You are on page 1of 18

Int J Fract (2007) 143:143–160

DOI 10.1007/s10704-007-9054-9

O R I G I NA L PA P E R

Deformation and failure modelling of high strength


adhesives for crash simulation
Lars Greve · Florence Andrieux

Received: 23 October 2006 / Accepted: 29 January 2007 / Published online: 16 March 2007
©Springer Science+Business Media B.V. 2007

Abstract The deformation and failure mechan- Keywords Adhesives · Yield · Plasticity ·
isms of toughened high strength adhesives used in Failure · Modelling · Crash
the automotive industry are very complex and req-
uire advanced numerical models for crashworthi- Nomenclature
ness simulation. The theoretical background of two
new modelling approaches for thin adhesive lay- GIC , GIIC Critical strain energy release rate for
Mode I and Mode II loading
ers is presented: firstly, a simplified elastic damag- δI , δII Relative out-of-plane (Mode I) and in-
ing node-to-element tied interface model approach plane (Mode II) interface element dis-
for convenient and efficient modelling, and sec- placements
ondly a detailed modelling approach for improved σI , σII Out-of-plane (Mode I) and in-plane
(Mode II) interface element stresses
accuracy using an elasto-viscoplastic solid element E, G Young’s modulus, shear modulus
representation of the adhesive layer. The material dA, dW Infinitesimal area and energy
model parameters required for both approaches dI , dII Out-of-plane (Mode I) and in-plane
are determined by a comprehensive set of exper- (Mode II) damage functions
<x>+ Operator returns ‘x’ if x > 0 and 0 oth-
iments, including quasi-static and dynamic adhe- erwise
sive coupon testing, fracture toughness testing, and supt≤τ (x) Supremum of x: Maximum of x within
quasi-static tension/shear (and combined) testing the range 0 ≤ t ≤ τ
of thin adhesive layers. A more complex adhesively I1 , J2 First invariant of the stress tensor and
second invariant of the stress deviator
joined assembly of two aluminium extrusions sub- σm Hydrostatic stress
jected to quasi-static (QS) and dynamic loading σe Equivalent von Mises stress
serves as the final validation example for both mod- I Identity matrix
elling approaches. Good agreement of experiments σ , εe , εp Stress tensor, elastic and plastic strain
tensor
and numerical predictions was observed for both E Elasticity matrix
modelling approaches. σx , τxy Axial stress and torque shear stress
 p
σT εT Tensile strain hardening curve
 Yield function
L. Greve (B) α, β Shape parameters for the yield function
Volkswagen AG, 38436 Wolfsburg, Germany and the flow potential
e-mail: lars.greve@volkswagen.de σ0 Shift stress value representing the cen-
F. Andrieux tre of the yield locus
Fraunhofer Institut für Werkstoffmechanik, σ̃ Special effective equivalent stress
p
Woehlerstr. 11, 79108 Freiburg, Germany ε̇T Plastic strain rate threshold
144 L. Greve, F. Andrieux

C Fitting parameter for the strain rate The von Mises yield criterion (von Mises 1928) is
model
λ̇ Plastic multiplier
based on the assumption that material yield solely
depends on deviatoric stresses, σ ij . It has been
Abbreviations particularly useful for predicting yield of various
isotropic metals, where the yield condition is deter-
DCB Double Cantilever Beam mined by the second invariant
DF Deshpande and Fleck  J2 of the
 deviator-
DYN Dynamic
ic stress tensor, φvonMises = f J2 σ  ij . However,
Exp, Sim Experiment and Simulation the yield stress of the high strength epoxy adhe-
FE Finite Element sive investigated in this paper also depends on
JC Johnson and Cook the hydrostatic stress σm , which is related to the
QS Quasi-static
first invariant of the stress tensor I1 , φAdhesive =
f (I1 , J2 ). Schlimmer (Schlimmer 1974) proposed
1 Introduction such a hydrostatic stress dependent plasticity model
suitable for adhesives, which generalises the yield
Automotive body design increasingly incorporates criteria of Drucker and Prager (1952) and von
a mixture of different materials for optimised light- Mises (1928). The model covers the features (a)
weight design, e.g., Audi TT (2006). This requires and (b) outlined above. Recently, Mahnken and
advanced jointing techniques such as riveting or Schlimmer (2005) have implemented the model in
adhesive bonding, since conventional spot welding a commercial code for the simulation of elasto-
can not be applied. Thin layered adhesive bonding plastic deformation of high strength adhesives sub-
is an attractive option since it also can significantly jected to tensile, shear, and combined tensile/shear
improve the bending, torque, and dynamic stiff- loading.
ness in comparison to conventional spot welding. A similar plasticity model has been proposed
Of particular interest are toughened high strength by Deshpande and Fleck (DF) (2000) for model-
epoxy adhesives (BetamateTM 1496), which offer ling the deformation of compressible foams, which
high strength in conjunction with increased duc- addresses features (a) and (b) as well. Due to its
tility for improved crashworthiness. The ductility simplicity, the DF model is used as the foundation
limit of these adhesive materials must be known to of the plasticity model and is extended in this paper
ensure structural integration of a vehicle during a to meet the requirements for adhesive materials by
crash. addressing features (c) and (d).
Today, vehicle design is significantly supported The fracture model by Johnson and Cook (JC)
by virtual prototyping, including numerical crash- (1985) has been successfully applied to predict the
worthiness analysis using crash simulation codes. plastic failure strain of isotropic metals. The model
Prediction of the complex deformation and failure is based on the principle of accumulated dam-
mechanisms of high strength adhesives requires age, where the rate of damage can depend on
advanced material models and modelling techni- the stress triaxiality σm /σeq and also on the strain
ques, which must cover the following phenomena: rate. In this paper the JC model is adopted to pre-
dict adhesive failure since it allows modelling fea-
(a) Hydrostatic stress dependent yield. Simple von tures (e) and theoretically also (f) outlined above.
Mises (von Mises 1928) deviatoric stress the- It operates in conjunction with the enhanced DF
ory (J2 plasticity) does not apply. model in order to provide a comprehensive 3D
(b) Material deformation without conservation of (solid element) material model for high strength
the volume. adhesives.
(c) Non-symmetric yield. The compressive yield Accurate experimental measurement of the fail-
stress is significantly higher than the tensile ure strains of thin adhesive layers still remains an
yield stress. unresolved issue, since conventional test methods
(d) Strain rate dependent yield and hardening. can not be applied. Literature resources
(e) Hydrostatic stress dependent fracture. concerning dynamic testing provide only informa-
(f) Strain rate dependent fracture. tion about the dynamic strength of high strength
Deformation and Failure modelling of high strength adhesives for crash simulation 145

