was required, due to the complex inter-
action of model parameters, numerical results demonstrate key trends of experiments and provide a
starting point for the development of more predictive approaches for simulating fastener failure.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction a growing need to ensure that crashworthiness is not compromised

by lightweighting measures.
In recent years, significant weight reductions have been Aircraft skin structure plays a crucial role in absorbing energy in
achieved in commercial aircraft design. The newly developed a crash situation. The composite fuselage skin structure of the
Airbus A350 consists of 53% composite material, resulting in CO2 Airbus A350 features bolted joints, where countersunk fasteners
emissions per passenger which are 25% lower than the current air- are used for aerodynamic reasons. As damage tends to localise at
craft in its class [1]. Such remarkable improvements in structural stress concentrations, joints have a significant influence on the
efficiency have been greatly aided by developments in finite ele- level of energy absorbed in a crash situation. In fact, in a global
ment (FE) modelling, especially in the area of composite damage aircraft crash simulation, if the failure of fastened joints is not
modelling. Following publication of results from the World Wide modelled accurately, global kinematics and the predicted
Failure Exercise (WWFE) outlined in [2], research involving US overall crash behaviour can be completely incorrect [8]. Thus,
and European researchers led to the development of ‘‘LaRC’’ (Lang- accurate simulation of catastrophic failure of dynamically loaded
ley Research Centre) failure criteria [3–5]. Crucially, LaRC criteria countersunk composite joints is an important research goal for
are physically-based and incorporate the Puck criteria of [6,7], the aircraft industry.
which had performed well in the WWFE. Coupled with greater Dynamic structural analyses are normally carried out using an
availability of computational power, more widespread use of explicit solver, which offers important benefits for modelling
physically-based composite damage models has considerably catastrophic failure of composite bolted joints, including a robust
improved predictions of failure onset. However, as the drive handling of contact and material softening [9]. For these reasons,
towards even lighter, more efficient structures continues, there is explicit FEA is set to play an increasingly important role in simulat-
ing failure of complex aircraft structures [10]. Pearce et al. [11]
used the PAM-CRASH explicit solver to simulate failure of
⇑ Corresponding author. Tel.: +353 61 234334; fax: +353 61 202944. multi-bolt, composite structures under quasi-static and dynamic
E-mail address: [email protected] (C.T. McCarthy). loading. A highly efficient global modelling approach was devel-
1
Present address. oped, in which laminates were modelled using shell elements,

http://dx.doi.org/10.1016/j.compstruct.2015.07.004
0263-8223/Ó 2015 Elsevier Ltd. All rights reserved.
 B. Egan et al. / Composite Structures 133 (2015) 1198–1208 1199

