Computational Mechanics 31 (2003) 69–77 Ó Springer-Verlag 2003

DOI 10.1007/s00466-002-0394-z

A cohesive segments method for the simulation of crack growth

J. J. C. Remmers, R. de Borst, A. Needleman

69
Abstract A numerical method for crack growth is de- cracks in another phase or by propagating across phase
scribed in which the crack is not regarded as a single boundaries, e.g. [1]. Another example is the transition
discontinuity that propagates continuously. Instead, the from subsonic to intersonic crack speeds via the nucle-
crack is represented by a set of overlapping cohesive ation of a micro-crack ahead of the main crack, [2]. Also,
segments. These cohesive segments are inserted into finite in quasi-brittle materials micro-cracking in front of the
elements as discontinuities in the displacement field by main crack tip plays a key role in setting the fracture
exploiting the partition-of-unity property of shape func- toughness [3]. Hence, a need for analysing discontinuous
tions. The cohesive segments can be incorporated at ar- crack growth arises in a wide variety of contexts.
bitrary locations and orientations and are not tied to any A cohesive surface methodology has emerged which
particular mesh direction. The evolution of decohesion of permits the analysis of fracture processes in which there is
the segments is governed by a cohesive law. The inde- no dominant flaw. In fact, an initial crack-like defect is not
pendent specification of bulk and cohesive constitutive required since crack nucleation can occur naturally during
relations leads to a characteristic length being introduced the loading history. The basic assumption of the cohesive
into the formulation. The formulation permits both crack surface framework is that the separation process is con-
nucleation and discontinuous crack growth to be mod- fined to a set of discrete planes (or lines in a two-dimen-
elled. The implementation is outlined and some numerical sional context). A constitutive relation is then specified for
examples are presented. each cohesive surface that allows separation to occur.
The cohesive approach to fracture was pioneered by
Keywords Crack growth, Fracture, Cohesive zones, Barenblatt [4], Dugdale [5] and Hillerborg et al. [6]. In
Partitions of unity these formulations, a dominant flaw was assumed present
as in conventional engineering fracture mechanics, but a
1 cohesive zone was introduced ahead of the existing crack
Introduction tip. The relation between the work expended in this co-
In conventional engineering fracture mechanics, crack hesive zone and that in the crack tip field is typically such
growth is assumed to occur by the extension of a single that the stress singularity is cancelled and the near tip
dominant crack. However, there are a wide variety of stresses are finite. The slip weakening model of Andrews
circumstances where the fracture process involves the [7] introduced in the geophysics literature did not require
nucleation and growth of multiple crack-like flaws. For crack growth to be continuous. The cohesive framework
example, in heterogeneous materials, multiple cracks that was extended to finite deformations and to situations
initiate and grow in one phase may link up by nucleating without an initial crack in [8]. Subsequently, a wide variety
of fracture phenomena have been analysed using the
cohesive surface methodology.
In a cohesive surface formulation, constitutive relations
J. J. C. Remmers (&), R. de Borst are specified independently for the bulk material and for
Faculty of Aerospace Engineering, Koiter Institute Delft, one or more cohesive surfaces, see [8, 9]. The cohesive
Delft University of Technology, PO Box 5058, constitutive relation embodies the failure characteristics of
2600 GB, Delft, The Netherlands the material and characterises the separation process. The
e-mail: [email protected]
bulk and cohesive constitutive relations together with
A. Needleman appropriate balance laws and boundary (and initial) con-
Division of Engineering, Brown University, ditions completely specify the problem. Fracture, if it takes
Providence, Rhode Island 02912, USA place, emerges as a natural outcome of the deformation
process without introducing any additional failure crite-
Dedicated to the memory of Prof. Mike Crisfield, for his rion. The simplest cohesive constitutive relation is one
cheerfulness and cooperation as a colleague and friend over many where the cohesive surface traction is a function of the
years.
displacement jump across the cohesive surface. Such a
The authors wish to express their thanks to Erik-Jan Lingen for cohesive constitutive relation incorporates as parameters
his help in the implementation of the model in the JIVE finite the strength ft and the work of separation (or fracture
element toolbox. AN is grateful for support from the Office of energy) Gc . From dimensional considerations, this intro-
Naval Research through grant N00014-97-1-0179. duces a characteristic length.
 When fracture takes place along well-defined interfaces of the continuum element [20, 24–27]. The magnitude of
as, for example, in a lamellar solid, the placement of co- the displacement jump is determined by additional de-
hesive surfaces is clear. Also when the crack path is known grees of freedom which are added at the existing nodes. A
in advance from experiments, accurate predictions can be key feature of the method is the possibility of extending
obtained, since interface elements can be placed along the the cohesive crack during the calculation in an arbitrary
known crack path [10]. However, for a solid that is ho- direction, independent of the structure of the underlying
mogeneous on the scale modelled, the placement of cohe- finite element mesh. When cohesive surfaces are added
sive surfaces is problematic. In Xu and Needleman [9] during the calculation, there is no need for a high initial
multiple cohesive surfaces were used and in their calcula- stiffness to minimise the effect of increasing the initial
tions cohesive interface elements were placed along all compliance of the medium due to the presence of cohesive