adhesives (Sato and Ikegami 1999). An ongoing 2 Theory of material modelling


research project is dedicated to accurate dynamic
strain measurement (FOSTA P676 2005–2007) in 2.1 The simplified modelling approach (Interface
order to resolve this discrepancy. Due to these cur- modelling)
rent experimental limitations the effect of strain
rate dependent failure is not investigated in this The tied connection interface material model (type
paper. 303) in PAM-CRASH Manual uses an elastic dam-
Large scale Finite Element (FE) models used aging material model. Although the investigated
in the automotive industry encourage simple and high strength adhesive exhibits considerable plas-
efficient modelling. Representing adhesive layers tic deformation, the total displacement prior to
using solid elements, as proposed above, obviously failure is very small. Considering pure shear load-
leads to increased modelling efforts. Specialised ing of a 0.5 mm thick adhesive layer the shear
interface elements have been established in the failure strain will be about 80%, which however
past for tying two separated components. Vari- corresponds to a total shear displacement of only
ous different tying methods have been developed, 0.4 mm prior to failure. The maximum displace-
ranging from discontinuous point connections to ments are even lower under tensile or combined
continuous line or area connections. A valuable loading. Based on these considerations plastic def-
summary of references is given in (Camanho et al. ormation in the adhesive interface model is neg-
2001). lected for the interface model, which is not the case
The main purpose of the interface models was for the advanced solid element approach discussed
to predict delamination of quasi-brittle laminated in the next section.
composites. Simple quadratic stress failure crite- For the interface model the adhesive layer is rep-
ria, e.g., (Camanho and Matthews 1999), or more resented by a series of discontinuous connections
advanced criteria based on the critical strain energy of slave nodes and master elements, Fig. 1, which
release rate (Johnson et al. 2001; Greve and Pickett can have different mesh topologies. The relative
2005) have been utilised. movement of the slave node is decomposed into
More recently, “pseudo-solid” interface ele- out-of-plane (Mode I) and in plane shear (Mode
ments, using elasto-plastic constitutive laws, have II) displacements.
been applied to model adhesive deformation and
failure (Mahnken and Schlimmer 2005). However,
this interface approach uses 4Node-To-4Node con-
2.1.1 The critical strain energy release rate
nection elements (ABAQUS Manual), which
requires congruent meshes of the bonded compo-
The critical strain energy release rate GiC repre-
nents and increases the modelling effort. To over-
sents the heart of the interface model. For Mode I
come these limitations the mesh independent Node-
loading, the link of this Fracture Mechanics cri-
To-Element tied interface provided by the explicit
terion and the corresponding Damage Mechan-
crash code PAM-CRASH Manual is used in this
ics formulation used for the interface elements,
paper as an alternative to the solid element mod-
is shown in Fig. 2. The Damage Mechanics law is
elling approach. The interface approach benefits
based on the work by Hillerborg et al. (1976).
from available automatic partner search and con-
nection procedures for joining parts, and from a
significantly reduced set of material model param-  
dW dWEle FI
eters. The major drawback of the interface ap- GIC = = = dδI = σI dδI
dA dAEle AEle
proach is the negligence of visco-plastic deforma-
(1)
tion. Showing the influence of this simplification in
comparison with the detailed solid element mod-
elling technique for the predictability of adhesive Linear coupling of the strain energy release rates
failure of practical complex structures is another for crack initiation (GI0 − GII0 ) and crack
goal in the present study. propagation (GIC −GIIC ) is assumed between pure
146 L. Greve, F. Andrieux

Fig. 1 Slave node to I (out of-plane displacement*)


Master element
connection technique and Adhesive Slave
displacement layer
node Shell layers
decomposition of the tied thickness II (in-plane displacement*)
with arbitrary
interface material model
mesh topology
(type 303) provided in Adhesive
PAMCRASHTM interface
elements

Master
element *) Relative displacement with respect to the Master element movement

Fig. 2 Link between Force-displacement curve of


Fracture Mechanics and the global system
Damage Mechanics for Crack opens
the adhesive interface (a) F
dA F
element demonstrated for
dW
pure Mode I loading.
Parameter dW is the Reduced system stiffness
released elastic energy in s
the considered system to s
create a crack of the area
F
dA. The corresponding
Stress-displacement curve of
FE approach defines the
the interface element
absorbed energy WEle as
the area under the (b) Slave I I

stress–displacement curve Node E


using a simple triangular
stress–displacement Adherent layers Adhesive interface
relation element GIC
I
Master
Element

Mode I and pure Mode II loading, Fig. 3, where E is the Young’s modulus, G is the shear
modulus, dI,II are the damage functions, and εI,II
GI GII GI GII are the elastic strains.
+ = 1; + = 1. (2)
GI0 GII0 GIC GIIC The following formulae describe Mode I load-
ing. Identical equations can be established for
For arbitrary combined load paths, the strain en- Mode II by interchanging indices I with II and the
ergy release rates are decomposed into (GIA , GIIA ) out-of-plane modulus E with the in-plane shear
for crack initiation and (GIB , GIIB ) for complete modulus G, respectively.
failure, Fig. 3. A simple triangular shape is imposed for the
elastic damaging stress–strain relation, Fig. 4. Point
2.1.2 The elastic damaging stress–strain relation A represents damage initiation and point B repre-
sents the fully damaged state (dI = 1) for arbitrary
A simple elastic damaging stress–strain relation is combined loading. The strains for crack initiation,
used for the interface elements, εIA, , and ultimate failure, εIB , follow from the
    
