Engineering Fracture Mechanics 70 (2003) 19631987

www.elsevier.com/locate/engfracmech

On the practical application of the cohesive model

Alfred Cornec *, Ingo Scheider, Karl-Heinz Schwalbe
Institute for Materials Research, GKSS Research Centre Geesthacht, Max-Planck-Strasse 1, 21502 Geesthacht, Germany
Received 11 November 2002; received in revised form 12 March 2003; accepted 12 March 2003

Abstract
The cohesive model has been formulated such that it can be used for practical application. A specic traction
separation law is proposed which is mainly given by the cohesive stress, T0 and the cohesive energy, C0 . Experimental
procedures have been developed which allow the determination of these material parameters. By means of experiments
on three dierent materials and dierent specimen geometries it has been demonstrated that the proposed procedure
provides very good predictions.

2003 Elsevier Science Ltd. All rights reserved.
Keywords: Cohesive model; Cohesive stress; Cohesive energy; Simulation of crack extension; Validation; Finite element analysis

1. Introduction
The classical fracture mechanics concepts provide the basis for comprehensive structural assessment
methods. These methods comprise procedures for the determination of the relevant material properties,
many of which are cast in national and international standards, e.g. [14], whereas some are available in
formats close to formal standards, see for example the ESIS procedures [5,6], a draft ISO standard [7] and
the GKSS test method [8].
The assessment routines as suchwhich have a much higher complexity than test methodsare now
also close to being standardized; a prominent example is the European SINTAP procedure [9], the core
assessment modules of which are based on the R6 procedure [10], developed by British Energy and the
ETM [11,12] developed by GKSS. SINTAP is presently extended within the EU Network FITNET [13] to
cover a wider range of failure mechanisms.
This is to be regarded as a very positive development and countless successful applications to failure
investigations, failure avoidance measures, material improvement, inspection strategies and so on have
demonstrated that fracture mechanics is a powerful and indispensable engineering tool.
However, this success story cannot ignore the limits of classical fracture mechanics: in the beginning,
fracture mechanics wasand in most applications still isbased on a single parameter treatment of
cracked bodies which gives only incomplete descriptions of the near crack tip stress and strain conditions.

Corresponding author. Tel.: +49-4152-87-2535; fax: +49-4152-87-2534.

E-mail address: [email protected] (A. Cornec).

0013-7944/03/$ - see front matter
2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0013-7944(03)00134-6

1964

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Nomenclature
vLL
load line displacement (for C(T) only)
F
applied force
B
specimen thickness
Bnet
net section thickness for side-grooved specimens
C(T)
compact tension specimen
CMOD crack mouth opening displacement (for M(T) only)
CTOD crack tip opening displacement
J
J -integral
Ji
J -integral dened at fracture initiation
M(T) centre cracked specimen (panel)
NRB notch round bar
Rp0:2
yield stress at 0.2% strain oset
Rm
ultimate stress
SZW stretch zone width at crack tip
SZWi stretch zone width dened at fracture initiation
T
applied cohesive stress (traction)
T0
maximum cohesive stress level
W
specimen width
d
applied cohesive separation
d0
maximum cohesive separation dening fracture
d1 , d2 dened separation point on the tractionseparation law
d5
specic dened CTOD with 5 mm gauge length
Da
crack extension
DaB
crack extension by blunting

Since fracture properties are heavily dependent on these constraint conditions, it is evident that this way the
fracture behaviour of cracked bodies can only approximately be described by a single parameter. In fact,
single-parameter treatment is fairly accurate under very limited conditions of widespread yielding and a
state of two-dimensional plane strainthe latter restriction is due to the fact that test standards are mostly
restricted to plane strain conditions.
This poses a problem which is known as the transferability problem: the conditions in a test piecedue
to the high degree of stress triaxiality at the crackprovide lower bound fracture properties, whereas
structural conditions cover a wide range of arbitrary stress triaxiality (or constraint) [14].
There are three dierent ways out of this dilemma:
ii(i) The most frequently used route is the conservative assessment: plane strain data are used to assess the
behaviour of components with conditions likely to be less severe than a state of plane strain. In most
cases this has been satisfactory.
i(ii) A further solution can be characterized by adjusted testing, which means testing specimens modelling
more closely the structural constraint conditions than the standard plane strain C(T) or SE(B) specimens do [7,15]. The wide plate test is the most prominentand classical example. It models membrane-like stress states which are present in many structural elements, e.g. for pipeline or aerospace
applications.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1965