finite element mesh boundaries. Although this approach surface elements. As a consequence, numerical anomalies,
70
simplifies the simulation of complex crack phenomena such as stress oscillations at the interface [28] or spurious
such as crack branching and crack initiation away from a stress wave reflections are avoided, but other anomalies
main crack tip, it is not mesh independent. In fact, since the may be introduced in dynamic problems [12, 13].
interface elements are aligned with the element boundaries, A drawback of the current version of partition-of-unity
the orientation of cracks is restricted to a limited number of based cohesive zone numerical methods is that the crack is
predefined angles. In addition, if, as in [9], the cohesive regarded as a single entity. Crack propagation is modelled
surfaces are taken to have a nonzero initial compliance, the by extending the current displacement jump. Thus, con-
presence of the cohesive surfaces contributes to the overall tinuous growth of a crack as in the original cohesive zone
compliance of the body. Then, if cohesive surfaces are formulations of Barenblatt [4], Dugdale [5] and Hillerborg
added between all elements as the computational mesh is et al. [6] can be modelled, but not discontinuous crack
refined, the overall compliance depends on the mesh and growth involving crack initiation at multiple locations and
an ill-posed problem results. A well-posed problem is ob- the subsequent growth and coalescence of the nucleated
tained if a mesh-independent cohesive surface spacing is cracks.
used, but it is unclear how to set that spacing for homo- Here, we develop a cohesive finite element method
geneous solids. To overcome limitations associated with based on incorporating segments of cohesive surfaces into
initially compliant cohesive surfaces, Camacho and Ortiz continuum finite elements that is applicable when crack
[11] used initially rigid cohesive surfaces in conjunction growth is discontinuous. The crack is not regarded as a
with adaptive mesh refinement. The use of initially rigid single entity. Instead, it is modelled as a collection of
cohesive surfaces introduces other difficulties, at least in overlapping cohesive segments, which are added as dis-
modelling dynamic crack growth [12, 13]. placement jumps by using the partition-of-unity property
For a certain class of problems, an alternative is to adopt of finite element shape functions. A combination of
a smeared crack approach, in which the separation energy overlapping crack segments can behave as a continuous
Gc is distributed over the element width, e.g. [10, 14–16]. crack. In addition, since crack segments can be added at
Finite element methods with embedded discontinuities arbitrary positions and with arbitrary orientations, the
provide a means of implementing smeared cohesive models method allows for complex crack patterns including the
[17, 18]. The embedded discontinuity approaches enhance simulation of crack nucleation at multiple locations, fol-
the deformational capabilities of the elements, especially lowed by growth and coalescence. Branching of an existing
when the standard Bubnov-Galerkin approach is replaced crack is also allowed for.
by a Petrov-Galerkin method, which properly incorporates We begin with a short description of the approach.
the discontinuity kinematics [19]. The high local strain Then, the underlying kinematic relations for a domain
gradients inside localisation bands are captured more ac- with multiple displacement jumps is discussed with
curately at the expense of obtaining a nonsymmetric stiff- attention restricted to a small deformation formulation.
ness matrix. However, a true discontinuity is not obtained Next, some details regarding the implementation are giv-
because the kinematics of the embedded localisation band en. The paper concludes with some numerical examples.
are diffused over the element when the governing equations
are cast in a weak format, either via a Bubnov-Galerkin or 2
via a Petrov-Galerkin procedure. Several authors [20, 21] Cohesive segments model
have proved the equivalence between embedded disconti- The physics of crack initiation and crack growth in a
nuity approaches and classical smeared-crack models in heterogeneous quasi-brittle material is illustrated in Fig. 1
which the separation energy is smeared out over the ele- [1] which shows a concrete specimen loaded in tension.
ment width. Accordingly, the embedded discontinuity ap- The heterogeneity of the material, i.e. the presence of
proaches inherit many of the disadvantages of conventional particles of different sizes and stiffnesses, leads to a
smeared-crack models, including the sensitivity of crack complex stress field in which new cracks nucleate (‘‘a’’ in
propagation to the direction of the mesh lines. Fig. 1) and existing cracks branch (‘‘b’’ in Fig. 1). Smeared
There are advantages to incorporating the cohesive (cohesive-zone) models are not able to capture these
surfaces (or zones) into continuum finite elements by processes of crack initiation, growth, coalescence and
using the partition-of-unity property of finite element branching properly, since essential characteristics are lost
shape functions [22] in conjunction with a discontinuous in the averaging process.
mode incorporated at the element level [23]. The cohesive The cohesive segments approach can, at least in prin-
zone is then modelled as a jump in the displacement field ciple, describe the physical processes observed in Fig. 1.
 3
Kinematic relations
The key feature of the cohesive segments approach is the
possible emergence of multiple cracks in a domain. Con-
sider the domain X with boundary C as shown in Fig. 2. It
contains m discontinuities Cd;j , where j ¼ 1; m. Each dis-
continuity splits the domain in two parts, which are de-
noted as X þ
j and Xj . For all discontinuities the following
relation must hold:
X þ
j [ Xj ¼ X 8 j ¼ 1; m : ð2Þ
71
The displacement field in the domain X consists of a
continuous regular displacement field u ^ plus m additional
continuous displacement fields u
~j , cf. [29]:
X
m
uðx; tÞ ¼ u
^ðx; tÞ þ HCd;j ðxÞ~
uj ðx; tÞ ; ð3Þ
j¼1