σI E (1 − dI ) 0 εI requirement of letting the area under the stress–
= ;
σII 0 G (1 − dII ) εII displacement curve equal the corresponding strain
    energy release rate, GIA,B (as explained in
εI 1 δI
= , (3) Fig. 2):
εII t δII
Deformation and Failure modelling of high strength adhesives for crash simulation 147

Fig. 3 Coupling of Mode GI


I and Mode II strain Damage initiation
energy release rates for
GIC Load path
crack initiation and
propagation (ultimate
damage)
GI / GII
B
Ultimate damage
GIB
GI0 Point (GI, GII)
A
GIA
Damaging zone

GII
GIIA GII0 GIIC
GIIB
Note: exaggerated display for the initial
strain energy release rates Gi0. Usually: Gi0 << GiC

1 max 1 max 2GIA below:


GIA = σ δIA = σ εIA t ⇒ εIA = max ,
2 I 2 I σI t
1 1 max 2GIB εI  2
GIB = σImax δIB = σ εIB t ⇒ εIB = max . t σI 2+ t σImax
2 2 I σI t GI = t σI dεI = ⇒ GI0 = ,
2E 2E
(4) 0
(6)
For pure Mode I loading GIA,B corresponds to
GI0,C . where only positive tensile stresses contribute to
The imposed triangular shape of the stress–strain potential fracture initiation. Using Eq. (2) and
relation leads to the following definition of the substituting the energy release rates by the expres-
damage function: sions in terms of stresses, Eq. (6), yields the non-
symmetric quadratic stress failure locus for crack

⎪ ε (εI (τ )−εIA ) initiation,
⎨ supt≤τ εIB
I (τ )(εIB −εIA )
if (εI (τ ) > 0)
+
dI = , (5)  2  2

⎩ σI + σII
0 otherwise + = 1, (7)
σImax σIImax

where it has to be noted that, according to the where maximum tensile and shear stresses, σImax
definition, no damage occurs under Mode I com- and σIImax , have to be determined by experiments.
pression loading. In the case of unloading, the
stress–strain curve intersects the origin (no plastic
deformation), Fig. 4. This distinction is irrelevant 2.2 The detailed modelling approach (Solid
for shear (Mode II) loading. element modelling)

The DF model (Deshpande and Fleck 2000) pro-


2.1.3 The fracture criterion for crack initiation vides the foundation of the solid element approach
and is enhanced to address unsymmetrical yield
According to Fig. 4, the following relation of str- and strain rate hardening. The enhanced DF model
esses and corresponding strain energy release rates operates in conjunction with the JC fracture crite-
exists at the threshold to crack initiation and rion (Johnson and Cook 1985) and has been imple-
148 L. Greve, F. Andrieux

Fig. 4 Triangular I
stress–strain relation and
graphical representation max
A
of the energy release rates I

shown for Mode I loading E


GIB/t (1-dI)E

GIA/t
B
I
IA I IB

mented in a user subroutine of the commercial stress σe ,


crash code PAM-CRASHTM . A similar routine has

been implemented in the crash code LS-DYNA 3   
(Memhard et al. 2005). σe = σ σ = 3J2 ; σ  = σ − σm I. (11)
2 ij ij
2.2.1 Stress–strain relation
The enhanced DF yield criterion is
Isotropic behaviour is assumed for the adhesive
material. The corresponding relation between  = σ̃ − B ≤ 0, (12)
stresses σ and elastic strains εe is
where σ̃ is the effective equivalent stress,
σ = Eε e , (8)
σ̃ 2 = σe2 + α 2 (σm − σ0 )2 , (13)
where E is the symmetric 6 × 6 stress–strain matrix
(Eij = Eji ), with
which is a function of the von Mises equivalent
E11 = E22 = E33 = (1 − ν e ) c; stress and of the hydrostatic stress. The yield func-
E44 = E55 = E66 = G; tion  takes the shape of an ellipse in the σe − σm
E12 = E23 = E31 = ν e c;
(9) stress space, Fig. 5, defined by B and A = B/α.
c = (1+ν e )(1−2ν
E
; G= E Parameter σ0 allows improved adjustment to expe-
e) 2(1+ν e ) .
rimental observations by shifting the ellipse along
the σm -axis.
Parameter E is the elastic modulus, G is the shear
Strain hardening and strain rate hardening The
modulus, and ν e is the elastic Poisson’s ratio.
quasi-static true tensile stress
 pversus
 plastic tensile
2.2.2 Yield condition, strain rate hardening, and strain hardening curve σTsta εT can be obtained by
flow rule a quasi-static tension test of an adhesive coupon.
Any monotonically increasing curve function can
Yield condition For isotropic materials it is be input. A direct relation exists between uniaxial
convenient to use invariant stress and strain ten- tensile stress σT and yield function parameter B,
sor expressions for modelling plastic deformation using Eq. (12) with σe = σT , and σm = σT /3:
and fracture. The hydrostatic stress, σm , is related 
to the first invariant of the stress tensor I1 :  2
1 σ0
B= 1 + α2 − σT . (14)
3 σT
1 I1
σm = Trσ = , (10)
3 3
For addressing strain rate hardening a simple log-
whereas the second invariant of the deviator stress arithmic model is proposed (Johnson and Cook
tensor, J2 , is related to the equivalent von Mises 1985),
Deformation and Failure modelling of high strength adhesives for crash simulation 149

Fig. 5 Principle plot of


the yield locus and the
flow potential in the stress
invariant space
represented by σm and σe

⎧  p 
⎪ ε̇  p  p p

⎨ 1 + C ln Tp σTsta εT if ε̇T > ε̇0
 p p ε̇0
σT εT , ε̇T = , (15)