and joint behaviour was represented by one-dimensional ‘‘PLINK’’ (ABS0873), and steel nuts (ASNA2536). Bolt-hole clearance was
connector elements, which were calibrated using single-fastener maintained at less than 57 lm, which represents a H8 tolerance
joint tests. The approach was very robust and good agreement of ISO 286-2 (1988) and is considered a realistic upper tolerance
was obtained with experiment. More recently, Pearce et al. [12] for aerospace joints. Two layups were studied. The 2.125 mm thick,
used a stacked shell approach in PAM-CRASH to predict bearing 17-ply ‘‘C’’-layup ([45/90/+45/0/0/0/+45/0/90/0/45/0/0/0/+45/
failure in a dynamically-loaded, single-bolt countersunk joint, 90/45]) represented a fuselage skin panel capable of sustaining
which included more detail than the global models in [11], but low axial loads. The thicker, 3.125 mm, 25-ply, ‘‘E’’-layup ([45/9
were more efficient than three-dimensional (3D) models. Failure 0/+45/0/0/0/45/+45/0/0/0/45/90/+45/0/0/0/+45/45/0/0/0/+45
of the composite fabric material was predicted using a Tsai-Wu /90/45]) was characteristic of fuselage regions experiencing
failure surface, and damage variable growth was controlled by higher axial loads.
strain-driven functions. A cohesive zone model was used to simu- Joint geometries are shown in Fig. 1, with relevant dimensions
late delamination. The model showed good agreement with exper- given in Table 1. The tests which featured a test velocity (TV) of
iment in the initial loading and failure onset regions, but 10 m/s are the main focus of this paper so unless stated otherwise,
significant differences arose after extensive hole crushing. It was a 10 m/s loading rate is considered. Regarding the abbreviations
proposed that a better treatment of debris in the model, possibly used, ‘‘CLam_B1’’ for example, refers to a joint with ‘‘C’’-layup lam-
using smooth particle hydrodynamics (SPH), could improve inates, and one bolt (‘‘B1’’). A test repeat number is also included
post-failure predictions. The same stack-shell approach was used (e.g. ‘‘CLam_B1_1’’ indicates the first instance of a number of
to simulate pull-through loading in [13]. Here, correlation with 10 m/s test repeats).
experiment improved when additional link elements were In [14], the type of final (i.e. catastrophic) failure had a substan-
included to account for the inhibiting effect on delamination tial effect on the amount of energy absorbed by the joint, which
of compressive through-thickness stress arising from fastener contributes significantly to overall structural crashworthiness.
pre-load. Two types of final failure mode were observed in [14] – fastener
In our recent quasi-static modelling study of [9], bearing failure fracture at a thread (denoted ‘‘FF’’ failure) and fastener pull-through
of countersunk composite joints was simulated using 3D explicit (denoted ‘‘FP’’ failure). As shown in Table 2, all C-layup joints
FE models. A physically-based damage model was used to predict (CLam_B1 and CLam_B3) exhibited FP as the final failure mode.
composite failure and very good correlation was obtained with For the ELam_B1 joints, which feature thicker laminates, three of
experiment. Quasi-static numerical studies are generally used in the six specimens exhibited FF failure, while the other three exhib-
sizing aircraft joints and do not feature prediction of fastener ited FP failure. An FF failure of an E-layup, single-bolt joint is shown
failure, as joint sizing is based on the initial rather than ultimate in Fig. 2(a), while an FP failure of the same joint type is shown in
failure loads. However, studies involving dynamic loading are used Fig. 2(b). It can be seen that the FF failure is associated with far less
to investigate crashworthiness behaviour, so the simulation of bearing damage, and hence energy absorption, than the FP failure,
complete joint separation is important. Prediction is needed not because the fastener failure at a thread interrupts the crushing
just of damage initiation, progression and preliminary failure, e.g. process.
bearing failure, but also ultimate failure of the joint, which is often A crack in the countersunk head can be seen in Fig. 2(a), appear-
sudden or ‘‘catastrophic’’ (e.g. net-tension failure, bolt failure or ing on the side where the bolt rotates down into the laminate
bolt pull-through). (i.e. the side which is not reacting the in-plane bearing load). In
Although catastrophic failure of dynamically-loaded composite the tests of [14,16,17], this type of crack was evident in the thicker
bolted joints has been analysed using global and stacked shell (E-layup) joints, but not in the thinner C-layup joints. This was
models, high fidelity 3D models have not yet been applied to this attributed in [14] to the higher moment being reacted by the
problem. Furthermore, little or no attention has been given to fastener (due to the more eccentric load path) in thick, single-lap
the prediction of fastener fracture, with most numerical studies joints. A further observation in high speed experiments was the
focusing solely on composite damage, or including only a simple removal of the cadmium coating from the nut (see ring markings
plasticity law to deal with nonlinearity in the response of the fas- on non-countersunk (NCSK) laminates in Fig. 2(a) and (b)), which
tener material. The aim of the current paper is to use detailed 3D indicated significant normal loads arising between the nut and
solid FE modelling to simulate catastrophic failure of laminate in this region.
dynamically-loaded composite bolted aircraft joints for the first The filtered load–deflection curves from the 10 m/s single-bolt
time. Far greater attention is given to the prediction of fastener joint tests are shown in Fig. 3. The CLam_B1 joints, which all failed
damage than in previous studies, with a Johnson–Cook failure by FP (fastener pull-through), exhibit a gradual softening after the
model assigned to the bulk fastener material. A novel approach ultimate load is reached. The thicker ELam_B1 joints exhibit nearly
of using cohesive elements at critical failure locations allowed rep- twice the ultimate load of the CLam_B1 joints. Repeats 1 and 3 of
resentative fractures to be captured where the bulk metal failure the ELam_B1 joints, which fail by FF (fastener fracture at a thread),
model was not quite so predictive. This type of high-fidelity FE exhibit a complete drop-off in load after ultimate load is attained.
modelling of dynamically-loaded joints has significant potential Even for the one E-layup joint that did not fail in this catastrophic
to further improve predictive capabilities and provide valuable manner (i.e. repeat 2), the load initially drops much more sharply
inputs to global crashworthiness models. than in the thinner C-layup joints, and then, at about 2.5 mm joint
deflection, the response settles to a more gradual unloading slope,
similar to that of the C-layup joints. The initial sharp drop is attrib-
2. Experimental results uted to the aforementioned countersunk head fracture.

High speed tensile tests of countersunk composite bolted joints,

representative of fuselage skin joints, were carried out in [14], as 3. Finite element models
part of a crashworthiness study in the MAAXIMUS project [15].
The joints were designed to promote bearing failure (w/d = 6.25) 3.1. Model details and objectives
Ò
and were manufactured from HexPly M21E/IMA carbon–epoxy
pre-preg. Joints were fastened with 4.8 mm diameter, aerospace This numerical investigation focuses on 10 m/s loading of
grade titanium (Ti–6Al–4V) bolts, having 130° countersunk heads single-bolt and three-bolt joints. Three models (CLam_B1,
1200 B. Egan et al. / Composite Structures 133 (2015) 1198–1208

Fig. 1. (a) Single-bolt joint geometry, and (b) multi-bolt joint geometry. DIC speckle refers to the high speed tests. See Table 1 for values of e, w, t, g and L. All dimensions are
in mm.

Table 1
Codes describing layups and joints geometries. C-layup is [45/90/+45/0/0/0/+45/0/
for the simulation of composite failure. Maximum stress criteria
90/0/45/0/0/0/+45/90/45], E-lay-up is [45/90/+45/0/0/0/45/+45/0/0/0/45/90/
+45/0/0/0/+45/45/0/0/0/+45/90/45]. All dimension are in mm.
were used to predict tensile and compressive fibre failures, while
the crack band model [20] was used to mitigate mesh sensitivity.
Code Configuration/ Refer g L w t e p
Although delamination was not modelled explicitly, the use of
Layup to:
Puck criteria allowed for the prediction of ply failures which mim-
CLam_B1 C-Layup (CLam) Fig. 1(a) 30 2.125 15 – icked delaminations [21]. Furthermore, the Mohr–Coulomb type
Single-Bolt (B1)
Joint
intralaminar failure criteria ensure that through-thickness com-
ELam_B1 E-Layup (ELam) Fig. 1(a) 50 160 30 3.125 15 – pressive forces increase resistance to laminate splitting, which is
Single-Bolt (B1) physically realistic (and was found to be important in [13]). For
Joint further details on the damage model and the joint modelling
CLam_B3 C-Layup (CLam) Fig. 1(b) 50 190 30 2.125 22 22
approach, see [9]. Some notable differences between the
Three-Bolt (B3)
Joint quasi-static modelling of [9] and these analyses are:

 Joint end displacements were applied using a 10 m/s velocity

Table 2 boundary condition instead of the smooth step displacement
Final (catastrophic) failure mode (FF = Fastener fracture at a thread, FP = Fastener used in [9].
pull-through, TV5_1 means test velocity = 5 m/s, instance 1).  A compressive fibre fracture energy (Gfc) of 143 kJ/m2 was used
TV5_1 TV5_2 TV5_3 TV10_1 TV10_2 TV10_3
in all models. Based on results obtained in [9], this was consid-
ered suitable for predicting bearing damage development in
CLam_B1 FP FP FP FP FP FP
ELam_B1 FP FF FP FF FP FF
this toughened carbon–epoxy material.
CLam_B3 3  FP 3  FP 3  FP 3  FP 3  FP 3  FP  Near-hole meshes featured 35 rings of elements (having one ply
thickness in the radial direction), as shown in Fig. 4. This
highly-refined region of mesh was required to adequately sim-
ELam_B1 and CLam_B3) were developed in Abaqus/Explicit, with ulate the extensive hole crushing occurring prior to separation.
the bolt-hole contact region of the CLam_B1 model shown in All composite elements which could potentially come in contact
Fig. 4. Images such as those in Figs. 4 and 5 have half of the model with the fastener or nut were included in interior surface con-
removed to give a better view of the bolt-hole contact region. A tact definitions.
bolt pre-load of 2.28 kN was introduced by applying a thermal  A material model incorporating plasticity and damage was
expansion to the nut, as outlined in [18]. Contact surfaces were applied to the titanium fastener regions.
defined at part level and could be included in multiple  Cohesive elements were included in two locations, as shown in
assembly-level interactions. Prior to any calibration, the friction Fig. 4, to allow fastener fractures to be simulated at the fracture
coefficients used in [19] were utilised, i.e. 0.7 for locations of the experiments (see [14]). Cohesive zones were
laminate-laminate interactions, 0.3 for nut-laminate interactions, defined at the thread location (to simulate FF fracture) and in
and 0.1 for bolt-laminate interactions. a ring inside the countersunk head (to simulate head fracture
The simulations featured a physically-based damage model, which facilitates FP joint failure). The cohesive zones could be
implemented in VUMAT, with non-linear in-plane shear behaviour made inactive by using very large strengths (i.e. 10,000 MPa)
 B. Egan et al. / Composite Structures 133 (2015) 1198–1208 1201

Fig. 2. Failed single-bolt E-layup joints – (a) specimen: ELam_B1_TV10_1, final failure mode: fastener failure (FF), and (b) specimen: ELam_B1_TV5_1, final failure mode:
fastener pull-through (FP) (CSK: countersunk laminate, NCSK: non-countersunk laminate).

The Johnson–Cook [23] material model which is available in

Abaqus/Explicit was used to simulate the mechanical response of
the titanium fastener. This material model incorporates
rate-dependent plasticity and a progressive damage formulation,
although effects of temperature on the material behaviour have
been omitted for these analyses. The yield stress (r
 ) which governs
rate-dependent plasticity is defined as:
  
n e_ pl
r ¼ ½A þ Bðepl Þ  1 þ C ln _ ð1Þ
e0
where epl is the equivalent plastic strain and A, B, C and n are mate-
rial parameters. Relevant parameters for Ti–6Al–4V, shown in
Table 3, were obtained from an FAA study conducted by Lesuer
[23]. The reference strain rate ðe_ 0 Þ was taken to be 1.0.
The Johnson–Cook criterion, defining the equivalent plastic
strain at damage onset, is given by:
  pl 
epl
e_
Fig. 3. Experimental response curves from the 10 m/s tests on CLam_B1 and D ¼ ½d1 þ d 2 expðd3 Þ 1 þ d4 ln ð2Þ
ELam_B1 joints. e_ 0
where d1–d4 are parameters, given in Table 3.
to enforce elastic behaviour across a particular cohesive zone, or
In addition to the mechanical behaviour of the fastener, a large
activated at the FF or FP zone (or both) simply by lowering the
number of model parameters can be expected to vary with loading
strengths at the relevant zone.
rate. For example, most high-rate testing studies on carbon–epoxy
 Due to the inclusion of the cohesive elements, selective mass
suggest enhancement of matrix-dominated properties with load-
scaling was applied to retain the 5  109 second stable time
ing rate [24–27]. The mechanical response of the joint materials
increment of simulations without cohesive elements.
may also be affected by temperature increases caused by the fric-
 Rather than pure stiffness hourglass control, which is suited to
tion forces acting in the hole crushing region [28]. Such effects are
quasi-static simulations [22], a combined stiffness-viscous con-
not so easily accounted for in these FE models. Thus, to adequately
trol, with a weighting factor of 0.5, was utilised in the dynamic
capture the complex failure process, a model calibration procedure
models.

Fig. 4. (a) Meshing strategy and material property assignments in the bolt-hole region (shown for CLam_B1 model) and (b) cohesive element locations.
1202 B. Egan et al. / Composite Structures 133 (2015) 1198–1208

Fig. 5. Model deformations observed in calibration study, at 5 mm end displacement.