(iii) The third approach tries to solve the problem closer to its origin: it acknowledges the fact that a single
parameter is not sucient to cover the whole range of structural constraint to compensate for the deviations of the actual stress elds from the reference stress elds. For linear elastic conditions the concepts of the T -stress [16], and for elasticplastic conditions the J Q concept [17] were introduced. A
further complication arises when yield strength mismatched welded joints are to be assessed: for this
class of problems the J QM concept was introduced [18], where M accounts for the additional constraint variations due to the yield strength mismatch.
For the latter two cases, the eorts needed are tremendous. Not only have numerous classes of structural
problems to be treated numericallythis adds another two dimensions to the compilations of J -integral
solutions such as the EPRI Handbook [19]the fracture resistance of the materials under consideration has
also to be determined as functions of Q and M (N.B.: even the M term alone is not sucient to describe the
fracture behaviour of welded joints, i.e. in the neighborhood of a tip of an interface crack the strain eld
exhibits intensications on the softer side and on the harder side, as compared to the respective homogeneous cases, thus aecting the local fracture process evolution).
In spite of the tremendous success that classical fracture mechanics has achieved so far, more advanced
methods are being sought for, in particular for application cases where high exploitation of the load carrying capacity is required, without compromising safety.
In this attempt, the various damage mechanics models have opened a new perspective: whereas classical
fracture mechanics treats the local fracture event using parameters which describe the conditions near the
crack tip only for limiting conditions, damage mechanics provides physical models of the fracture process
which are embedded into a two- or three-dimensional nite element model. If the model parameters are
known, then the deformation and fracture behaviour of the component can be well described. This way the
transferability problem is avoided since the material parameters are supposed to be independent of geometrical inuence.
Among the various damage models, the cohesive model (CM) seems particularly attractive for practical
application since it is very computer eective and requires only two parameters, which can be determined in
experiments with relative ease.
The fundamental idea for the CM was formulated by Barenblatt [20] as a tractionseparation law for the
decohesion of atomic lattices. This way, the unrealistic continuum mechanics stress singularity at the crack
tip could be avoided.
Dugdale [21] used this model earlier to describe analytically the plastic deformation near the crack tip
whereby the normal stress was limited by the yield strength of an elastic-ideally plastic material. Its application as a fracture model occurred substantially later, using the nite element analysis method.
Applications of the CM to the fracture behaviour of concrete have been put forward by Hillerborg et al.
[22] and Petersson [23]. Further developments in this area have been described by Elices et al. [24], Planas
et al. [25] and Ba
zant [26]. A comprehensive overview of concrete structures and the use of the CM is given
by Carpinteri and Ferro [27].
On the mesolevel, Tijssens [28] was able to model crack initiation and extension by taking into account
the presence of hard aggregates in a visco-plastic matrix. Similarly, Tijssens et al. [29] applied the CM to
craze formation in polymers.
Fundamental work on the application of the CM to metallic materials has been performed by Needleman [30,31], Tvergaard and Hutchinson [3234], and Tvergaard [35]. Further work on crack extension in
metallic materials has been published by Lin et al. [36,37] and Cornec et al. [38]. Also dynamic fracture, e.g.
[39,40], time dependent fracture processes, e.g. [28,41], and recently near-threshold fatigue behaviour by
Deshpande et al. [42] have been treated with the CM.
It has thus been demonstrated that the CM is able to cover a wide range of materials and fracture
mechanisms. The various authors have used cohesive laws with dierent tractionseparation formulations.

1966

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

There is, however, no natural tractionseparation law given a priori: the CM is a phenomenological
model which does not claim to model the real physical fracture process and the choice of the traction
separation law is basically free as there is no unequivocal correlation between the law and the result of the
analysis.
Up till now, the CM has been mainly used to perform systematic parameter studies to demonstrate the
sensitivity of the fracture process, expressed by the parameters, on the macroscopic component.
The GKSS Institute of Materials Research develops the structural assessment method Engineering Flaw
Assessment Method (EFAM) [43] which is based on experimental and analytical methods of classical
fracture mechanics. However, as the methods of damage mechanics are becoming suciently mature for
engineering application, they are considered to be included into an extended GKSS assessment procedure.
Therefore, the present paper describes a proposed method for practical application of the CM with the
emphasis on metallic materials for which several validations will be presented. The method used consists of
a dened shape of the tractionseparation law,
determination of the parameters for the tractionseparation law,
suggestions for the nite element analysis.

2. Choice of the cohesive model for ductile fracture

The basic idea of the CM is shown in Fig. 1. The ductile tearing process, consisting of initiation, growth
and coalescence of voids is represented by a tractionseparation law, simulating the deformation and nally
the decohesion of the material in the immediate vicinity of the crack tip, see left-hand side of Fig. 1. The
centre of Fig. 1 shows schematically the implementation of the CM in a nite element model. Interface
elements representing the damage are implemented between the continuum elements representing the
elasticplastic properties of the material.

Fig. 1. Representation of the ductile failure process by the CM.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1967

One of the key problems in the application of the CM is the choice of the material law within the cohesive zone. Since the tractionseparation law is not known a priori a decision has to be taken as a choice of
an appropriate function. Several investigations deal with the eect of the shape of the tractionseparation
function on the resulting fracture behaviour [24,29,31,32]. Tvergaard and Hutchinson [32,33] came to the
conclusion that this eect can be relatively weak.
It seems that a shape like that shown in the top diagram on the right-hand side of Fig. 1 is appropriate
for ductile materials, whereas the bottom diagram is characteristic of brittle materials. We propose the
shape function in Fig. 2 for modelling ductile fracture. Its three sections are given by the following analytical expressions:
8
"

#
2
>
d
d
>
>
T
for d < d1

2

>
0
>
>
d1
d1
<
1a; b; c
T T0 "
# for d1 6 d 6 d2


3


2
>
>
>
d  d2
d  d2
>
>
3
1
for d2 < d < d0
>
: T0  2 d0  d2
d0  d2
With d1 0:01d0 and d2 0:75d0 , the tractionseparation law is then determined by three material parameters:
cohesive stress, T0 ,
cohesive energy, C0 ,
separation at material decohesion, d0 .
However, only two parameters are independent since the third one can be determined from



Z dc
1 1 d1 1 d2
C0
T dd T0


2 3 d0 2 d0
0

which in our case simplies to

C0 0:87T0 d0

An even simpler tractionseparation law would be given by the rectangle

C0 T0 d0

Fig. 2. Proposed tractionseparation law.