⎪  p

σTsta εT otherwise

p
where C is a fitting parameter, ε̇T is the plastic ten- Damage occurs once the accumulated damage
p
sile strain rate, and ε̇0 is the plastic tensile strain exceeds D = 1.
rate threshold below which the static yield stress
p
σTsta εT is used.
Flow potential and flow rule A non-associated 3 Material model parameter identification
flow rule is used to address non-conservation of procedure
the volume observed for the adhesive material (see
Sect. 3.1), where the flow potential is defined by an The material model parameters for the simplified
unsymmetrical elliptic function, interface approach and the detailed solid element
approach are obtained from the same set of exper-

p ∂G iments. The high strength epoxy adhesive Beta-
G= σe2 + β 2 σm 2+ ; ε̇ij = λ̇ , (16)
∂σij mateTM 1496 is used for all tests. The whole set
of material model parameters is summarised in
where β is the shape parameter, and λ̇ is the plas- Appendix Table 1.
tic multiplier. Negative hydrostatic stresses do not
contribute to volumetric changes, Fig. 5.
Fracture Criterion The JC fracture criterion 3.1 Uniaxial tension and compression tests using
(Johnson and Cook 1985) is based on the idea of adhesive specimens
accumulated damage, defined as the ratio of equiv-
alent von Mises plastic strain increments dεp to the The quasi-static mechanical properties of Beta-
corresponding failure strain εf , mateTM can be obtained by several sources, e.g.,
(FOSTA P593 2002–2004), where usually simple
  adhesive tension and compression coupons were
dεp p 2 p p
D= ; dε = ε̇ ε̇ dt (17) used. Some experimental results for dynamic ten-
εf 3 ij ij sion loading are also published, e.g., (FOSTA P477
2000–2002) or (Disse 2004). The basic elastic and
where dt is the time increment and the plastic fail- yield material properties obtained from these
ure strain is an exponential function of the stress sources are summarised in Appendix Table 1.
triaxiality σm /σe , The shape parameter β for the modified Fleck
σm
model can be obtained from the plastic Poisson’s
εf = d1 + d2 e−d3 σe . (18) ratio ν p in a tension test. Using Eq. (16) yields
150 L. Greve, F. Andrieux

 2 
p 1
− β critical strain energy release rate of GIC = 5 kJ/m2
ε1 2 3 3 1 − 2ν p
νp = − p =  2 ⇒ β = √ , leads to a very good agreement of the force-dis-
ε2 1 + νp
1+ β 3
2 placement curves in comparison with the experi-
ments, Fig. 9.
(19) Various test procedures have been proposed in
the literature for testing the shear fracture tough-
For BetamateTM , a νp = 0.23 has been measured, ness of composites and adhesives, e.g., the End
corresponding to β = 1.4. The quasi-static hard- Notch Flexure (ENF) test (ASTM draft 1993). In
ening curve is approximated by a linear function, contrast to Mode I loading using the DCB test,
p
Fig. 6a, the plastic strain rate threshold ε̇T = 0.01, unstable crack propagation and potential overlay
and the strain rate model fitting parameter C = with Mode I loading remain challenging issues for
0.36 can be determined using Eq. (15), Mode II fracture toughness testing. Ongoing re-
Fig. 6b. search indicates that the Mode II critical strain
energy release rate for the considered adhesive
material is significantly higher than under Mode
3.2 Fracture Toughness testing using the double I loading (FOSTA P676 2005–2007). Due to the
cantilever beam lack of existing data, the critical strain energy re-
lease rate under Mode II loading is estimated,
Quasi-static Mode I fracture toughness testing us- GIIC = 2GIC = 10 kJ/m2 .
ing the Double Cantilever Beam test (DCB) is a The strain energy release rates are considered
widely accepted test procedure for measuring the strain rate independent in this study. However, it
critical strain energy release rate GIC for quasi- has to be noted that strain rate dependency has
brittle materials, e.g., composites (ISO DIS15024). been indicated within an ongoing project (FOSTA
More recently, the test procedure has been adopted P676 2005–2007).
for adhesive testing, e.g., (Steinbrecher et al. 2006;
Blackman et al. 2003). The critical strain energy
release rate is input to the interface model only. 3.3 Adhesively butt-joined tubes subjected
Two aluminium beams are adhesively bonded, to combined quasi-static loading
Fig. 7, imposing an adhesive layer thickness of
t = 0.5 mm. According to the test standard an arti- Adhesively butt-joined thin-walled cylindrical
ficial initial crack is introduced by a polyimide film tubes are utilised for the mechanical characterisa-
insert, having a thickness of 12.5 µm. tion of thin adhesive layers under pure and com-
According to the test standard the specimens are bined quasi-static tension and torsion loading,
tested on an Instron 5500R tensile testing machine. Fig. 10.
The applied force p and the crack propagation dis- The nominal stresses and strains are calculated
tance a are monitored and recorded. The critical from the measured forces and moments (F, M) and
strain energy release rate can be estimated by clas- displacements (u, v):
sical beam theory, F 4F  u
σx = =  2  ; εx = ln 1 +
  A π do − d2i t
4p2 3a2 1  
GIC = + , (20) M 16Mdo v
EAl h2 h3 h τxy = =  4  ; γxy = arctan .
W π do − d4i t+u

where EAl = 70 GPa is the Young’s modulus for (21)


aluminium. A Finite Element model of the DCB The tests are performed quasi-statically imposing
test is established, Fig. 8, where the thick alumin- a principal strain rate of 0.001/s on an MTS 809.55
ium beams are represented by solid elements, biaxial tension/torsion test machine. Different
whereas the adhesive layer is discretised using int- strain ratios, R = γxy (2εx ) = const, were investi-
erface elements. The material model parameters gated: R = 0 (pure tension), R = 1/2, R = 2, R = ∞
summarised in Appendix Table 1 are applied. The (pure torsion). The obtained stress–strain curves
Deformation and Failure modelling of high strength adhesives for crash simulation 151

(a) (b)

Fig. 6 Quasistatic hardening curve and strain rate harden- testing (FOSTA P593 2002–2004); (b) approximated log-
ing for Betamate 1496: (a ) quasi-static hardening curve arithmic strain rate mode (yield stresses are obtained from
approximated from uniaxial tension adhesive coupon literature sources Disse 2004)