Table 3 Two other features, which appeared in Mod_1, but not in the
Parameters for the Ti–6AL–4V Johnson–Cook material model [23]. tests, were (1) penetration of highly distorted elements into the
A (MPa) B (MPa) n C d1 d2 d3 d4 bolt and (2) the development of significant NCSK hole crushing
1098 1092 0.93 0.014 0.09 0.25 0.50 0.014
rather than CSK hole crushing. Neither of these issues arose in
the quasi-static models of [9] (which were required to predict far
less hole crushing). To rectify this, modifications were made to
the element deletion criterion:
is considered justified. In this, certain parameters are varied to
tþDt tþDt tþDt
obtain reasonable correlation with experiment for a baseline case f erosion 1 ¼ maxðabsðc12 Þ; absðc23 Þ; absðc31 ÞÞ  X; f erosion 1 P0
(i.e. CLam_B1_TV10). Using the calibrated parameters, other cases ð3Þ
are modelled to see if experimental trends can be captured.
tþDt tþDt tþDt
where c12 ; c23 and c31 are the shear strains at the end of the cur-
3.2. Model calibration rent time increment and X is the shear strain threshold.
The modifications made in Mod_2 were:
CLam_B1 joints failed via fastener pull-through (FP), so a key
goal of the calibration was to achieve this failure in simulations. 1. Penetration of distorted elements into the bolt was prevented
It was also necessary to eliminate some non-characteristic by initially reducing the element deletion criterion from the
behaviour which arose in trial simulations and improve model pre- very high base model value (of X = 2.0) to X = 1.0.
dictions for the baseline case. Initially, both cohesive zones were 2. The shift in hole crushing from the CSK hole to the NCSK hole,
deactivated, to investigate if pull-through could be simulated after the development of extensive bearing damage, was attrib-
solely using the built-in Johnson–Cook plasticity and metal dam- uted to insufficient element deletion in the countersunk area. A
age model of Abaqus/Explicit. The complete model calibration very conservative value of 1.0 had been chosen for the element
applied to the CLam_B1 model is summarised in Table 4 and deletion criterion, in order to leave the elements in the analysis
outlined here. for as long as possible. However, as elements in the countersunk
The first simulations showed ‘‘bursting’’ to occur, i.e. the nut region are initially highly skewed due to the high (130°) head
burst through the non-countersunk (NCSK) laminate at the side angle, in order to be deleted, they must become far more
of the hole not in contact with the bolt, as shown in ‘‘Mod_0’’ in distorted than those located elsewhere. Lowering the shear
Fig. 5(a). This did not occur in the experiments, but our simulations strain threshold (X) for the material in the countersunk region
of a countersunk joint in [29] showed this region of the NCSK lam- (identified as ‘‘VUMAT_CSK’’ in Fig. 4), from 1.0 to 0.4, allowed
inate to be under severe compressive through-thickness stress due these elements to be more readily deleted, permitting CSK hole
to the bolt rotation and consequent pressing of the nut into the crushing to dominate, as had been seen in experiments.
laminate. In addition, the removal of the cadmium nut coating
(see Section 2) via friction is also indicative of high normal forces The next adjustment was to the friction parameters (see Mod_3
between the nut and NCSK laminate. Thus, even though bursting action of Table 4), but this still did not yield pull-through failures.
did not occur in the tests, it is likely that it could have done with Instead stable composite crushing dominated, with no mechanism
minor changes in test parameters, and therefore prediction of its for catastrophic joint failure (see Fig. 5)). In [30], a range of John-
occurrence is not unreasonable. To prevent bursting in the model, son–Cook material parameters are presented for Ti–6Al–4V, indi-
in ‘‘Mod_1’’ (simulation with first modification), we assigned elas- cating a degree of uncertainty associated with these inputs. With
tic properties to surface layers at this side of the hole in the NCSK this in mind, metal plasticity and damage parameters were varied
laminate, as shown in Fig. 4. significantly to promote fastener damage. However, in simulations
 B. Egan et al. / Composite Structures 133 (2015) 1198–1208 1203

Table 4
Sequence of actions in calibration of the CLam_B1 joint model.