where x denotes the position of a material point, t is time

and HCd;j are Heaviside step functions, defined as:

0 if x 2 X
j ,
HCd;j ðxÞ ¼ ð4Þ
1 if x 2 Xþ
j .
Fig. 1. Experimentally observed ‘diffuse’ crack pattern [1] and
possible representation with cohesive segments The strain field in the bulk can be found by taking the
derivative of the displacement field, Eq. (3):
Although a wide variety of cohesive constitutive relations X
m
can be incorporated into the cohesive segments frame- ðx; tÞ ¼ rs u
^ðx; tÞ þ HCd;j ðxÞrs u
~j ðx; tÞ ; ð5Þ
work, in the computations here no cohesive segments are j¼1
present at the beginning of a calculation which means that
an initially rigid cohesive constitutive relation is assumed. where a superscript s denotes the symmetric part of a
When decohesion initiates, a cohesive segment is inserted differential operator. Note that at the discontinuities Cd;j ,
through the integration point. The segment is taken to the strains are not defined. There, the magnitude of the
extend throughout the element to which the integration displacement jump
point belongs and into the neighbouring elements, see vj ðx; tÞ ¼ u
~j ðx; tÞ x on Cd;j ; ð6Þ
Fig. 3. The evolution of the crack segment is governed by a
decohesion constitutive relation. Thus, we need to specify: is the relevant kinematic quantity. Although the formula-
(i) a criterion for the initiation of decohesion; (ii) a cri- tion can be applied to finite deformations [25, 26], we
terion for the orientation of the added cohesive segment; restrict attention to small displacement gradients.
and (iii) the decohesion constitutive relation.
In the illustrative examples here only normal (mode-I) 4
decohesion is considered. The initiation of decohesion is Equilibrium equations
taken to occur when the maximum principal stress at a Consider the quasi-static equilibrium equations without
finite element integration point exceeds the cohesive body forces and the corresponding boundary conditions:
strength ft . The orientation of the cohesive surface is rr¼0 x2X ; ð7Þ
specified as being normal to the maximum principal stress
direction. The decohesion constitutive relation is taken to
nt  r ¼ t x 2 Ct ; ð8Þ
have the exponential form, u¼u  x 2 Cu ; ð9Þ