Fig. 7 Specimen 12.5 35 Artificial crack introduced


dimensions (in mm) of the by film insert Aluminium Adhesive
p
Double Cantilever Beam beams layer
(DCB) specimen for
h=8 a
quasi-static Mode I
adhesive fracture t =0.5*
toughness testing
Width: b = 25
p
150

*) Not drawn to scale

Fig. 8 Finite Element


discretisation of the
Double Cantilever Beam
specimen using interface
elements for the
representation of the
adhesive layer

Fig. 9 Force-displace-
ment curves of Mode I
fracture toughness tests
using the Double
Cantilever Beam
specimen: Comparison of
experimental and
simulation results
152 L. Greve, F. Andrieux

Fig. 10 Combined do = 60
tension/torsion device for
quasi-static testing of thin di = 50
adhesive layers M
(dimensions in mm) Relative
u displacements to
F

v
t= 0.5

Ring-shaped
x
adhesive layer

Aluminium y
tubes

Fig. 11 Adhesively 0.04


butt-joined tubes 0.035 R0 : Exp1
subjected to quasi-static
0.03 R0 : Exp2
pure tensile loading
(R = 0): comparison of 0.025 R0 : Exp3
s x GPa

axial stress–strain curves 0.02 R0 : Exp s Y


x
of experiments and 0.015
simulations R0 : Exp s M
x
0.01
R0 : Sim Solid
0.005
R0 : Sim Interface
0
0 0.05 0.1 0.15 0.2 0.25 0.3
x

Fig. 12 Adhesively 0.04


butt-joined tubes 0.035 R0.5 : Exp1
subjected to quasi-static
0.03
combined loading R0.5 : Exp2
(R = 1/2): comparison of 0.025
s x GPa

R0.5 : Exp s Y Y
x ,t xy
axial stress–strain curves 0.02
of experiments and 0.015 R0.5 : Exp s M
x
simulations
0.01 R0.5 : Sim Solid
0.005
R0.5 : Sim Interface
0
0 0.05 0.1 0.15 0.2 0.25 0.3
x

Fig. 13 Adhesively 0.015


butt-joined tubes R0.5 : Exp1
subjected to quasi-static
combined loading 0.01
R0.5 : Exp2
(R = 1/2): comparison of
t xy GPa

R0.5 : Exp t Y
xy
shear stress–strain curves
of experiments and R0.5 : Exp t M
xy
simulations 0.005
R0.5 : Sim Solid

R0.5 : Sim Interface


0
0 0.1 0.2 0.3 0.4
g xy
Deformation and Failure modelling of high strength adhesives for crash simulation 153

Fig. 14 Adhesively 0.025


butt-joined tubes R2 : Exp1
subjected to quasi-static 0.02
combined loading R2 : Exp2
(R = 2): comparison of 0.015
R2 : Exp s Y

s x GPa
x
axial stress–strain curves
of experiments and 0.01 R2 : Exp s M
x
simulations
R2 : Sim Solid
0.005
R2 : Sim Interface
0
0 0.05 0.1 0.15
x

Fig. 15 Adhesively 0.03


butt-joined tubes R2 : Exp1
subjected to quasi-static 0.025
combined loading 0.02
R2 : Exp2
(R = 2): comparison of
t xy GPa

R2 : Exp t Y
xy
shear stress–strain curves 0.015
of experiments and R2 : Exp t M
xy
simulations 0.01
R2 : Sim Solid
0.005
R2 : Sim Interface
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
g xy

Fig. 16 Adhesively 0.04


butt-joined tubes 0.035 R : Exp1
subjected to quasi-static
0.03
pure shear loading R : Exp2
(R = ∞): comparison of 0.025
t xy GPa

R : Exp t Y
xy
shear stress–strain curves 0.02
of experiments and 0.015 R : Exp t M
xy
simulations
0.01 R : Sim Solid
0.005
R : Sim Interface
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
g xy

are shown in Figs. 11–16. Yield (σxY , τxY ) and maxi- whereas for the solid element model the adhesive
mum (σxM , τxM ) stresses are obtained by curve anal- layer is discretised by four layers of solid elements
ysis. distributed over the thickness, Fig. 17 section (b).

3.3.1 Finite Element modelling of the adhesively 3.3.2 Model parameter identification for the
butt-joined tubes interface model

Finite Element simulation models of the adhesively The quadratic stress failure criterion for interface
butt-joined tubes are established, where only the crack initiation has been discussed in Sect. 2.1.3.
ring-shaped adhesive layer is modelled, whereas The maximum stresses of the adhesively joined tube
the aluminium tubes are considered quasi-rigid and experiments in normal (Mode I) and shear (Mode
are represented by rigid shell layers on top and at II) direction are used to determine the fracture
the bottom of the adhesive layer, Fig. 17. For the locus. Considering σI = σxM and σII = τxy M , the

interface model the Node-To-Element connections stresses are collected in one chart, Fig. 18 and can
are visualised by bar elements, Fig. 17 section (a), be reasonably approximated by the quadratic stress
154 L. Greve, F. Andrieux

Fig. 17 Finite Element


simulation models of the
adhesively butt-joined
aluminium tubes. The
ring-shaped adhesive
layer is represented by:
(a) simple
node-to-element interface
elements visualised by bar
elements; (b) four layers
of solid elements
distributed over the
adhesive layer thickness

