Académique Documents
Professionnel Documents
Culture Documents
ABSTRACT This study proposes an algorithm of embedding cohesive elements in Abaqus and
develops the computer code to model 3D complex crack propagation in quasi-brittle materials in
a relatively easy and ecient manner. The cohesive elements with softening traction-separation
relations and damage initiation and evolution laws are embedded between solid elements in re-
gions of interest in the initial mesh to model potential cracks. The initial mesh can consist of
tetrahedrons, wedges, bricks or a mixture of these elements. Neither remeshing nor objective
crack propagation criteria are needed. Four examples of concrete specimens, including a wedge-
splitting test, a notched beam under torsion, a pull-out test of an anchored cylinder and a notched
beam under impact, were modelled and analysed. The simulated crack propagation processes and
load-displacement curves agreed well with test results or other numerical simulations for all the
examples using initial meshes with reasonable densities. Making use of Abaquss rich pre/post-
processing functionalities and powerful standard/explicit solvers, the developed method oers a
practical tool for engineering analysts to model complex 3D fracture problems.
KEY WORDS nite element method, cohesive elements, three-dimensional crack propagation,
discrete crack model, concrete structures, Abaqus
I. INTRODUCTION
Crack propagation is an inherent feature in quasi-brittle materials such as concrete due to the low
material tensile strength, and it is a predominant source of nonlinearity and a main culprit for ultimate
failure of structures made from these materials. Accurate understanding of the crack propagation
behaviour so as to assess the structural load-carrying capacity, by either experimental studies or numerical
modelling, remains a worldwide challenge, especially for three-dimensional (3D) problems with complex
geometries and loading conditions.
Numerical modelling of crack propagation has been an active research eld since 1960s[1, 2]. Nowadays
there exist a large number of numerical models. In terms of how the cracks are modelled geometrically,
there are discrete crack models explicitly separating crack surfaces and modelling discontinuity, smeared
crack models based on continuum mechanics, and more indirectly, lattice models. In terms of numerical
methods, both the traditional nite element methods (FEM) and boundary element methods (BEM)
Corresponding author. Email: zjyang@liv.ac.uk
Project supported by EPSRC UK (No. EP/F00656X/1). Xiangting Sus one-year visit to the University of Liverpool
was supported by the China Scholarship Council and the National Natural Science Foundation of China (No. 50579081).
272 ACTA MECHANICA SOLIDA SINICA 2010
have been widely applied to crack propagation modelling. Extensive literature review of these models
and methods in 2D problems has been given elsewhere[36] .
It appears now that the discrete crack models, mostly based on the cohesive crack model (CCM) devel-
oped in ductile materials by Barenblatt[7] and Dugdale[8] and in quasi-brittle materials by Hillerborg[9] ,
are becoming more and more popular, because of their ability to model macroscopic cracks with strong
discontinuity, the CCMs capability of realistically representing the energy dissipation during fracture
processes, and the ease of its implementation as cohesive interface elements (CIE) in the FEM and
the BEM. In general, two types of approaches are used to model discrete cohesive crack propagation.
The rst type of approaches is based on sophisticated remeshing procedures that constantly change
the meshes as cracks propagate[3, 1012] . For problems with crack paths unknown a priori, objective
crack propagation criteria are needed to judge when and in which direction a crack propagates. This
usually involves calculation of stress intensity factors (SIFs) and stresses at crack tips, whose accuracy
can only be ensured by ne crack-tip meshes or using singular elements. This in turn exacerbates the
complexity of the remeshing procedure. It is also very dicult for the remeshing-based approaches to
deal with complex crack propagation situations, such as multiple cracking and 3D problems. Indeed, the
success of the remeshing-based approaches has been largely limited to 2D static fracture problems with
single or a few cracks, although they are computationally ecient because a relatively small number of
nonlinear CIEs are inserted into the meshes. Another type of approaches pre-insert or pre-embed CIEs
between the nite elements in the initial meshes[6, 1315] . Crack propagation is modelled by automatic
opening and merging of the CIEs under applied loadings. Neither remeshing procedures nor crack prop-
agation criteria are needed so complex crack propagation can be modelled. However, these approaches
restrict the cracks to the nite element edges or surfaces and the predicted crack patterns may thus be
mesh-dependent. Another disadvantage is the high computational cost due to the use of a large number
of nonlinear CIEs. The CIEs can also be dynamically inserted when certain crack initiation criterion
is satised[16, 17] so that the number of CIEs changes with the loading and the total number of CIEs
remains relatively low. These procedures of dynamically inserting CIEs need objective crack initiation
criteria and change meshes although the remeshing operation is relatively straightforward.