ft nd;j  r ¼ tj x 2 Cd;j ; ð10Þ
tn ¼ ft exp  j ; ð1Þ
Gc
where tn is the normal traction across the cohesive surface,
Gc is the work of separation and j, which has dimensions
of reciprocal length, is a specified cohesive parameter.
Since only mode-I separation is considered, the shear
traction across the cohesive surface vanishes. The cohesive
segments are inserted into existing finite elements by ex-
ploiting the partition-of-unity property of their shape
functions, thus ensuring that the discrete character of the
cohesive segment is preserved in the discretisation pro-
cess. Note that the orientation of the cohesive segments is Fig. 2. Domain X crossed by two discontinuities, Cd;1 and Cd;2
not tied to any direction associated with the discretisation. (dashed lines)
 where r is the Cauchy stress in the bulk material, t are the X
m
prescribed tractions on Ct with outward normal vector nt , u ¼ Na þ HCd;j Nbj ; ð17Þ
 are the prescribed displacements on Cu and tj are the
u j¼1
tractions at discontinuity Cd;j . The normal nd;j points from where the vector a contains the regular nodal degrees of
X þ
j to Xj . Equilibrium can be expressed in a weak form by freedom of the element and bj contains all additional nodal
multiplication with an admissible variational displacement degrees of freedom associated with discontinuity Cd;j . The
field g: matrix N contains the conventional element shape func-
Z
tions. The discretised strain field can be derived by
g  ðr  rÞdX ¼ 0 : ð11Þ straightforward differentiation:
X X
m
72 e ¼ Ba þ HCd;j Bbj ; ð18Þ
Taking the space of the admissible variations to be the
j¼1
same as the actual displacement field, Eq. (3), the varia-
tions of the displacements can be decomposed as: where B ¼ LN contains the spatial derivatives of the ele-
X
m ment shape functions. L is a differential operator matrix,
g¼g
^þ HCd;j g
~j : ð12Þ which, for two-dimensional elements, is:
2 o
3
j¼1
ox 0
6 o 7
Substituting the variations into Eq. (11) gives: L¼40 oy 5 : ð19Þ
Z m Z
X
o o
oy ox
^  ðr  rÞdX þ
g HCd;j g
~j  ðr  rÞdX ¼ 0 :
Finally, the discrete displacement jump at discontinuity
j¼1
X X Cd;j , see Eq. (6), is equal to:
ð13Þ vj ¼ Nbj : ð20Þ
We now apply Gauss’ theorem, use the symmetry of the In the spirit of a Bubnov-Galerkin approach, the variations
Cauchy stress tensor, eliminate the Heaviside functions in of the displacement fields can be discretised as:
the bulk by changing the integration domain and use the
^ ¼ Nda
g ~j ¼ Ndbj
g
boundary conditions at the external boundary Ct and at ð21Þ
the discontinuity planes Cd;j to give: r g ¼ Bda rs g
s^ ~j ¼ Bdbj :
Z Xm Z Xm Z Inserting them into the weak form of the equilibrium
s s
rg ^ : r dX þ rg ~j : r dX þ ~j  tj dC equation (10) yields:
g
j¼1 þ j¼1 Z X m Z m Z
X
X Xj Cd;j
ðBdaÞT rdX þ ðBdbj ÞT rdX þ ðNdbj ÞT tj dC
Z m Z
X j¼1 j¼1
X Xþ Cd;j
¼ ^  t dC þ
g ~j  t dC :
HCd; j g ð14Þ j
Z m Z
X
j¼1
Ct Ct
¼ ðNdaÞTt dC þ HCd; j ðNdbj ÞTt dC : ð22Þ
j¼1
Ct Ct
5
Finite element formulation By taking all the variations da and dbj respectively, the
For the finite element formulation, we use the partition-of- weak equilibrium equations can be separated in a set of
unity property of finite element shape functions [22]. A m þ 1 equilibrium equations:
Z Z
collection of functions /i , associated with node i T
(1 i n) form a partition of unity if: B r dX ¼ NTt dC ;
X n X Ct
Z Z Z
/i ¼ 1 : ð15Þ T T
i¼1
B r dX þ N t1 dC ¼ HCd;1 NTt dC ;
þ Cd;1 Ct
For any set of functions that satisfy this equation, a field u XZ1 Z Z
can be interpolated as follows:
! BT r dX þ NT t2 dC ¼ HCd;2 NTt dC ; ð23Þ
X
n X
m
uðx; tÞ ¼ /i ðxÞ ai ðtÞ þ bj ðxÞbij ðtÞ ; ð16Þ Xþ
2
Cd;2 Ct