failure criterion, Eq. (7). Fitting parameters σImax into the invariant stress space using Eqs. (10) and
and σIImax are determined by a least square fitting (11):
operation and are summarised in Appendix Table 1.
The tube simulations are evaluated using the deter- 
1 + 2κ Y  2  2
mined maximum stress parameters and the critical σm = σx ; σe = (1 − κ)2 σxY + 3 τxy
Y .
strain energy release rates obtained in the previous 3
section, Figs. 11–16. The maximum stresses can be (23)
well predicted whereas, as expected due to the sim-
plified pure elastic constitutive law of the interface The yield points of all adhesively butt-joined tubes,
elements, the plastic domain can not be accurately accompanied by the yield points of the uniaxial ten-
represented. The post-peak domain is character- sion and compression tests, are shown in Fig. 19.
ised by the critical strain energy release rates and The quasi-static yield locus can be well approxi-
the imposed triangular stress–strain relation. mated using Eqs. (12) and (14), where fitting
parameters α and σ0 are determined by a least
3.3.3 Model parameter identification for the solid square fitting method. The yield locus for dynamic
uniaxial tensile loading at ε̇T = 156s−1 is also
p
element model
included in the figure.
Determination of the yield locus parameters Determination of the fracture model parameters
Assuming quasi-homogeneous stress distribution An inverse approach is used for the determina-
and plane strain in the adhesive layer the stress tion of the JC fracture parameters. The different
tensor at the threshold to yielding is FOSTA P593 strain ratios (R = 0, 1/2, 2, ∞) are applied to the
(2002–2004): simulation model. The stress–strain curves of the
butt-joined tubes display a stress maximum after
⎡ ⎤ which the material softens and the stresses gradu-
σxY τxyY 0
⎢ ⎥ νe ally drop, Figs. 11– 16. This softening phase will be
σ = ⎣ τxyY κσxY 0 ⎦ ; κ= , (22)
1 − νe neglected by the Johnson–Cook fracture model.
0 0 κσxY Hence, the maximum stresses are assumed to de-
fine the state of fracture initiation. Each simulation
where σxY and τxyY are the tensile and shear yield is run to this state, where the fracture criterion is
stresses, respectively. For each stress–strain curve being deactivated. At the corresponding state of
in Figs. 11–16 the yield stresses can be transformed failure, the computed equivalent von Mises plastic
Deformation and Failure modelling of high strength adhesives for crash simulation 155

a load cell in the punch records the force, whereas


0.04
the displacement is measured using a laser extens-
ometer. The relative displacement of the punch
0.03
with respect to the large extrusion is measured.
s II GPa

This avoids potential measurement errors intro-


0.02 sM M
x , t xy duced by rotation of the whole set-up due to bend-
s max , s max ing of the mounting plate.
I II
0.01 For dynamic testing an impactor with a weight
Fracture locus
of m = 12.5 kg and an impact velocity of v = 4 m/s
0 is utilised. The force and displacements are com-
–0.02 –0.01 0 0.01 0.02 0.03 0.04 puted by integration of the recorded acceleration
s I GPa
measured in the impactor. In this case the rotation
Fig. 18 Approximation of the quadratic stress failure curve effect of the set-up can not be performed. Hence,
for interface crack initiation: fitting through the maximum the results of quasi-static and dynamic testing can
stresses obtained from the adhesively butt-joined tube tests not be truly compared.
For numerical simulation of the tests, an FE
strains and the corresponding stress triaxialities of model is established, using shell elements for the
the elements close to the inner ring boundary and representation of the extrusions. The thicker sup-
close to the outer ring boundary, which potentially porting parts are discretised by solid elements,
cause fracture initiation, are collected in one chart, Fig. 22. For the simplified modelling approach the
Fig. 20. These data points can be well approximated adhesive layer is represented by interface elements,
by Eq. (18), where the parameters d1 , d2 , and d3 whereas for the detailed modelling approach the
are determined by a least square fit (see Appendix adhesive layer is represented using one layer of
Table1). solid elements.
For confirmation, the simulations are repeated, It has to be noted that no congruent meshes are
now using the determined parameters for the frac- required for both approaches. For the solid element
ture criterion. Good agreement of computed and approach, the connection between adhesive layer
experimental stress–strain relations is observed for and the aluminium extrusions and supporting parts
most of the load cases, Figs. 11–16, including the is achieved by using additional mesh independent
prediction of initial elastic stiffness, yield stress, quasi-rigid tied interfaces without failure criterion,
elasto-plastic deformation, and failure. It has to Fig. 22.
be noted that the predicted initial normal stress Quasi-static loading The experimental force-
peak for the R = 2 simulation, Fig. 14, is signifi- displacement curves are displayed in Fig. 23,
cantly higher than in the experiment. It can be accompanied by the simulation results using the
shown that using a better (non-linear) approxima- interface and the solid element approach. Struc-
tion of the hardening curve reduces this discrep- tural failure of the joined extrusions is governed
ancy. by adhesive failure occurring in a sequence of two
phases. In the first phase, the upper L-shaped sup-
porting part, Fig. 22 (position a), gradually peels
4 Validation testing and simulation off. This phase is characterised by the transition
of a linear elastic response to a strongly nonlinear
For validation and final comparison of the estab- relation of force and displacement in Fig. 23. In the
lished modelling techniques, a more complex second phase the adhesive layer beneath the flat
adhesively joined structure is investigated. Two supporting part, Fig. 22 (position b), fails due to
aluminium extrusions are adhesively bonded, shear loading, characterised in Fig. 23 by decreas-
using one flat and two L-shaped supporting assem- ing load prior to part separation. The experimental
bly parts, Figs. 21 and 22. The large extrusion is force-displacement curves exhibited some scatter,
fixed and the smaller extrusion is subjected to quasi- where the results of both simulation approaches
static and dynamic loading. For quasi-static loading tend toward the experiment having the lowest
156 L. Greve, F. Andrieux

Fig. 19 Approximation 0.08


of the yield locus for the
0.07 Quasi static yield locus
high strength epoxy
adhesive BetamateTM 0.06 Dynamic yield locus
Uniaxial tension
1496 using the enhanced 0.05 Uniaxial compression
DF model

s e GPa
0.04 Symmetry s 0
Yield coupons
0.03 Yield R0
0.02 Yield R1 2
Yield R2
0.01 Yield R
0
0.05 0.025 0. 0.025 0.05
s m GPa

Fig. 20 Approximation
of the Johnson–Cook 1.2
JC Fracture Curve
fracture curve for the high Fracture points outer
1
strength epoxy adhesive Fracture points inner
BetamateTM 1496
0.8

0.6
f

0.4

0.2

–0.25 0 0.25 0.5 0.75 1


sm se

Fig. 21 Sketch Quasi-static or


(dimensions in mm) of the dynamic loading Large extrusion
adhesively bonded
aluminium extrusions 200 100
using two L-shaped and
one flat supporting parts;
all adhesive layers have a Supporting parts
thickness of 0.5mm and a
width of 15 mm Small extrusion