1968

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

However, the two increasing and decreasing branches in Fig. 2 were chosen for the following reasons:
the initial slope is needed to avoid numerical problems between the cohesive elements and the surrounding continuum elements, and
the descending slope models the rapid softening during void growth and coalescence.
Ductile fracture may occur in various modes:
(a) Normal fracture, where the fracture plane is perpendicular to the highest normal stress.
(b) Shear fracture, where the fracture plane coincides with the plane of maximum shear stress.
(c) A combination of both which is typical for the fracture behaviour of thin sections; in this case normal
and shear modes are present. Slant fracture is a further designation used for this case.
The parameters for shear fracture, case (b), are in general dierent from those modelling mode I fracture.
Fracture combination, case (c), i.e. mixed mode problems, is modelled by a combined normal and tangential separation. The theory for this issue is investigated for example by Bocca et al. [44], Tvergaard and
Hutchinson [45], Xu and Needleman [39], Camacho and Ortiz [46] and will be discussed further in Section
6.2.
Slant fracture in thin sheets is a special case of the mixed mode problem, since this issue can be solved by
quasinormal separation in conjunction with plane stress or shell elements, i.e. the fracture surface is
modelled by a projection onto a plane normal to the applied stress. The reduction to the two-dimensional
space leads to a crack path which is modelled by cohesive line elements, opening under normal separation
[47,48].
The cohesive parameters for slant fracture are dierent from those used for mode I fracture, T0N , C0N ,
but also dierent from the parameters for mode III fracture, T0S , C0S . For a simulation of slant fracture,
using the method described above, it is assumed that the crack deviates immediately from the initial mode I
condition to the shear dominated 45 direction, which is an approximation for thin sections where the real
transition range is often sucient small.
For a more fundamental treatment of the problem of the transition behaviour from normal to slant
fracture, a full 3D analysis would be required. In such an analysis, a certain volume has to be equipped with
cohesive elements between the surfaces of continuum elements. A rst step have already been done in
[49].

3. Aspects of numerical implementation of the cohesive model

The CM has been implemented using the programming language FORTRAN as a user dened element
within the FE code ABAQUS [50].
The polynomial function of the tractionseparation law, dened in Eq. ((1a)(c)), has been implemented
by Scheider [51]. User dened interface elements have been developed for two- and three-dimensional
applications. Since they behave like contact elements they have one dimension less than the surrounding
continuum elements. In the undeformed state, the nodes of the upper and the lower faces of the contactpair have the same coordinates. The cohesive elements used throughout the paper for two-dimensional
applications have four nodes with a linear displacement formulation.
Contrary to the linear formulation of the cohesive elements for two-dimensional applications, implementation of the three-dimensional elements is such that they can be built using eight nodes and a linear
displacement formulation or with quadratic displacement formulation using 16 or 18 nodes. Fig. 3 shows
these elements.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1969

Fig. 3. Element nodes and integration point position for the cohesive elements.

The separations and the stresses are calculated in every increment for all cohesive elements at the integration points according to the tractionseparation law. When the critical separation energy is reached,
the element has failed at this point. The contribution of the integration point to the stiness vanishes and
the point obtains the status failed. An integration point that has lost its stiness once, can never obtain
another status.
The stiness matrix for the cohesive element can be derived based upon the principle of virtual work. For
clarication, since the symbol d denotes the variation of a quantity in variational mechanics, the displacement jump [u]
d u u u
is used.
The internal virtual work of stresses for the cohesive element is dened by
Z
dPi T  du dA

where T is the vector of the cohesive stresses acting on the separation. The reformulation for an incremental
analysis gives
Z
oT
 Dudu dS
7
dPi
oB ou
In the following a compatible formulated approach for the coordinates and the separations [u], du
according to the nite element formulation of the connected continuum elements is used (linear, quadratic,
two- or three-dimensional, etc.). The displacements u are replaced by the product of the matrix of shape
functions Vu and the displacements at the nodes ue
u Vu  ue

The rank of Vu depends on the degree of the polynomials, which describe the element geometry and displacements. With Eq. (8), the increment of the energy of the cohesive element, Eq. (7), writes:

1970

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

dPi du 

Z
oB

VTu 

oT
 Vu dA  Due du  K  Due
ou

with K being the stiness matrix of the cohesive element.

For the calculation of the stiness of the cohesive element, i.e. the rate of the load in dependence on the
displacement, it is necessary to evaluate the derivative of the normal, TN , and tangential tractions, TS , to the
cohesive separations including the mixed terms oTN =odS and oTS =odN , which are necessary for general
mixed modes of fracture.
The integration over the surface of the cohesive element is done numerically in local coordinates and
transformed into global ones afterwards.
Another implementation aspect regards the coordinate system used for the deformed cohesive element.
Contrary to a conventional contact algorithm, there is no master and slave surface for the cohesive element,
from which the coordinate system and reference plane can be dened. Instead, there are two possibilities to
dene the reference plane:
i(i) If the original position of the cohesive element is taken as the reference for the coordinate system during
the entire analysis, this means the assumption of small deformations.
(ii) According to the theory of large deformations in an updated Lagrangian formulation, the coordinate
system can be dened by moving with the element using a mid section face, that is the bisector between
upper and lower surfaces.
The choice of the coordinate system is important, since it aects the normal and the tangential portions
of the separation of the cohesive element for arbitrary deformations.