where the fastener was more damageable, the damage tended to unfiltered numerical response cannot be meaningfully interpreted,
localise in the shank just below the countersunk section and so the filtered response is characteristic of the load–deflection curves
the fastener fracture was different to that of experiment (where obtained in the high speed tests of [14]. Due to the significant
the angled part of head had been torn away from the shank). It effects of machine slack on the stroke readings of high speed tests,
was thus concluded that catastrophic failure could not be pre- digital image correlation was used to obtain displacement mea-
dicted solely by adjusting the Johnson–Cook bolt material param- sures in [14], and these are compared with nodal displacements
eters. Only by activating the cohesive elements, with tuned at the same locations in the model. Force readings were taken from
strength values, (see Mod_4 of Table 4/Fig. 5), could realistic sim- the fixed ends of the models, as this was the side where the load
ulation of fastener pull-through (FP) be achieved. cell was positioned in the tests. In Fig. 6(b), the filtered
force-extension response of the model is plotted alongside the
4. Results experimental results (which were available to extensions of
5 mm). Remarkably good agreement is seen between experiment
4.1. Load–deflection and damage in the calibrated model versus and simulation in the elastic loading phase, for which the model
experiment response is largely independent of calibrated parameters. The sub-
sequent failure onset and crushing loads in the model (which was
Reaction forces obtained from dynamic models generally fea- calibrated according to Table 4) also captured the experimental
ture large-amplitude, high-frequency content, and require filtering response well over the 5 mm extension for which experimental
to remove frequencies which are not of structural significance. Fre- data was available. A peak load of 7.3 kN was predicted, which is
quencies of structural significance are typically two to four orders within the experimental values, and this peak was followed by a
of magnitude less than the highest frequency which can be cap- significant drop, which was also apparent in repeat 3 of the exper-
tured by the model [22], which is related to the stable time incre- iments. In the model, the force recovered slightly between exten-
ment. Here, reaction forces at the fixed end of the joint were output sions of 2.5 mm and 4 mm, whereas the experimental readings
at every increment. This ensured that all solution frequencies were showed a gradual load drop-off. Beyond 4 mm, the experimental
present in the raw signal, preventing any possibility of aliasing and model response again match well. The load in the model finally
during post-processing. In the post-processing phase, this raw data reduced to zero at an extension of 7 mm signifying complete sep-
was filtered to remove unwanted high-frequency noise in the aration of the joint parts. The ability to simulate bearing damage to
numerical solution. Various low pass filters were trialled, and a fil- complete separation, involving a joint displacement of 7 mm, with-
ter of 5 kHz was found to adequately remove high frequency con- out the occurrence of a numerical instability, indicates exceptional
tent without excessively smoothing the predicted response. It is model robustness. A video of the joint behaviour, through to final,
noted that the raw force signals from the high speed tests of [14] catastrophic failure of the bolt head, is included in the Supplemen-
were low-pass filtered at a frequency of 2.5 kHz to remove oscilla- tary material.
tions associated with the test rig. The higher filter cut-off applied The development of compressive fibre damage in the CLam_B1
to the numerical results gave confidence that the structural model is shown in Fig. 7. At a displacement of 2 mm, the counter-
response of tests was adequately represented in the filtered sunk head has begun to separate. After separation of the counter-
response. sunk head, the assembly remains intact, with the bolt shank
The filtered and unfiltered force signals for the calibrated gradually pulling through the countersunk hole (see inset at
CLam_B1 joint model are presented in Fig. 6(a). While the 6 mm displacement). At this point, the crushing composite
1204 B. Egan et al. / Composite Structures 133 (2015) 1198–1208

Fig. 6. Response of (calibrated) CLam_B1 model – (a) load–deflection curve (filtered and unfiltered) and (b) load–extension curve (including experiments).

Fig. 7. Separation of the CLam_B1 model, showing development of compressive fibre damage (deformation scale: x1).

Fig. 8. FP failure in the CLam_B1 joint – (a) 10 m/s test and (b) simulation (deformation scale: x1).

‘‘flows’’ up under the nut, at the non-clamped side of the NCSK 4.2. Single-Bolt failure prediction
hole. A high speed test image taken just prior to joint separation
is compared to the CLam_B1 model in Fig. 8. Similar bolt With model parameters kept unchanged from the calibrated
rotations are observed in experiment and in simulation. Very model, the thicker joint exhibiting fastener pull-through (FP) fail-
significant countersunk hole crushing is evident in both model ure was simulated in a model denoted ‘‘ELam_B1_FP’’. The pre-
and test at final separation. dicted response curve is plotted alongside the experimental
 B. Egan et al. / Composite Structures 133 (2015) 1198–1208 1205

Fig. 9. Response of ELam_B1 models compared to experiment – (a) FP model and (b) FF model filtered load–extension curves.