i¼1 j¼1 ..
.
with ai ðtÞ the regular nodal degrees-of-freedom, bj ðxÞ the Z Z Z
enhanced basis terms and bij the additional degrees of BT r dX þ NT tm dC ¼ HCd;m NTt dC :
freedom at node i, representing the amplitude of the jth
Xþ Cd;m Ct
basis term bj . The displacement fields HCd;j u
~j can be con- m

sidered as enhanced basis fields. By replacing bj ðxÞ by the The equilibrium equation that is related to the regular de-
corresponding Heaviside function HCd;j , we can cast the grees of freedom is identical to the equilibrium equation for
displacement field in Eq. (3) in the following discrete form: an element without a discontinuity. Therefore, it is possible
to add a discontinuity to an element during the calculations The internal forces are given by:
with a minimal effort by adding the additional equilibrium Z
relations and the corresponding degrees of freedom bj . f int
a ¼ BT r dX ;
X
5.1 Z Z ð28Þ
Constitutive relations f int
bj ¼ BT
r dX þ NT tj dC :
The stress rate in the bulk material r_ is a function of the
strain rate e_ and can be written as, see also Eq. (18): Xþ
j
Cd;j
!
Xm Finally, the expression for the external forces is:
r_ ¼ De_ ¼ D B_a þ _
HCd;j Bbj ; ð24Þ Z
ext
j¼1 fa ¼ NTt dC ; 73
where ð_Þ denotes oð Þ=ot and D is the tangent stiffness Ct
Z ð29Þ
matrix of the bulk material (rate independent material
behaviour is assumed here). The traction rates t_ j at the jth f ext
bj ¼ HCd;j NTt dC :
discontinuity can be expressed in terms of the corre- Ct
sponding enhanced nodal velocities v_ j , Eq. (20):
Note that if the tangent matrices D and T are symmetric,
t_ j ¼ Tv_ j ¼ TNb_ j ; ð25Þ symmetry of the submatrices Kaa , Kabj and Kbj bk is
preserved. Consequently, the total stiffness matrix also
where T is the tangent stiffness of the traction-separation remains symmetric.
law at the discontinuity. The latter relations are defined in
a local frame of reference, aligned with the discontinuity. 6
Therefore, they must be transformed into the element local Implementation
frame of reference. The procedure has been implemented using a four-node
quadrilateral continuum finite element. A new cohesive
5.2 segment is added when the major principal stress at an
Linearisation of the equilibrium equations integration point within an element reaches the cohesive
The deformation history is calculated in an incremental strength ft in Eq. (1). The added cohesive segment passes
fashion. At each time step the rate equilibrium equations through the integration point and extends through the
are differentiated with respect to the displacement vari- entire element and into the neighbouring elements to the
ables a and bj . Differentiating the rate form of the dis- boundary of a patch of elements influenced by the added
cretised equilibrium equations (23) leads to degrees of freedom, see Fig. 3. Various criteria can be used
2 32 3
Kaa Kab1 . . . Kabm a_ to determine the direction of the cohesive segment, but
6 Kab Kb b . . . Kb b 76 b_ 7 here the cohesive segment is taken to be normal to the
6 1 1 1 1 m 76 1 7
major principal stress direction at the integration point.
6 76 . 7
6 . . . . .
.. 7 6 7 Since the cohesive segment is taken to be a straight line,
4 . .. .. 54 .. 5
the normal vector nd;j is constant along the patch of ele-
Kabm Kb1 bm . . . Kbm bm b_ m ments. The magnitude of the displacement jump of the
2 ext 3 2 int 3 segment is governed by a set of additional degrees of
fa fa
6 ext 7 6 int 7 freedom, which are added to all four nodes of the central
6 f b1 7 6 f b1 7
6 7 6 7
¼6 . 76 . 7 ; ð26Þ
6 .. 7 6 .. 7
4 5 4 5
f ext
bm f int
bm

where the terms in the stiffness matrix are:

Z
Kaa ¼ BT DB dX ;
X
Z
Kabj ¼ BT DB dX ;
Xþ
j
Z Z ð27Þ
Kbj bj ¼ BT DB dX þ NT TN dC ;
Xþ Cd;j Fig. 3. A single cohesive segment in a quadrilateral mesh. The
j
Z segment passes through an integration point () where the co-
hesive strength is attained. The dark nodes contain additional
Kbj bk ¼ BT DB dX if j 6¼ k : degrees of freedom bj that determine the magnitude of the dis-
Xþ \ Xþ placement jump. The gray shade denotes the elements that belong
j k
to the patch that is influenced by the cohesive segment
 element of the patch. Since the nodes at the edge of the candidate is Simpson’s rule. Although the error with this
patch do not contain additional degrees of freedom, the integration scheme is greater than that associated with the
displacement jump at the edge vanishes, thus ensuring a first alternative the effect on the global accuracy of the
zero opening at the tip of the cohesive segment [24]. computation will generally be small, since these elements
Since the additional degrees of freedom cannot be are usually nearly stress free.
condensed at an element level, they influence all the
surrounding elements in the patch (denoted by the gray 7
shade). For the elements that belong to the patch, but do Numerical examples
not contain a cohesive segment, the additional displace- Some features of the cohesive segments method are illus-
ment u ~ j is not premultiplied with a Heaviside function, but trated in a few simple problems. The material behaviour is
with a value, which is zero when the element belongs to the taken to be described by isotropic elasticity and only
74
Xj part of the domain, and equals one when the element is mode-I (tensile) separation is considered. However, the
in the Xþ j part. cohesive surfaces formulation places no restriction on the
A key feature of the approach is the possibility of having material constitutive relation and allows for mixed mode
overlapping cohesive segments. In the situation sketched crack growth. A double-cantilever beam with an initial
in Fig. 4, a new cohesive segment is added next to an notch with length 1 mm as shown in Fig. 7 [30] is con-
existing segment. The displacement jump of the new co- sidered. The beam is loaded by peel forces P. The two
hesive segment is supported by an additional set of de- layers of the beam have identical elastic properties:
grees of freedom, and is added to the nodes of the central Young’s modulus E ¼ 100 N/mm2 and Poisson’s ratio
element of the new segment (denoted by black squares). m ¼ 0:3. The tensile strength of the adhesive is ft ¼ 1:0
This new set has no relation to the additional degrees of N/mm2 and the work of separation is Gc ¼ 0:1 N/mm. It is
freedom of the first segment. The hatched elements are
elements in the patches that belong to both segments. In
these elements two additional sets of degrees of freedom
are present, although each set of degrees of freedom is
added to a different set of nodes.
Another possible configuration is shown in Fig. 5 where
two segments cross to form a cohesive zone with a sharp
bend. Here, the patches overlap in such a way that one
node contains additional sets of degrees of freedom as-
sociated with both segments. Again, this configuration will
act as a single cohesive zone on a global level.
Numerical integration of an element that contains one
or more discontinuities is not trivial. Since the displace-
ment fields are only piecewise continuous, standard inte-
gration techniques are no longer accurate. One alternative
is to divide the element into subdomains, see Fig. 6. Since
the displacement field is continuous within each subdo-
main, the integrals over the subdomain can be evaluated Fig. 5. Crossing of two cohesive segments. The lighter shade of
numerically with standard integration techniques. A sec- gray denotes the patch of elements to the left hand cohesive
ond possibility is to integrate such an element using a segment. The dark circles the nodes that contains the enhanced
large number of fixed integration points. A possible degrees of freedom. The darker shade of gray and the black
squares denote the elements and the nodes that belong to the
right-hand segment. The dark triangle denotes the node that
contains additional degrees of freedom of both segments

Fig. 4. Interaction of two cohesive segments. The hatched ele-

ments have two sets of additional degrees of freedom. The dark Fig. 6. Numerical integration of an element with two disconti-
circles denote the nodes that contain the additional degrees of nuities (heavy lines). The subdomains (denoted with dashed
freedom for the segment on the left and the dark squares denotes lines) are integrated with a Gauss scheme with one integration
those on the right point (þ) each
 75