75

Relative
displacement measurement

Mounting of system
12
to solid ground
Mounting plate
Deformation and Failure modelling of high strength adhesives for crash simulation 157

Fig. 22 Finite Element


model and different
modelling approaches for
the adhesively bonded
aluminium extrusions
subjected to quasi-static
and dynamic loading

Fig. 23 Force-displace- 8
ment curves of adhesively Exp 1
joined aluminium
6 Exp 2
extrusions subjected to
quasi-static loading: Exp 3
F kN

comparison of 4
experimental and Sim: Solid
simulation results Sim: Solid
2
Sim: Interface

0
0 2 4 6 8 10
s mm
*) Imposing increased penalty spring stiffness of elastic tied interfaces due to
reduced solution time step

Fig. 24 Experimental 8
force-displacement curves
of an adhesively joined Exp 1 DYN
aluminium extrusions 6
assembly subjected to
F kN

dynamic loading 4 Exp 2 DYN

2 Exp 3 DYN

0
0 2 4 6 8 10 12 14 16 18 20 22 24
s mm

Fig. 25 Force-displace- 14
ment curves of an Interface
12
adhesively joined
aluminium extrusions 10
assembly subjected to Solid w. rate dep.
8
F kN

dynamic loading:
comparison of 6 Solid w o. rate dep.
experimental and
4
simulation results (using
Exp 1 DYN
moving average filter) 2
0
0 2 4 6 8 10 12 14 16 18 20 22 24
s mm
158 L. Greve, F. Andrieux

maximum load, Fig. 23. Both numerical approaches complete adhesive failure. A rebound is observed
predict the correct sequence of failure according to in the simulation at approximately 14 mm displace-
the experiments. ment.
The interface element approach can predict the For the interface approach the higher force level
initial stiffness. However, it can not reproduce the can be explained by the neglect of plastic deforma-
following nonlinear relation of force and displace- tion. For the solid element approach using strain
ment. The predicted response is brittle and the rate hardening, the higher force level could be
maximum force is slightly overrated. Nevertheless, due to the neglect of strain rate dependent frac-
the overall agreement of this strongly simplified ture. Under dynamic loading the adhesive poten-
approach is encouraging. tially embrittles, which is not taken into account
The solid element approach leads to slightly in the simulation model. Interestingly, the solid
lower elastic stiffness prediction in comparison to element approach without strain rate hardening
the experiment. This effect is due to the finite is in very good agreement with the experiments.
stiffness of the additional elastic tied interfaces This would be explained by the assumption that
between the adhesive layer and the aluminium higher dynamic strength is compensated by lower
parts. The stiffness of these elastic spring elements dynamic failure strain.
depends on the solution time step and increases as Analogous to the quasi-static test, the results us-
the time step decreases (PAM-CRASHTM Man- ing the simplified interface model are encouraging.
ual) . A principle study is added in Fig. 23, in which
the solution time step is artificially reduced by fac-
tor 10. Now, the initial elastic stiffness can be well 5 Conclusions
predicted; however, at an expense of a significantly
higher computation time. A new 3 D material model for high strength adhe-
In contrast to the interface approach, the solid sives has been developed and implemented in the
element approach can represent the nonlinear explicit crash code PAM-CRASHTM as a user sub-
force-displacement relation prior to structural fail- routine. It enhances the work by Deshpande and
ure, where the maximum force level is slightly Fleck, incorporates the fracture criterion by
underestimated. Johnson and Cook and covers the specific material
Dynamic loading Good repeatability of the behaviour of thin layered high strength epoxy adhe-
experiments has been observed, Fig. 24. The ini- sive joints, including hydrostatic stress dependent
tial contact force peak is followed by a force drop yield and failure, non-symmetric yield and plastic
after which the force increases again and remains flow, and strain rate hardening. Furthermore, an
level until complete structural failure. The same existing node-to-element interface model has been
sequence of structural adhesive failure is observed investigated, which originally has been developed
for the dynamic and quasi-static tests. for the simulation of delamination of quasi-brittle
The numerical computations using the interface composites. The interface model uses a simplified
and the solid element approach (with and without elastic damaging law including a stress based failure
strain rate dependency) can predict the general criterion for crack initiation and an energy based
trend of the experimental force displacement rela- criterion for crack propagation.
tion, Fig. 25, where all simulations predict slightly The required material model parameters for
amplified dynamic noise in the first phase of the both approaches are obtained from a comprehen-
impact. A potential reason is the lack of physical sive test programme, including uniaxial and multi-
damping in the computation models. axial loading, and fracture toughness testing.
The interface approach and the solid element A validation example, using two adhesively
approach using the strain rate hardening model joined aluminium extrusions subjected to quasi-
predict higher force plateaus after the initial force static and dynamic loading, has been investigated
peak than those observed in the experiments. Due in order to compare the modelling approaches for
to the higher force level the initial kinetic energy situations close to industrial real life conditions.
introduced to the system is insufficient to cause Both approaches could predict the sequence of
Deformation and Failure modelling of high strength adhesives for crash simulation 159

adhesive failure observed in the experiments. The case involving not only a sequence of separated
force-displacement curves of experiments and sim- mode I (peeling) and mode II (shearing) failure,
ulations were in fair agreement. The detailed solid but also a combination of these failure modes. To
element modelling approach performed slightly improve the numerical prediction of the behav-
better than the simplified interface model; nev- iour of the adhesive layer after the maximum load,
ertheless, the good agreement of the simplified the material model for the solid element modelling
interface model was encouraging for industrial should be extended by coupling the stress tensor
application. Both modelling approaches are con- with the damage variable as usually dealt with in
sidered being useful tools for numerical crashwor- classical damage mechanics.
thiness analysis of adhesive failure.
Acknowledgements The experimental analysis of the
adhesively butt-joined tubes by Prof. Schlimmer and co-
6 Future work workers from Kassel University is acknowledged as well
as the work by Prof. Pickett and co-workers at Cranfield
The potential existence of rate dependent frac- University for the evaluation of the fracture toughness
ture must be clarified in the future, which requires and joined extrusion tests. Furthermore, the authors
acknowledge the work of the different partners of the
advanced dynamic strain measurement test meth- FOSTA project “Methodenentwicklung zur Berechnung
ods. Furthermore, it would be interesting to per- von höherfesten Stahlklebverbindungen des Fahrzeug-
form another validation test for a different loading baus unter Crashbelastung” (FOSTA P676 2005–2007).