response of the sample which exhibited fastener pull-through and resulted in further hole crushing. This led to the final
(repeat 2) in Fig. 9(a). It is evident that the elastic response of drop-off slope in the load deflection curve (Fig. 9(b)) being more
the FP model accurately captures experimental behaviour in the gradual than in the tests. This discrepancy could be minimised
early loading region. However, the onset of damage occurs earlier by reducing the interior friction parameter which controls the fric-
than in experiments and the predicted ultimate strength, at tion forces acting between the fastener and eroding holes. Reduced
10 kN, is 1–2 kN lower than experimental values. Inaccuracy in bolt-hole friction forces would facilitate an easier release of the
the ultimate strength could be due to the use of cohesive elements fastener from the NCSK hole, limiting crushing during separation.
(calibrated for the CLam_B1 model) to simulate head fracture, the However, employing reduced interior friction across all models
properties of which are not rate dependent. Ultimate strengths of would diminish the stable crushing loads in the other models
ELam_B1 joints in high speed tests were considerably higher than and so was not invoked.
those attained in quasi-static tests, which was attributed to a In the work of Pearce et al. [11], stacked shell model predictions
rate-enhancement in the mechanical behaviour of the fastener. of a 10 m/s joint response, diverged significantly from experimen-
This would not be captured in the failure of the FP cohesive zone. tal results after extensive bearing damage. This was attributed to
After the ultimate strength has been attained in the simulation, the inability of the model to reproduce behaviour of debris mate-
there is a sharp drop-off in load and the slope of the predicted rial (of woven fabric composite) and it was proposed that smooth
drop-off correlates very well with experiment. This drop-off is particle hydrodynamics (SPH) be considered to overcome this. In
due to the crack in the countersunk head. Following this, further contrast, although an initial calibration was necessary, the 3D sim-
hole crushing develops until complete pull-through, as is evident ulations conducted here showed reasonably good agreement with
by the levelling-off in the response after the initial load drop. In experimental response curves, even after significant bearing dam-
the simulation, a levelling-off was observed at an extension of age development. This improvement may be partially attributed to
3 mm, where the crushing load was just over 4 kN. Again this is the fact the shell elements do not calculate the through-thickness
lower than the post-failure crushing load in the experiment, which shear or normal stresses accurately [31].
began at about 6 kN. Strengths and crushing loads attained in the Images from the high speed tests taken just prior to the separa-
FP model naturally depend on the choice of cohesive strengths tion of ELam_B1 joints are compared with model images in Fig. 10.
and friction coefficients. However, this model, which utilises The FF model, shown in Fig. 10(b), exhibits a crack in the counter-
parameters calibrated on a separate baseline test case, shows rea- sunk bolt head at the side of the hole not in contact with the bolt,
sonable qualitative agreement with experiment. in addition to detachment of the nut section. As discussed in [14],
In a separate model (‘‘ELam_B1_FF’’), fastener fracture at a the head crack on the non-load-bearing side of the countersunk
thread was captured by activating cohesive elements at the thread hole is caused by the high out-of-plane forces required to maintain
location. In trial simulations, where the mode 1 strength was pro- equilibrium. In line with experiment, this was solely a feature of
gressively reduced from the 10,000 MPa (‘‘de-activated’’ value), the ELam_B1 simulations - the CLam_B1 model did not exhibit
1950 MPa was determined to be the highest value which permitted crack initiation at the non-loaded side (but the head was rather
FF failure. The response of the FF model, shown in Fig. 9(b), is ‘‘ripped off’’ from the load-bearing side). The FP model of
broadly similar to that of the FP model of Fig. 9(a) up to an exten- Fig. 10(b) shows separation of the countersunk bolt head section,
sion of about 2.5 mm, i.e. a similar elastic response, gradual stiff- as in the CLam_B1 (FP) model. Interestingly, a detached shard of
ness loss, ultimate strength and initial load drop-off are the countersunk head is also observable in corresponding high
predicted. The countersunk head crack which limited the strength speed test footage. The final failure in the ELam_B1_FF model is
of the FP model, also occurred in the FF model, as had been the case compared with experiment in Fig. 11. The limited bearing damage
in experiment. However, unlike that of the FP model, the at both holes (compared to a CLam_B1 model), the exposed frac-
ELam_B1_FF model response does not exhibit a levelling-off to sig- ture surface at the shank, and the crack in the countersunk head
nify stable crushing beyond 2.5 mm, but shows the load dropping are all features of the experimental test.
to zero much sooner. This is in line with experimental behaviour, The single-bolt joint models demonstrated that fastener frac-
where fastener fracture at a thread (FF failure) led to earlier termi- ture at a thread (FF failure) was more likely to occur in an ELam_B1
nation of the crushing process and reduced energy absorption. joint than in a CLam_B1 joint. To examine this further, the same
Separation of the FF joint model, after fracture in the FF zone, mode 1 strength (1950 MPa) was applied to the cohesive elements
was somewhat delayed compared to the joint separation observed at the FF zone in a CLam_B1 model. In this model, the FF cohesive
in high speed test footage. This was due to model friction forces elements did not fracture, and an identical FP response to that
acting between the fastener shank and non-countersunk hole, observed for the original CLam_B1 (FP) model was again observed.
which prevented immediate release/separation after FF fracture This showed that FF failure, which dramatically reduced energy
1206 B. Egan et al. / Composite Structures 133 (2015) 1198–1208

Fig. 10. FF and FP failures in the ELam_B1 joint – (a) 10 m/s tests and (b) simulation (deformation scale: x1).

Fig. 11. Post-failure deformation in parts of the ELam_B1_FF model (deformation scale: x1) compared to experiment (ELam_B1_1).

absorption, would occur more readily in an E-layup joint than in a in experiments, this indicates that energy absorption (which is
C-layup joint. This had also been an important conclusion of the considered in the next section) is under-predicted in the simula-
experiments in [14]. tion. This is corroborated by comparisons of post-failure counter-
sunk hole elongations of model and experiment (see Fig. 13). It is
4.3. Multi-Bolt failure prediction clear that crushing at each hole in the model is marginally less than
that of experiment. In spite of this, the predicted post-failure defor-
The CLam_B3 model is shown in Fig. 12(a). The filtered and mation of the countersunk laminate appears highly satisfactory.
unfiltered forces extracted from the fixed end of the CLam_B3 joint
are plotted in Fig. 12(b). The peak load of the filtered curve, at 4.4. Energy absorption
21.5 kN is about three times that obtained in the single bolt
(CLam_B1) joint simulation of Fig. 6(a). Force is plotted against dis- Energy absorption was calculated by integrating the area under
placement at the DIC location, for both experiments and simula- the filtered load–deflection curves. In the high speed joint testing
tion, in Fig. 12(c). The maximum available displacement from DIC study of [14], energy absorption values exhibited significant scat-
in experiments was 10 mm. As can be seen in the force-extension ter. The scatter was not unique to this test series and had been pre-
curves of Fig. 12(c), the predicted elastic response captures that of viously attributed by Li et al. [32] to the instabilities which arise as
experiment reasonably well, and the prediction of ultimate fasteners progressively pull through composite material. Consider-
strength, at 21.5 kN, is very close to the experiments. However, ing both the significant experimental scatter and the limitations of
the predicted rate of load drop-off (following ultimate failure) modelling debris using mesh-based methods [12], accurate
was significantly greater than in the experiments. Combined with prediction of energy absorption is considered to be beyond
the fact that joint separation occurred sooner in the model than the capability of the current models. Thus, the focus here is on
 B. Egan et al. / Composite Structures 133 (2015) 1198–1208 1207

Fig. 12. CLam_B3 model – (a) Joint mesh, (b) load versus end displacement and (c) load versus displacement at DIC point.