In the last two decades, the partition of unity nite element method (PUFEM)[18, 19] and the extended
nite element method (XFEM)[2022] are becoming popular in modelling crack propagation. These
methods introduce discontinuity into the nite elements. The crack growth is described independently
of the mesh, so that remeshing is not needed. Meshless methods are also receiving much attention[23, 24] .
These innovative methods, showing high potential in crack propagation modelling, generally need ne
crack tip meshes like the traditional FEM to calculate accurate stresses or SIFs used in evaluating the
crack propagation criterion. In this sense, ne initial meshes are needed if the crack paths are unknown
in priori, which means high computational cost. The newly-developed semi-analytical scaled boundary
nite element[25] is able to compute accurate SIFs without using ne meshes or singular elements. It has
recently been applied to model discrete cohesive crack propagation[4, 5, 26], but so far only 2D problems
were modelled.
In summary, all the above crack models and numerical methods have their own advantages and
disadvantages in terms of eectiveness, eciency and applicability, and the selection largely depends
upon the problems analysed and the preference and experience of the analysts. Most of the existing
studies have modelled 2D crack propagation. 3D modelling of complex cohesive crack propagation has
been sporadically conducted, and there is a long way to go before it becomes a routine exercise in
research and practical design due to the inherent diculties. In addition, most of the studies have
used specialised in-house computer programs (e.g., Cornell Universitys FRANC) that may be dicult
for practical engineers to use. Most of the general-purpose commercial FEA packages, such as Ansys
and Abaqus, are unable to model complicated crack propagation without user intervention or second
development.
This study aims at developing a practical nite element method for modelling complex 3D crack
propagation in quasi-brittle materials under static and dynamic loadings. The method makes use of
a special type of CIEs called cohesive elements, designed for modelling bonded interfaces in Abaqus
V6.5 or higher[27] , to model potential cracks. An ecient algorithm was devised to insert these CIEs
into concerned regions in FE meshes consisting of various types of solid elements. The method is simple
to implement. The rich pre/post-processing functionalities and powerful implicit and explicit solvers
Vol. 23, No. 3 Xiangting Su et al.: Complex 3D Static and Dynamic Crack Propagation 273
of Abaqus can be fully exploited. Its eectiveness is validated by modelling various fracture problems
of concrete structures.
Fig. 1. FE meshes before and after cohesive elements are inserted and 3D cohesive elements.
The cohesive elements in Abaqus[27] are based on the cohesive crack model. The constitutive response
of cohesive elements, dened in terms of traction-separation laws, assumes initially linear elastic behav-
iour followed by the initiation and evolution of damage. Several damage initiation criteria are available
in Abaqus. This study used the quadratic nominal stress criterion in all the examples. The damage
evolution criteria can be classied into three categories according to the relation between traction and
separation: linear damage evolution, exponential damage evolution, and tabular damage evolution. All
these criteria were used in this study.
2.2. Inserting Cohesive Elements into 3D FE Meshes
Although inserting cohesive elements into 2D FE meshes is relatively straightforward[6], it is not a
trivial task for 3D problems, especially if the FE mesh is not regular and consists of dierent types of
solid elements. One challenge is how to robustly deal with the changes in the complicated 3D nodal and
elemental connectivity due to the insertion of CIEs. This is tackled by an eective embedding algorithm
implemented in a MATLAB[28] program with carefully designed data structures listed in Table 1. It
has the following steps:
Table 1. Data structures used in the algorithm of inserting cohesive elements in 3D meshes
(1) Reading the nodal coordinates and the nodal connectivity of the solid elements from the Abaqus
input le. These data are then used to generate three arrays of type Structure: NODES, ELEMS
and FACES. Most of the members in these structures are self-explanatory. For example, the members
elemconn and faceconn in NODES structure are arrays containing the elements and faces the node is
connected with, respectively. It is worth noting that only those nodes, elements and surfaces in the
concerned or dened regions in the mesh are considered, in order to reduce the computational cost.