4. Determination of the cohesive parameters

4.1. Determination of the cohesive stress
For the determination of the cohesive stress, T0 in the case of normal fracture a hybrid technique has
been developed. Fig. 4 depicts qualitatively the procedure for normal fracture: a notched specimen with a
circular cross section undergoes a tensile test. Using conventional elasticplastic nite element analysis, the
distribution of the axial stress across the notch section of the specimen is determined for the instant of crack
initiation in the centre of the specimen, which coincides well with nal fracture. At that event, the axial
stress exhibits a maximum in the centre of the specimen, which is supposed to be equal to T0 .
For slant fracture, thin at tensile specimens serve for the determination of T0 . Due to the simple
geometry, the hybrid technique using a nite element analysis is not needed: the applied force at fracture is
divided by the instantaneous cross section as in a thin at tensile specimen a uniform stress state can be
assumed.
4.2. Determination of the cohesive energy
The cohesive energy, C0 , can be determined in a fracture mechanics test by assuming that C0 equals the
J -integral at initiation of stable crack extension, Ji . Fig. 5 demonstrates the procedure which is taken from
standard test methods [2,5,8]: The stretch zone width at initiation, SZWi , is determined on at least three
specimens exhibiting ductile tearing beyond the 0.2 mm oset line, see Fig. 5(a). The intersection point of
the average SZWi and the J Da curve denes Ji . It should be noted that the standard methods require the
determination of Da for this purpose using the multiple specimen method, Fig. 5(b), since the indirect

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1971

Fig. 4. Hybrid technique for determining the cohesive stress, T0 : combination of experiments and conventional elasticplastic FE
calculation (qualitative description).

Fig. 5. Determination of the cohesive energy, C0 , by three dierent methods, each setting C0 equal the J -integral at initiation, Ji : (a)
determination of the J -integral at initiation, Ji , using stable crack extension of the resistance curve and the stretch zone width, SZWi ,
after [2,5,8]; (b) determination of Ji , using multiple specimens each with some amount of stable crack extension and the stretch zone
width, after [2,5,8]; (c) determination of Ji using the analytical blunting line and the nal stretch zone width, SZWi . SZWi has to be
determined on at least three specimens, each with sucient amount of stable crack extension.

1972

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Fig. 6. Overview on the specimens used for the determination of the cohesive parameters.

techniques are not suciently reliable for the small amounts of crack extension required for the determination of Ji . Alternatively, the intersection of the analytical blunting line [2,5,8]
J 3:75Rm  DaB

10

with the critical stretch zone width, SZWi , can be used for determining Ji , see Fig. 5(c), with Rm being tensile
strength and DaB crack extension due to blunting. In this case, no determination of the J -integral is needed.
The drawback of the method is that the angle between the blunting line and the constant line for SZWi is
relatively small, so that small errors in SZWi may have a strong eect on the value of Ji .
For the determination of the cohesive energy for slant fracture, the best way to determine crack initiation
experimentally is from a fracture specimen with a pre-existing slant fracture plane. This can be realized by a
machined slit in a specimen inclined under 45 to the surface of the specimen. The fatigue crack emanating
from this slit should be driven as short as possible because otherwise it may run into a plane perpendicular
to the applied load, i.e. into a mode I fracture plane. Fig. 6 illustrates the various steps for determining the
material parameters.
If this procedure is not applicable, crack initiation can also be determined using a thin fracture specimen
with a conventional mode I fatigue crack. The corresponding value of Ji can be regarded as an upper bound
and can be used as a good approximation for the value of C0 . This has been done in the present investigation.

5. Validation
A number of validations have been performed in order to check the suitability of the procedures described above. To this end three materials with a wide variety of properties have been examined.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1973

Fig. 7. Specimen types for 2024-FC used to generate the cohesive parameters and used for the validation of the CM: (a) NRB; (b) C(T)
specimen; (c) surface cracked tension specimen; (d) eng. stressstrain curves.

5.1. Aluminum 2024-FC

This material represents a model material with a very low yield strength (Rp0:2 81 MPa) and high
hardening capacity and has been extensively investigated with regard to its crack extension properties, e.g.
[52,53]. It was made by overaging a 100 mm thick plate of aluminum 2024-T351. Experimental data were
generated on dierent specimen types as shown in Fig. 7 [37].
5.1.1. Determination of cohesive parameters
Conventional elasticplastic nite element analysis of the round tensile specimen shown on the left-hand
side of Fig. 7 served for the determination of T0 . Hereby the rotational symmetry of the specimen was
exploited. The result is presented in Fig. 8. The maximum axial stress at fracture was set equal to the
cohesive stress providing T0 420 MPa.
The cohesive energy, C0 Ji , was determined using the intersection of SZWi (SZWi 1012 lm) with
the R-curve from the potential drop method, Fig. 9(a), the R-curve with multiple specimen method, Fig.
9(b), and the blunting line, Fig. 9(a). The diagrams demonstrate that all three approaches yield the same
value for crack initiation. For the simulations of the material 2024-FC the above determined cohesive
parameters are used nally: T0 420 MPa, C0 10 N/mm.
5.1.2. Simulations with the cohesive model
The rst simulation was done on the tensile specimen to check the appropriateness of the value for T0
and is shown in Fig. 10. The calculation using cohesive elements in a line through the necking region meets
the experimental fracture point very precisely with T0 420 MPa. This result is remarkable in so far as the
onset of the fracture point is very sensitive with respect to the value of T0 .
For the C(T) specimen of Fig. 7, the results are shown in Figs. 11 and 12. The simulated force versus the
loadline displacement (F vLL ) curve meets the experimental values very accurately. Moreover, the shape of
the crack front is also very well predicted.

1974

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Fig. 8. Determination of the cohesive stress, T0 , for 2024-FC using the hybrid technique of test and FE calculation.

multiple specimen method

Fig. 9. Determination of crack initiation, Ji , for 2024-FC by three dierent methods: (a) R-curve from single specimen with crack length
from potential drop method and its intersection with the nal stretch zone width; (b) R-curve from multiple specimen method with
optical determined crack lengths and its intersection with the nal stretch zone width; for small crack extension a linear regression line
is used; (c) intersection of the analytical blunting line with the nal stretch zone width.