Fig. 13. CLam_B3 countersunk laminate post-failure-test (CLam_B3_1) and simulation.

the prediction of variations in energy absorption across the range of 5. Conclusions

joints considered.
The energy absorbed in the ELam_B1_FP model, at 32.7 J, was Catastrophic failure of dynamically-loaded countersunk com-
only marginally greater than that absorbed in the CLam_B1 model posite bolted joints was simulated using fully three-dimensional
(31.9 J). This accurately reflects the important experimental obser- FE models for the first time, with a strong focus on modelling fas-
vation, that a thicker joint is not necessarily a better tener failure. This is in contrast to previous 3D numerical studies
energy-absorber. The energy absorbed in the ELam_B1 model which focus solely on composite damage or include only a simple
exhibiting thread fracture (ELam_B1_TV10_FF), was 26.1 J, signifi- plasticity law to capture nonlinearity in the fastener.
cantly lower than the FP model. A similar observation was made A degree of model calibration was required for these complex
in the experiments, where the fastener thread fracture was found analyses which feature both composite and fastener damage. How-
to terminate the energy-absorbing crushing process sooner than ever, following calibration of certain model parameters for the
if the joint had proceeded to fail by fastener pull-though. Energy CLam_B1 baseline case, the simulations exhibited key trends
absorption in the CLam_B3 model, at 111.2 J, was over three times observed in experiments:
that of the CLam_B1 model, again reflecting a trend of experiment.
Although the 3D models could accurately capture experimental 1. The ultimate load achieved in the ELam_B1 joint simulation was
trends, the predicted energy magnitudes were generally as low higher than that of the CLam_B1 model. However, it was lower
as half what was obtained in the experiments of [14]. Accurate pre- than that of experiment, which may be attributed to an
diction of energy absorption is considered a challenge for future enhanced mechanical response of the fastener in dynamic tests
modelling work, where the treatment of both fastener fracture which would not have been accounted for in the cohesive zone
and composite debris will need to be improved. model.
1208 B. Egan et al. / Composite Structures 133 (2015) 1198–1208