This is indicated by the member isIndenedArea. A consistent nodal ordering approach (clockwise or
anti-clockwise) in nodeconn of FACES should be used to facilitate determination of correct nodal order
when cohesive elements are created in the 4th step.
(2) Inserting new nodes. For each existing node with isIndenedArea=1, all the N number of solid
elements connected with this node are rst identied. N number of new nodes, each connected with
one solid element, are then generated with the same coordinates of the existing node. All the new nodes
are stored in an array of POINT, which is a member of a Structure BALLS associated with the existing
node. The member elemindex in POINT represents the solid element the new node is connected with.
An array of BALLS is then generated for all the existing nodes. The design of BALLS greatly facilitates
subsequent generation of cohesive elements and is at the heart of the embedding algorithm.
(3) Updating the arrays NODES and ELEMS. Since a large number of new nodes are generated,
the elemental connectivity (elemconn) of NODES and the nodal connectivity (nodeconn) of ELMS are
updated. It should be noted that the array FACES is not updated at this step.
(4) Generating cohesive elements. This is based on the array FACES. There are two solid elements
(elemconn) connected with each face. For each vertex of a face (nodeconn), a pair of new nodes connected
to the two solid elements is found in POINT of BALLS. The three or four pairs of nodes found for this
face comprise a cohesive element (COH3D6 or COH3D8). As the nodal order in FACES is carefully
designed in the 1st step, the requirement on the nodal order of cohesive elements is readily satised.
(5) Creating the Abaqus input le. The le contains the index and coordinates of POINT, the index
and nodal connectivities of the updated ELEMS and the cohesive elements.
To minimise the mesh-dependence for problems with crack paths unknown a priori, a reasonably
ne initial mesh should be used. This means generally a large number of solid elements, nodes, surfaces
Vol. 23, No. 3 Xiangting Su et al.: Complex 3D Static and Dynamic Crack Propagation 275
and CIEs are used. Computer implementation of the above inserting algorithm must be carried out
with care, or the time spent on inserting CIEs may be unduly long. As an example, for an initial mesh
with 46800 C3D8 elements and 52140 nodes, it takes about 30 minutes for the developed MATLAB
program to insert 135,200 CIEs using a PC with an Intel Xeon CPU@3.16 GHz and 3 GB physical
memory.
Figure 5 shows a deformed FE mesh with 774 C3D6 and 1392 C3D8R solid elements, 1854 CIEs
and 6852 nodes in total. It can be seen that the crack path was vertical as expected because of the
symmetry of the mesh and loading conditions. Finer meshes were also modelled but they led to little
dierence from the predicted crack path and the load-crack mouth opening displacement (CMOD)
curves. Figure 6 compares the load-CMOD curves from the experiment and modelling using dierent
solvers and solid elements. The standard solver always failed by divergence at the post-peak stage, e.g.,
it failed at CMOD=3.2 mm for C3D8 as shown in Fig.6. This also often occurred in 2D simulations[6] .
In contrast, the explicit solver always managed to model the whole fracture process when small enough
time increments were used. The explicit solver using C3D8R with reduced integration and hourglass
control (HC) resulted in nearly the same pre-peak responses as the standard solver using C3D8, while
the explicit solver using C3D8R but without hourglass control led to too soft pre-peak responses. C3D8
was also tried with the explicit solver but the computational time nearly doubled when C3D8R was
276 ACTA MECHANICA SOLIDA SINICA 2010
used. Therefore, the explicit solver was used in all the following simulations and C3D8R elements with
hourglass control were chosen when 8-noded brick elements were used.
Fig. 5. Deformed mesh (CMOD=4 mm, scale=100). Fig. 6. Load-CMOD curves: eect of solid element types
and solvers.
Fig. 13. Crack surfaces after failure for the torsion test.
model. The predicted load-CMOD curve is also shown in Fig.14 and it is very close to that from Mesh
B in this study.
3.3. Pull-out Test of a Steel-Embedded Concrete Anchor
This example models the pull-out behaviour of a steel anchor embedded in concrete. The detailed
geometry, applied boundary and loading conditions of one-quarter specimen are illustrated in Fig.15.
This example was also modelled in Ref.[19,22,23] using other methods. The anchor bolt, excluded from
the FE model, was simulated by a boundary condition of vertical displacement imposed on the top
surface of the groove. The concrete properties are: E = 30000 MPa, = 0.2 and = 2.5106 kg/mm3 .