A similar exercise was performed for the tensile panel with a surface crack, depicted in Fig. 7. Fig. 13
shows that both the force-crack mouth opening displacement (F-CMOD) curve, Fig. 13(a), and the crack
front shape, Fig. 13(b) and (c) were also predicted with high accuracy.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1975

Fig. 10. Simulation of the global normalized forcedisplacement curve (engineering axial stress vs. strain) of the NRB for 2024-FC with
variation in the cohesive stresses.

Fig. 11. Simulation of the forcedisplacement curve (F vLL ) of the C(T) specimen in Fig. 7 in comparison with the test result for 2024FC.

5.2. Aluminum 2024-T351

This material is typically used in the form of thin sheets, in particular for aerospace structures. Crack
extension resistance curves were determined on 1.6 mm thick C(T) specimens with W 50, 100, 200, and
1000 mm, of which only the C(T) specimens with W 50 mm and W 1000 mm were simulated [48].

1976

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Fig. 12. Crack front shapes in C(T) specimen of 2024-FC: (a) development of simulated crack front shapes; (b) comparison of the crack
front shapes from the simulation and test.

Fig. 13. Surface cracked tension specimen of 2024-FC: (a) comparison of simulated and experimental F-CMOD curves; (b) development of simulated crack front shapes; (c) comparison of crack front shapes at the interrupted test and the corresponding simulation.

5.2.1. Determination of cohesive parameters

The at tensile specimen shown in Fig. 14, served for the determination of the stressstrain curve and
for T0 . The material is a high strength aluminum with Rp0:2 300 MPa and Rm 427 MPa. A value of

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1977

Fig. 14. Stressstrain curve and cohesive stress, T0 , for 2024-T351 determined on a at tensile specimen failed by a slant fracture.

550 MPa was found for T0 . Note that this value is used only for slant fracture most likely occurring in thin
sheets, which is modelled by plane stress elements. However, the cohesive strength for at fracture would
yield higher values.
The cohesive energy was determined on C(T) specimens tested several years ago [52] which had precracks perpendicular to the applied force, unlike the procedure for shear crack evolution as depicted on the
right-hand side in Fig. 6. Crack initiation, Ji , was determined in the same fashion as already outlined in Fig.
9, resulting in values of Ji 9:5, 10, and 8.8 N/mm as presented in Fig. 15(a)(c). Although this initiation
value belongs to mode I from Fig. 15(a) it can be expected that this value will be close to an extrapolation
value from the R-curve representing full slant fracture after small crack extension.
5.2.2. Simulations with the cohesive model
The R-curves of the C(T) specimens were available in the form of crack tip opening displacement,
CTODd5 , versus crack extension, Da. Duplicate tests had been done for both simulated tests W 50 mm
and W 1000 mm. The simulations were performed before the cohesive law of Fig. 2 was established: at
that time a rectangular cohesive law was used [36,37], which provides results very similar to those that
would have been obtained with the law in Fig. 2. For C0 the value 9.5 N/mm was chosen, together with
T0 550 MPa.
The results are shown in Fig. 16. The behaviour of the small specimens is accurately predicted by the
simulation, see Fig. 16(a), whereas for the wide specimens the crack extension resistance is underestimated,
Fig. 16(b). The substantial friction between the specimens and the anti-buckling plates used in these tests
and the free window for observation may have inuenced to a nominal increase in crack extension resistance. Further arguments on size-scales transitions are given in [27].
5.3. Pressure vessel steel 20 MnMoNi 55
The steel 20 MnMoNi 55 is typically used for reactor pressure vessel applications. The stressstrain
curves of ve standard round tensile specimens tested for the determination of tensile properties exhibited

1978

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Fig. 15. Determination of crack initiation, Ji , for 2024-T351 by three dierent methods: (a) R-curve from single specimen with crack
length from potential drop method and intersection with the nal stretch zone width; (b) R-curve from multiple specimen method with
optical determined crack lengths and intersection with the nal stretch zone width; for small crack extension a linear regression line is
used; (c) intersection of the analytical blunting line with the nal stretch zone width.

Fig. 16. Simulation and experimental R-curves of 50 and 1000 mm wide C(T) specimens made of 1.6 mm thick 2024-T351.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1979

Fig. 17. Stressstrain curves for 20 MnMoNi 55 as determined by ve tests; for the simulations the represented average one up to the
ultimate stress is considered.

some variation as shown in Fig. 17. It is expected that this variation will aect the simulation results accordingly. The open circles in Fig. 17 indicate unloading of the specimens prior to nal fracture. A replot of
these nal points in terms of true stress versus true strain using Bridgmans correction is shown in Fig. 18. A
power law was tted through these points and used in the subsequent simulations.

Fig. 18. True stressstrain curve for 20 MnMoNi 55 beyond ultimate stress used for the simulations.

1980

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

5.3.1. Determination of cohesive parameters

The round notched bar (NRB) shown in Fig. 19(a) served for the determination of T0 . The experimental
average stress is plotted in Fig. 19(b) as a function of the measured reduction of the notched cross section.