2. Energy absorption in the E-layup joint model was not greater References
than that of the CLam_B1 model, in line with the somewhat sur-
prising experimental results. However there was a significant [1] Marsh G. Wing worker for the world. Reinf Plast 2010;54:24–8.
[2] Hinton MJ, Soden PD. Predicting failure in composite laminates: the
under-prediction of energy absorption across all models, which background to the exercise. Compos Sci Technol 1998;58:1001–10.
was attributed to limitations on accurately simulating hole [3] Dávila CG, Jaunky N, Goswami S. Failure criteria for FRP laminates in plane
crushing late in the loading history, where mesh-based FE mod- stress. AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics, and
materials conference; 2003.
els may not accurately represent debris material. [4] Dávila CG, Camanho PP. Failure criteria for FRP laminates in plane stress. NASA
3. Fastener fracture at a thread (FF failure) was shown to be more technical report: NASA; 2003.
likely to occur in the E-layup models, since FF failure did not [5] Pinho ST, Dávila CG, Camanho PP, Iannucci L, Robinson P. Failure models and
criteria for FRP under in-plane or three-dimensional stress states including
occur in a CLam_B1 model where the same (mode 1) strength,
shear nonlinearity. NASA/TM-2005-213530; 2005.
as that in the ELam_B1_FF model, was applied. [6] Puck A, Schürmann H. Failure analysis of frp laminates by means of
4. The multi-bolt CLam_B3 joint simulation exhibited ultimate physically based phenomenological models. Compos Sci Technol 1998;58:
loads and energy absorptions which were approximately three 1045–67.
[7] Puck A, Schürmann H. Failure analysis of FRP laminates by means of physically
times larger than the corresponding single-bolt joint model, based phenomenological models. Compos Sci Technol 2002;62:1633–62.
as had been the case in experiments. [8] Heimbs S, Hoffmann M, Waimer M, Schmeer S, Blaurock J. Dynamic testing and
modelling of composite fuselage frames and fasteners for aircraft crash
simulations. Int J Crashworthiness 2013;18:406–22.
Some practical experience gained in this high speed 3D mod- [9] Egan B, McCarthy MA, Frizzell RM, Gray PJ, McCarthy CT. Modelling bearing
elling may be helpful to other analysts: failure in countersunk composite joints under quasi-static loading using 3D
explicit finite element analysis. Compos Struct 2014;108:963–77.
[10] Ostergaard M, Ibbotson A, Roux O, Prior A. Virtual testing of aircraft structures.
 Filtering of the dynamic reaction force signal obtained from the CEAS Aeronaut J 2011;1:83–103.
model was necessary to remove high frequency numerical [11] Pearce G, Johnson A, Thomson R, Kelly D. Numerical investigation of
noise. This noise is purely numerical and so is completely differ- dynamically loaded bolted joints in carbon fibre composite structures. Appl
Compos Mater 2010;17:329–46.
ent to high frequency noise typically seen in high speed exper- [12] Pearce GMK, Johnson AF, Hellier AK, Thomson RS. A stacked-shell finite
imental force signals. element approach for modelling a dynamically loaded composite bolted joint
 Lowering the element deletion criteria may be helpful where under in-plane bearing loads. Appl Compos Mater 2013;20:1025–39.
[13] Pearce GMK, Johnson AF, Hellier AK, Thomson RS. A study of dynamic pull-
very highly distorted elements break down the contact enforce-
through failure of composite bolted joints using the stacked-shell finite
ment or cause the solution to terminate prematurely. element approach. Compos Struct 2014;118:86–93.
 It is worth considering if regions of (initially) highly skewed ele- [14] Egan B, McCarthy CT, McCarthy MA, Gray PJ, O’Higgins RM. Static and high-
ments (e.g. at meshes around the countersunk) should be given rate loading of single and multi-bolt carbon–epoxy aircraft fuselage joints.
Compos A Appl Sci Manuf 2013:97–108.
a lower element deletion criterion compared to more regular [15] Gaudin J. MAAXIMUS: Delivering Innovation. Aerodays Madrid; 2011.
mesh regions. If elements are initially very skewed, the large [16] Gray PJ, O’Higgins RM, McCarthy CT. Effect of thickness and laminate taper on
shear strain thresholds which are typically applied as element the stiffness, strength and secondary bending of single-lap, single-bolt
countersunk composite joints. Compos Struct 2014;107:315–24.
deletion criteria may be unsuitable. Alternatively, the meshing [17] Gray PJ, O’Higgins RM, McCarthy CT. Effects of laminate thickness, tapering
strategy may be altered, but it is difficult to avoid non-skewed and missing fasteners on the mechanical behaviour of single-lap, multi-bolt,
elements where the fastener head angle is high (e.g. 130°). countersunk composite joints. Compos Struct 2014;107:219–30.
[18] Egan B, McCarthy CT, McCarthy MA, Gray PJ, Frizzell RM. Modelling a single-
bolt countersunk composite joint using implicit and explicit finite element
In future, a number of developments should help bypass the analysis. Comput Mater Sci 2012;64:203–8.
calibration required here. These include: [19] McCarthy CT, McCarthy MA, Stanley WF, Lawlor VP. Experiences with
modeling friction in composite bolted joints. J Compos Mater 2005;39:
1881–908.
 Modelling of the countersunk fastener to a greater degree of [20] Bažant ZP, Oh BH. Crack band theory for fracture of concrete. Mater Struct
accuracy, by capturing small scale details such as the 1983:155–77.
[21] Pinho ST, Iannucci L, Robinson P. Physically based failure models and criteria
thread geometry, the hexagonal shape of the nut and even the
for laminated fibre-reinforced composites with emphasis on fibre kinking. Part
manufacturer’s stamp which act as local stress raisers. II: FE implementation. Compos A Appl Sci Manuf 2006;37:766–77.
 Using a sophisticated material subroutine for simulating plas- [22] Abaqus-Inc. Abaqus User Manual, Version 6.11. Dassault Systèmes Simulia
ticity, damage and fracture in Ti-6Al-4V. A comprehensive suite Corp, Providence, RI, USA; 2012.
[23] Lesuer DR. Experimental investigations of material models for Ti–6Al–4V
of test data for the titanium, identifying failure envelopes for titanium and 2024-T3 aluminium. Washington D.C.: U.S. Department of
shear and ductile metal failure modes, would also be needed. Transportation Federal Aviation Administration; 2000.
 Modelling the contact interactions using a rate-dependent [24] Gilat A, Goldberg RK, Roberts GD. Experimental study of strain-rate-dependent
behavior of carbon/epoxy composite. Compos Sci Technol 2002;62:1469–76.
friction formulation. [25] Hsiao HM, Daniel IM, Cordes RD. Strain rate effects on the transverse
compressive and shear behavior of unidirectional composites. J Compos
Improved facilities for modelling debris within commercial FE Mater 1999;33:1620–42.
[26] Koerber H, Xavier J, Camanho PP. High strain rate characterisation of
codes such as Abaqus, would also enhance the predictive capability unidirectional carbon–epoxy IM7-8552 in transverse compression and
of these types of failure analyses. in-plane shear using digital image correlation. Mech Mater 2010;42(11):
1004–19.
Acknowledgements [27] Daniel IM, Werner BT, Fenner JS. Strain-rate-dependent failure criteria for
composites. Compos Sci Technol 2011;71:357–64.
[28] Pearce GMK. High strain-rate behaviour of bolted joints in carbon fibre
The authors wish to acknowledge the Irish Research Council for
composites structures. Sydney: The University of New South Wales; 2009.
Science Engineering and Technology (IRCSET) and the SFI/HEA Irish [29] Egan B, McCarthy CT, McCarthy MA, Frizzell RM. Stress analysis of single-bolt,
Centre for High-End Computing (ICHEC) for provision of financial single-lap, countersunk composite joints with variable bolt-hole clearance.
support and computational facilities. Compos Struct 2012;94:1038–51.
[30] Hammer JT. Plastic deformation and ductile fracture of Ti–6Al–4V under
various loading conditions. The Ohio State University; 2012.
Appendix A. Supplementary material [31] Hinton MJ, Kaddour AS. Triaxial test results for fibre-reinforced composites:
the Second World-Wide failure exercise benchmark data. J Compos Mater
2012.
Supplementary data associated with this article can be found, in [32] Li QM, Mines RAW, Birch RS. Static and dynamic behaviour of composite
the online version, at http://dx.doi.org/10.1016/j.compstruct.2015. riveted joints in tension. Int J Mech Sci 2001;43:1591–610.
07.004.