Fig. 7. Double-cantilever beam with an initial notch subjected to

peel forces P

assumed that the fracture mode is purely mode-I, the shear

traction is identically zero and the decohesion relation is Fig. 9. Final deformation of double-cantilever beam. Note that
the deformations are not magnified
given by Eq. (1), which can be differentiated to give the
material tangent matrix:
" 2   # the crack tip in this example. To capture the high peak
ft ft
T¼  G c
exp  G c
j 0 : ð30Þ stresses accurately, a rather fine mesh is needed.
0 0 In order to demonstrate the ability to simulate discon-
tinuous crack growth, the previous example is slightly
The specimen is analysed with a mesh having 200  7 el- modified. A double-cantilever beam is now considered that
ements. The initial notch is modelled as a series of is identical to the one in Fig. 7 except for a small cavity
overlapping traction-free cohesive segments which are with length 0:25 mm and height 0:14 mm at a distance
added to the mesh beforehand. 2 mm from the loaded edge, see Fig. 10. During loading,
Since the adhesive is relatively weak, the crack growth the crack at the initial notch propagates into the cavity.
trajectory is known, i.e. the interface between the two Upon further loading, a new crack nucleates at the oppo-
beams. Therefore, we have only monitored the normal site side of the cavity and continues along the weak in-
stresses in the y-direction at the interface as the new terface. This transition is indicated by the steep jump in
segment is always aligned with the x-axis. The results are the load–displacement curve in Fig. 11. The position of
shown in Fig. 8 which gives the load–displacement curve cohesive segments immediately after the nucleation of the
and in Fig. 9 which gives the deformed specimen at the second crack, position (A) in the load-displacement curve,
final load step, when u ¼ 6 mm. The cohesive segments is shown in Fig. 12.
approach gives results that are nearly identical to those The third example illustrates crack nucleation without
obtained with a method in [30] where a continuous crack any initial crack. The specimen is now clamped at both
is modelled using the partition-of-unity property of finite sides and has no initial crack, see Fig. 13. The structure is
element shape functions. The areas below the curve, which loaded by forces P. In order to preserve symmetry of crack
are a measure of the energy dissipation, are virtually the growth, the specimen is modelled with an odd number of
same for the two calculations. It is noted that there is a elements (199 in the calculation here) in the x direction.
high stress concentration in a relatively small area around When P  0:45 N, a small crack initiates in the centre of
the beam. Due to the size of the discrete load steps, this
initial crack consists of three overlapping cohesive seg-
ments, which nucleate at the same load step. The crack
propagates in both directions with the same velocity. The
load increases until the external load is 1:2 N, see Fig. 14.

Fig. 8. Load–displacement curve of double-cantilever beam as

calculated using the cohesive segments method and a continuous
partition of unity method (PUM) [30] Fig. 10. Geometry of double-cantilever beam with a small cavity
76

Fig. 11. Load–displacement curve for double-cantilever beam

with cavity Fig. 14. Load–displacement curve of double-cantilever beam
under central lateral loading

Fig. 15. Deformation of double-cantilever beam under central

lateral loading

the partition-of-unity property of finite element shape

functions. Nevertheless, the actual implementation of the
method can be somewhat elaborate from a bookkeeping
point of view, since a single element can be crossed by
Fig. 12. Position of cohesive segments, denoted by the heavy multiple cracks, each with its own additional degrees of
lines, in the double-cantilever beam with cavity. The individual freedom.
segments are slightly shifted horizontally for the sake of com- Some capabilities of the cohesive segments method have
prehensibility been illustrated by simple two dimensional examples. The
extension of the formulation to more complex bulk and
cohesive constitutive relations and to three dimensions is
straightforward in principle. The ability of the method to
capture complex crack patterns accurately, such as shown
in Fig. 1, remains to be demonstrated. However, the initial
studies here suggest that the cohesive segments method
provides a promising approach for modelling complex
fracture behaviour.