Fig. 19. Determination of the cohesive stress, T0 , for 20 MnMoNi 55 using the hybrid technique of test and FE calculation: (a) NRB
geometry; (b) comparison of average axial stress versus the diameter reduction from conventional FE calculation and test and the
maximum true axial stress in the centre of the specimen from the FE calculation; (c) stress distribution from the FE-calculation across
the notch section until reaching nal fracture, known from the NRB test.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1981

Also shown are the stress distributions of an elasticplastic nite element analysis. The average stress
distribution of the nite element analysis meets very well the experimental curve. The maximum stress,
occurring in the centre of the specimen, is also plotted and at unstable fracture this maximum stress is set
equal to the cohesive stress T0 1460 MPa. Additionally, in Fig. 19(c) the axial stress distribution from the
previous elasticplastic FE analysis is plotted across the notch section of the round notched bar NRB until
the experimentally determined fracture load.
Fig. 20 shows the determination of the crack initiation, Ji , using the critical stretch zone width, with
SZWi 52 lm, and its intersection with the R-curve from the potential drop method, Fig. 20(a), the Rcurve from the multiple specimen method (four specimens), Fig. 20(b), and with the analytical blunting line,
Fig. 20(c), resulting in Ji C0 120 N/mm used in the simulations.
5.3.2. Simulations with the cohesive model
For the validation of the cohesive simulation procedure, crack extension experiments were available on
the two specimen geometries shown in Fig. 21: A side grooved C(T) specimen and a side grooved M(T)
specimen [54].
From the simulations by Scheider [49] the behaviour of the C(T) specimen is almost up to maximum
force determined by the elasticplastic properties of the material, Fig. 22. Prior to maximum force crack
extension starts to aect the global forcedisplacement characteristics. The curve from the simulation
follows closely the experiment.
The development of the crack front shape can serve as a very sensitive check of the simulation procedure.
Fig. 23 shows the crack front at the end of test of a C(T) specimen. Also shown are two simulations, one of
which was performed with a rectangular tractionseparation law and has a more rened element mesh

Fig. 20. Determination of crack initiation, Ji , for 2024-T351 by three dierent methods: (a) R-curve from single specimen with crack
length from potential drop method and intersection with the nal stretch zone width; (b) R-curve from multiple specimen method with
optical determined crack lengths and intersection with the nal stretch zone width; for small crack extension a linear regression line is
used; (c) intersection of the analytical blunting line with the nal stretch zone width.

1982

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

Fig. 21. Specimen types for 20 MnMoNi 55 used to used for the validation of the CM: (a) C(T) specimen; (b) M(T) specimen, with 20%
side grooves.

Fig. 22. Simulation of the forcedisplacement curve (F vLL ) of the C(T) specimen and comparison with the test record for 20 MnMoNi
55.

across the specimen width. Both of them model the real crack fairly well, in particular the ones with ner
mesh. However, the CM does not generally require very high mesh density.
The results obtained for the M(T) specimen are shown in Fig. 24(a) for the F-CMOD curve and Fig.
24(b) for the crack extension resistance curve, again both with very good agreement between simulation and
experiment.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1983

Fig. 23. Crack front shapes in C(T) specimen of 20 MnMoNi 55: comparison of simulations and one test (C(T)-R22). The simulated
crack fronts are shifted together in the centre on the experimental crack front; displacement vLL at end of test C(T)-R22 is
vLL;test 5:351 mm, in the simulation vLL;simul: 5:358 mm.

Fig. 24. Validation of the CM and the proposed procedures on a M(T) specimen of 20 MnMoNi 55: (a) force vs. CMOD; (b) J Da
curve.

1984

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

6. Determination of cohesive parameters for other fracture modes

6.1. Shear fracture (Modes II and III)
It is important to distinguish between global shear loading and local shear failure mechanism. Global
mode II loading may readily lead to local mode I separation where the crack extends perpendicularly to the
principal normal stress and by deviating from the original crack extension direction. For the treatment of
modes II and III there is no dierence as both cases are described by tangential separation. Even if there
exists a dierence between the global fracture mechanics parameters for modes II and III (i.e. KIIc and KIIIc
respectively) this does not necessarily mean dierent cohesive parameters for both tangential directions.
The authors, therefore, expect that the cohesive parameters are independent of the direction of the tangential separation leading to pure shear failure.
The accurate determination of the parameters for tangential separation is dicult:
it must be assured that the desired failure mode is active,
for shear fracture there are no generally accepted test procedures available for the determination of crack
initiation.
The procedures for the determination of the cohesive parameters, therefore, need further work.
6.2. Interaction between normal and tangential separations
The interaction between normal and tangential separations is often set equal to mixed mode fracture.
This is, however, only partially correct. According to our experience, in most cases the cohesive element is
activated in normal or tangential direction only, i.e. only the maximum normal stress, T0N , or the maximum
tangential stress, T0T , is reached [51]. If separation occurs in both normal and tangential directions, this is
mostly a numerical problem, e.g. if crack extension is calculated with an irregular FE mesh [28], or if in the
prediction of crack path direction the crack has to choose its path in a mesh dependent zig-zag manner [39].
Almost all cohesive laws have their own interaction rules, which, however, in most cases are not compatible
with each other. Interaction of failure modes is outside the scope of the present paper so that this item has
not been treated here.

7. Conclusions
A procedure has been developed for the application to the assessment of engineering structures. This
procedure consists of a specic tractionseparation law of the CM and methods for determining the material parameters. The tractionseparation law is characterized by a constant cohesive stress, T0 which is
preceded by a steep slope for numerical reasons and followed by a more gentle descending slope representing the softening behaviour of the material. The area under the tractionseparation law represents the
cohesive energy, C0 which also characterises the material properties in the process zone.
T0 is determined from a tensile test on a NRB with an accompanying elasticplastic nite element
analysis for normal (at) fracture and from a tensile test on a at specimen for slant fracture. The cohesive
energy, C0 is set equal to the J -integral at initiation of stable crack extension, Ji , which can be determined in
a standard fracture mechanics test.
Validation on three quite dierent materials in the form of C(T) specimens covering a wide size range, of
a tensile panel with a surface crack, and of a M(T) specimen with a through crack demonstrate the suit-

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1985

ability of the proposed procedure. Both, the predicted macroscopic forcedisplacement characteristics of
the specimens and the crack front shapes with related R-curves agree very well with the experimental results.