Appendix

Table 1 Collection of elastic, quasi-static/dynamic yield, fracture, and fracture toughness properties of the high strength
adhesive BetamateTM 1496.
Denomination Symbol Value Source Model*
S I

Elastic modulus E 1.82 GPa [FOSTA p593] x x


Shear modulus G 0.625 GPa [FOSTA p593] x x
Elastic Poisson’s ratio νe 0.4 [FOSTA p593] x –
Plastic Poisson’s ratio νp 0.23 [FOSTA p593] x –
Shape parameter of yield potential α 1.335 – x –
Shift stress of yield surface σ0 –0.026 GPa – x –
Shape parameter of flow potential function β 1.4 – x –
Uniaxial quasi–static tensile yield strength σTsta 0.032 GPa [FOSTA p593] x x
Uniaxial quasi–static compression yield strength σCsta  1.6 σTsta [FOSTA p593], [Disse 2004] x –
σT ε̇T → 156s−1
p
Uniaxial dynamic tensile yield strength 2 σTsta [Disse 2004] x –
0.01ms−1
p
Plastic Strain rate threshold ε̇T – x –
Strain rate model parameter C 0.36 – x –
JC fracture curve parameters d1 0.054 – x –
d2 0.356 – x –
d3 5.92 – x –
Mode I critical strain energy release rate GIC 5 kJ/m2 – – x
Mode II critical strain energy release rate GIIC 10 kJ/m2 estim. – x
Mode I strength σImax 0.032 GPa – – x
Mode II strength σIImax 0.034 GPa – – x

* Model parameters required for solid element model (S), or interface model (I).
160 L. Greve, F. Andrieux

References Hillerborg A, Modeer M, Petersson PE (1976) Analysis


of crack formation and crack growth in concrete by
ABAQUSTM Manual, http://abaqus.com means of fracture mechanics and finite elements. Ce-
ASTM draft standard D30.06. (1993) Protocol for inter- ment Concrete Res 6:773–782
laminar fracture testing, End-Notched Flexure (ENF) ISO international Standard DIS15024. Fibre-reinforced
Audi TT (2006) Automobil Industrie, special issue plastic composites—Determination of Mode I inter-
Blackman BRK, Kinloch AJ, Paraschi M, Teo WS (2003) laminar fracture toughness, GIC, for unidirectionally
Measuring the mode I adhesive fracture energy, reinforced materials.
GIC, of structural adhesive joints: the results of an Johnson AF, Pickett AK, Rozycki P (2001) Computational
international round-robin. Int J Adhesion Adhesives methods for predicting impact damage in composite
23(4):293–305 structures. Compos Sci Tech 61(15):2183–2192
BetamateTM 1496: Technical data sheet http://automotive. Johnson GR, Cook WH (1985) Fracture characteristics of
dow.com/materials/products/betamate/bmate96.htm three metals subjected to various strain, strain rates,
Camanho PP, Davila CG, Ambur DR (2001) Numerical temperatures and pressures. Eng Fract Mech 21(1):31–
Simulation of Delamination Growth in Composite 48
Materials. NASA Langley Research Center, Hampton Mahnken R, Schlimmer M (2005) Simulation of strength
Virginia, NASA TP-2001-211041 difference in elasto-plasticity for adhesive materials.
Camanho PP, Matthews FL (1999) Delamination onset pre- Int J Numer Method Eng 63(10):1461–1477
diction in mechanically fastened joints in composite Memhard D, Andrieux F, Sun D-Z, Feucht M, Frank
laminates. J Compos Mater 33:906–927 T, Kolling S (2005) Entwicklung und Anwendung
Deshpande VS, Fleck NA (2000). Isotropic constitutive von Ersatzmodellen für die Anwendung von Kleb-
models for metallic foams. J Mech Phys Solids 48:1253– verbindungen unter Crashbelastung. 4. LS-DYNA An-
1283 wenderforum, Bamberg 2005
Disse T (2004) Untersuchungen zum geschwindigkeit- PAM-CRASHTM Manual Engineering System Interna-
sabhängigen Festigkeits- und Verformungsverhal- tional, www.esi-group.com
ten von Klebverbindungen. Dissertation, Universität Sato C, Ikegami K (1999) Strength of adhesively-bonded
Paderborn, Shaker Verlag butt joints of tubes subjected to combined high-rate
Drucker DC, Prager W (1952) Soil mechanics and plas- loads. J Adhesion 70:57–73
tic analysis in limit design. Quarterly Appl Math Schlimmer M (1974) Fließverhalten plastisch kompressi-
10(2):157–165 bler Werkstoffe. Dissertation, Technische Hochschule
FOSTA P477 (2000-2002) Untersuchungen zum Crash- Aachen
verhalten geklebter und hybridgefügter Stahlblech- Steinbrecher G, Buchman A, Sidess A, Sherman D. (2006)
verbindungen Characterization of the mode I fracture energy of adhe-
FOSTA project P593 (2002-2004) Methodenentwick- sive joints”, International Journal of Adhesion and
lung zur Berechnung und Auslegung geklebter Adhesives 26(8):644–650.
Stahlbauteile für den Fahrzeugbau von Mises R (1928) Mechanik der plastischen Formande-
FOSTA Project P676 (2005-2007) Methodenentwicklung rung von Kristallen. ZAMM 8:161–185
zur Berechnung von höherfesten Stahlklebverbindun-
gen des Fahrzeugbaus unter Crashbelastung
Greve L, Pickett AK (2005) Delamination testing and mod-
elling for composite crash simulation, Composites Sci-
ence and Technology 66(6):816–826

You might also like