References
1. van Mier JGM (1997) Fracture Processes of Concrete. CRC
Press, Boca Raton, Florida
Fig. 13. Double-cantilever beam under central lateral loading 2. Coker D, Rosakis AJ, Needleman A (2003) Dynamic crack
growth along a polymer composite-homalite interface.
J. Mech. Phys. Solids 51: 425–460
Subsequently, crack growth continues under decreasing 3. Cotterell B, Mai YW (1996) Fracture Mechanics of Cementi-
load. Figure 15 shows the deformed specimen at the final tious Materials. Blackie, Glasgow, Scotland
load step. 4. Barenblatt GI (1962) The mathematical theory of equilibrium
cracks in brittle fracture. Adv. Appl. Mech. 7: 55–129
5. Dugdale DS (1960) Yielding of steel sheets containing slits.
8 J. Mech. Phys. Solids 8: 100–108
Concluding remarks 6. Hillerborg A, Modeer M, Petersson PE (1976) Analysis of
A method for modelling crack growth has been described crack formation and crack growth in concrete by means of
in which a crack is represented by a collection of cohesive fracture mechanics and finite elements. Cement Concrete Res.
segments with a finite length. The segments are added to 6: 773–782
finite elements by using the partition-of-unity property of 7. Andrews DJ (1976) Rupture velocity of plane strain shear
the finite element shape functions. The method permits cracks. J. Geophys. Res. 81: 5679–5687
8. Needleman A (1987) A continuum model for void nucleation
crack nucleation and discontinuous crack growth to be by inclusion debonding. J. Appl. Mech. 54: 525–531
modelled, irrespective of the structure of the finite element 9. Xu XP, Needleman A (1994) Numerical simulations of fast
mesh. The numerical formulation is a moderate extension crack growth in brittle solids. J. Mech. Phys. Solids 42: 1397–
to existing methods which capture discontinuities using 1434
10. Rots JG (1991) Smeared and discrete representations of lo- Rammerstorfer FG, Eberhardsteiner J (eds). Vienna Univer-
calized fracture. Int. J. Fract. 51: 45–59 sity of Technology, Vienna, Austria, Paper 81054
11. Camacho GT, Ortiz M (1996) Computational modelling of 22. Babuska T, Melenk JM (1997) The partition of unity method.
impact damage in brittle materials. Int. J. Solids Struct. 33: Int. J. Numer. Meth. Eng. 40: 727–758
2899–2938 23. Moës N, Dolbow J, Belytschko T (1999) A finite element
12. Falk ML, Needleman A, Rice JR (2001) A critical evaluation method for crack growth without remeshing. Int. J. Numer.
of cohesive zone models of dynamic fracture. Journal de Meth. Eng. 46: 131–150
Physique IV 11 Pr5: 43–50 24. Wells GN, Sluys LJ (2001) A new method for modeling
13. Papoulia KD, Sam CH, Vavasis SA (2003) Time continuity in cohesive cracks using finite elements. Int. J. Numer. Meth.
cohesive finite element modelling. Int. J. Numer. Meth. Eng. Eng. 50: 2667–2682
(submitted) 25. Belytschko T, Moës N, Usui S, Parimi C (2001) Arbitrary
14. Bažant ZP, Oh B (1983) Crack band theory for fracture of discontinuities in finite elements. Int. J. Numer. Meth. Eng.
concrete. RILEM Materials and Structures 16: 155–177 50: 993–1013 77
15. Crisfield MA (1982) Accelerated solution techniques and con- 26. Wells GN, de Borst R, Sluys LJ (2002) A consistent geomet-
crete cracking. Comput. Meth. Appl. Mech. Eng. 33: 585–607 rically non-linear approach for delamination. Int. J. Numer.
16. de Borst R, Nauta P (1985) Non-orthogonal cracks in a Meth. Eng. 54: 1333–1355
smeared finite element model. Eng. Comput. 2: 35–46 27. Moës N, Belytschko T (2002) Extended finite element method
17. Ortiz M, Leroy Y, Needleman A (1987) A finite element for cohesive crack growth. Eng. Fract. Mech. 69: 813–833
method for localized failure analysis. Comput. Meth. Appl. 28. Schellekens JCJ, de Borst R (1992) On the numerical inte-
Mech. Eng. 61: 189–214 gration of interface elements. Int. J. Numer. Meth. Eng. 36:
18. Belytschko T, Fish J, Engelman BE (1988) A finite element 43–66
with embedded localization zones. Comput. Meth. Appl. 29. Daux C, Moës N, Dolbow J, Sukumar N, Belytschko T (2000)
Mech. Eng. 70: 59–89 Arbitrary branched and intersecting cracks with the extended
19. de Borst R, Wells GN, Sluys LJ (2001) Some observations on finite element method. Int. J. Numer. Meth. Eng. 48: 1741–
embedded discontinuity models. Eng. Comput. 18: 241–254 1760
20. Wells GN (2001) Discontinuous modelling of strain localisa- 30. Remmers JJC, Wells GN, de Borst R (2001) Analysis of
tion and failure. Dissertation. Delft University of Technology: delamination growth with discontinuous finite elements. In:
Delft Solids, Structures and Coupled Problems (Proceedings of the
21. Mosler J, Meschke G (2002) A comparison of embedded Second European Conference on Computational Mechanics),
discontinuity approaches with fracture energy based Pamin J (ed.). Cracow University of Technology, Cracow,
smeared crack models. In: Proceedings of the Fifth Congress Poland, CD-ROM
on Computational Mechanics, WCCM V. Mang HA,