Acknowledgements
The authors gratefully acknowledge the experimental support of test results for the pressure vessel steel
by Dr. Dieter Hellmann and Kay Erdmann.

References
[1] ASTM E 1820-99. Standard test method for fracture toughness. West Conshohocken: American Society for Testing and Materials
(ASTM); 1997.
[2] ISO 12135. Metallic materialsunied method of test for the determination of quasistatic fracture toughness. International
Standard Organisation (ISO), in press.
[3] BS 7448, Part 1. Fracture mechanics toughness testsmethod for determination of KIc , critical CTOD and critical J values of
metallic materials. London: British Standards Institution; 1991.
[4] ASTM E 192197. Standard Test method for determination of reference temperature, T0 , for ferritic steels in the transition range.
West Conshohocken: American Society for Testing and Materials (ASTM); 1997.
[5] ESIS P1-92. ESIS reommendations for determining the fracture resistance of ductile materials. European Stuctural Integrity
Society; 1992.
[6] ESIS P2-92. ESIS procedure for determining the fracture resistance of ductile materials. European Stuctural Integrity Society;
1992.
[7] ISO/TC164/SC-N413. Metallic materialsmethod of test for the determination of resistance to stable crack extension using
specimens of low constraint. International Standard Organisation (ISO), Draft 6, 2002.
[8] Schwalbe K-H, Heerens J, Zerbst U, Kocak M. EFAM-GTP 02. The GKSS Procedure for determining the fracture behaviour of
materials. Geesthacht, Germany: Research Centre Publications, in preparation.
[9] SINTAP. Structural Integrity Assessment Procedure. EU-Project BE 95-1462: Final Report. Brite Euram Programme, Brussels,
1999.
[10] R6, Revision 4. Assessment of the Integrity of Structures Containing Defects. British Energy Generation Ltd.; 2001.
[11] Schwalbe K-H, Zerbst U, Kim Y-J, Brocks W, Cornec A, Heerens J, et al. EFAM ETM 97the ETM method for assessing the
signicance of crack-like defects in engineering structures, comprising the versions ETM 97/1 and ETM 97/2. Report GKSS 98/E/
6. Geesthacht, Germany: GKSS Research Centre Publications; 1998.
[12] Schwalbe K-H, Kim Y-J, Hao S, Cornec A, Kocak M. EFAM ETM-MM 96the ETM method for assessing the signicance of
crack-like defects in joints with mechanical heterogeneity (strength mismatch). Report GKSS 97/E/9. Geesthacht, Germany:
GKSS Research Centre Publications; 1997.
[13] FITNET (European Fitness-for service Network). Available from http://www.eurotnet.org.
[14] Schwalbe K-H, Heerens J. R-curve testing and its relevance to structural assessment. Fat Fract Engng Mat Struct 1998;21:1259
71.
[15] Zerbst U, Heerens J, Schwalbe K-H. Fracture mechanics analysis based on a local simulation principle. Fat Fract Engng Mat
Struct 1995;18:3716.
[16] Cherry AH, France CC, Goldthorpe MR. Compendium of T -stress solutions for two and three-dimensional cracked geometries.
Fat Fract Engng Mat Struct 1995;18:14155.
[17] Shih CF, ODowd NP, Kirk MT. A framework for quantifying crack tip constraint. In: Hacket EM, Schwalbe K-H, Dodds RH,
editors. Constraint eects in fracture ASTM STP 1171. Philadelphia: American Society for Testing and Materials; 1993. p. 220.
[18] Zhang Z, Thaulow C, Hauge M. On the constraint eects in HAZ-cracked wide plate weldments. In: Kocak M, editor. Mismatching of interfaces and welds 2nd International Symposium. GKSS Research Centre Publication; 1997. p. 74958. [ISBN 3-00001951-0].
[19] Zahoor A. Ductile fracture handbook. Novotech. Cop & EPRI, Research Project 175769, vol. 1: 1989, vol. 2: 1990, vol. 3: 1991.
[20] Barenblatt GI. The mathematical theory of equilibrium cracks in brittle fracture. Adv Appl Mech 1962;7:55129.
[21] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8:1004.
[22] Hillerborg A, Modeer M, Petersson PE. Analysis of crack formation and crack growth in concrete by means of fracture mechanics
and nite elements. Cem Conr Res 1976;6:77382.

1986

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

[23] Petersson PE. Crack growth and development of fracture zones in plain concrete and similar materials. In: Report TVBM-1006,
Division of Building Materials, Lund Institute of Technology, Denmark; 1981.
[24] Elices M, Guinea GV, G
omez J, Planas J. The cohesive zone model: advantages, limitations and challenges. Engng Fract Mech
2002;69:13763.
[25] Planas J, Guinea GV, Elices M. Size eect and inverse analysis in concrete fracture. Int J Fract 1999;95:36778.
[26] Ba
zant ZP. Concrete fracture models: testing and practice. Engng Fract Mech 2002;69:165205.
[27] Carpinteri A, Ferro G. Fracture assessment in concrete structures. In: Milne I, Ritchie RO, Karihaloo B, editors. Comprehensive
structural integrity, vol. 7. Amsterdam: Elsevier Science, 2003, in press.
[28] Tijssens MGA. On the cohesive surface methodology for fracture of brittle heterogeneous solids. PhD Thesis at Technical
University Delft Netherlands: Shaker Publishing B.V.; 2000 [ISBN 90-423-0129].
[29] Tijssens, van der Giessen E, Sluys LJ. Simulation of mode I crack growth in polymers by crazing. Int J Solids Struct 2000;37:7307
27.
[30] Needleman A. A continuum model for void nucleation by inclusion debonding. J Appl Mech 1987;54:25531.
[31] Needleman A. An analysis of tensile decohesion along an interface. J Mech Phys Solids 1990;38:289324.
[32] Tvergaard V, Hutchinson JW. The relation between crack growth resistance and fracture process parameters in elasticplastic
solids. J Mech Phys Solids 1992;40:137797.
[33] Tvergaard V, Hutchinson JW. Eect of T -stress on mode I crack growth resistance in a ductile solid. Int J Solids Struct
1994;31:82333.
[34] Tvergaard V, Hutchinson JW. Toughness of an interface along a thin ductile layer joining elastic solids. Philos Mag A
1994;70:64156.
[35] Tvergaard V. Resistance curves for mixed mode interface crack growth between dissimilar elasticplastic solids. J Mech Phys
Solids 2001;49:2689703.
[36] Lin G. Numerical investigation of crack growth behaviour using a cohesive zone model. PhD Thesis, University of HamburgHarburg. Geesthacht, Germany: GKSS Research Cenrte Publication, Report GKSS 98/E/8; 1998.
[37] Lin G, Cornec A, Schwalbe K-H. Three-dimensional nite element simulation of crack extension in aluminium alloy 2024-FC. Fat
Fract Engng Mat Struct 1998;21:115973.
[38] Cornec A, Lin G, Schwalbe K-H. Simulation von Riwiderstandskurven mit dem Kohasivmodell: Large scale yielding (simulation
of crack resistance using a cohesive model: large scale yielding). Mat-wiss u Werkstotechnik 1998;29:65261.
[39] Xu X-P, Needleman A. Numerical simulations of fast crack growth in brittle solids. J Mech Phys Solids 1994;42:1397434.
[40] Zavattieri PD, Espinosa HD. Grain level analysis of crack initiation and propagation in brittle materials. Acta Mater
2001;49:4291311.
[41] Allen DH, Searcy CR. A micromechanical model for a viscoelastic cohesive zone. Int J Fract 2001;107:15976.
[42] Deshpande VS, Needleman A, van der Giessen E. A discrete dislocation analysis of near-threshold fatigue crack growth. Acta
Mater 2001;49:3189203.
[43] Schwalbe K-H. The Engineering Flaw Assessment Method (EFAM)Document EFAM 96. Geesthacht, Germany: GKSS
Research Centre Publications, Report GKSS 98/E/40; 1998.
[44] Bocca P, Carpinteri A, Valente S. Size eects in the mixed mode crack propagation: softening and snap-back analysis. Engng
Fract Mech 1990;35:15970.
[45] Tvergaard V, Hutchinson JW. The inuence of plasticity on mixed mode interface toughness. J Mech Phys Solids 1993;41:1119
35.
[46] Camacho GT, Ortiz M. Computational modelling of impact damage in brittle materials. Int J Solids Struct 1996;33:2899
938.
[47] Siegmund T, Brocks W. Modelling of crack growth in thin sheet aluminum alloys. In: Halford GL, Gallagher JP, editors. Fatigue
and fracture mechanics ASTM STP 1389. Philadelphia: American Society for Testing and Materials; 2000. p. 47585.
[48] Lin G, Cornec A. Characterization of crack resistance: simulation with a new cohesive model. (Numerische Untersuchung zum
Verhalten von Riwiderstandskurven: Simulation mit dem Kohasivmodell, in German.). Mat-wiss u Werkstotechnik
1996;27:2528.
[49] Scheider I. Bruchmechanische Bewertung von Laserschweiverbindungen durch numerische Rifortschrittsimumlation mit dem
Koh
asivzonenmodell. PhD Thesis, University of Hamburg-Harburg, Report GKSS 2001/3, Geesthacht, Germany: GKSS
Research Centre Publication; 2000.
[50] ABAQUS FEM code. Hibbitt, Karlsson, and Sorensen, Inc., 1080 Main Street, Pawtucket, RI, USA: ABAQUS Users manual,
Version 5.8.
[51] Scheider I. Simulation of cupcone fracture in round bars using the cohesive zone model. In: Bathe K-J, editor. Computational
uid and solid mechanics (First MIT Conference), vol. 1. Elsevier; 2001. p. 4602.
[52] Hellmann D, Schwalbe K-H. Geometry and size eects on J R and dR curves under plane stress conditions. In: Sanford RJ,
editor. Fracture mechanics: Fifteenth Symposium ASTM STP 833. Philadelphia: American Society for Testing and Materials;
1984. p. 577605.

A. Cornec et al. / Engineering Fracture Mechanics 70 (2003) 19631987

1987

[53] Heerens J, Schwalbe K-H, Nix C. Inuence of specimen size on J , Jm and d5R-curves for side-grooved compact-tension
specimens. In: Hacket EM, Schwalbe K-H, Dodds RH, editors. Constraint eects in fracture ASTM STP 1171. Philadelphia:
American Society for Testing and Materials; 1993. p. 42972.
[54] Heerens J, Hellmann D. Thickness contraction and ligament contractiontwo quantities to analyse constraint eects on ductile
fracture toughness. In: Kocak M, editor. The life of a crack-initiation growth fracture, Manusscript No. IX. Geesthacht,
Germany: GKSS Research Centre Publications; 2001 [ISBN 3-00-007431-7].