A linear softening curve with Gf = 0.1 N/mm, tensile strength=3 MPa and kn0 = 6 104 MPamm
was used. An analysis time of 0.1 s was used in the explicit solver.
To investigate the mesh-dependence of results, two meshes were modelled. The coarse mesh in
Fig.16(a) has 31889 C3D4 solid elements and 60164 CIEs, and the numbers for the ne mesh in Fig.16(b)
are 69798 and 138169, respectively. Figure 17 shows the load (F )-displacement (d) curves from the two
meshes and other studies. It can be seen that the two meshes predicted very close responses, especially
the same failure point when the load suddenly dropped at d = 0.25 mm.
Similar crack propagation processes and nal crack patterns were predicted from the two meshes.
Figure 18 illustrates four stages of the predicted crack propagation using the ne mesh, where the
cracks are represented by grey areas consisting of highly-damaged cohesive elements (with the damage
index D 0.99). At F = 310 kN (60% of the peak load Fu = 503 kN), tensile cracks initiated at
the top edge of the groove, forming a cone around the groove (Fig.18(a)). As the load approached
the peak, a splitting-mode longitudinal crack initiated at the top surface and propagated in the radial
direction (Fig.18(b)). Beyond the peak load (Fig.18(c)), the longitudinal crack propagated rapidly,
leading to a drastic decline on the load-displacement curve (Fig.17). After the complete propagation
of the longitudinal crack (Fig.18(d)), the conical crack continued to advance with the load decreasing
slowly. The simultaneous propagation of the longitudinal crack and the conical crack, observed in the
experiments by Rots[34] , was well reproduced. This may explain the dierence between the predicted
load-displacement curves (Fig.17) by the present study and other studies[22, 23] which only modelled
the conical crack.
3.4. Impact Test of a Concrete Beam
The fourth example is a concrete beam under
impact tested and modelled in 2D by Du et al.[35] .
3D modelling was reported by Belytschko et al.[36].
The geometry and the boundary condition are
shown in Fig.19. The impact load history obtained
from the test (Fig.20) was input as the loading
condition. The concrete properties are: E = 34480
Fig. 19 Impact test of a concrete beam (unit: mm).
MPa, = 0.2 and = 2.5 106 kg/mm3 . An
exponential softening curve (Fig.21) with the exponential law parameter = 1.5 and the equivalent
dynamic fracture energy Gf = 0.152 N/mm, suggested in Ref.[35], was adopted to model the CIEs.
The initial stiness kn0 = 1 105 MPamm was used.
Figure 22 shows a deformed mesh at the post-peak stage when t = 1 ms. The mesh consists of
3645 solid elements and 606 cohesive elements. Figure 23 compares favourably the predicted load-
deection curve with the test data. Figure 24 compares the measured and computed strain histories
at three locations (SG01, SG02 and SG03). The strain history computed for SG03 agreed well with
the measured. For the other two locations, the strain histories were underestimated, probably due to
the negligence of shrinkage eects in the numerical modelling[37] . The crack-tip extension histories are
compared in Fig.25.
Fig. 24. Strain history for the impact test. Fig. 25. Crack-tip extension history for the impact test.
IV. CONCLUSIONS
A simple yet eective nite element methodology has been developed to simulate 3D complex cohesive
crack propagation in quasi-brittle materials, using Abaquss cohesive elements. An ecient algorithm
to insert the cohesive elements into initial nite element meshes was devised and implemented by an
in-house computer program. Several concrete examples were modelled to demonstrate the eectiveness
of the method. It is concluded that, making use of the rich pre/post-processing functionalities and
powerful standard/explicit solvers in Abaqus, the developed methodology seems promising to provide
a practical simulation tool for modelling realistic 3D fracture in various industries.
References
[1] Scordelis,A.C. and Ngo,D., Finite element analysis of reinforced concrete beams. Journal of the American
Concrete Institute , 1967, 64: 152-163.
[2] Rashid,Y.R., Ultimate strength analysis of prestressed concrete pressure vessels. Nuclear Engineering and
Design, 1968, 7(4): 334-344.
[3] Yang,Z.J. and Chen,J.F., Fully automatic modelling of cohesive discrete crack propagation in concrete
beams using local arc-length methods. International Journal of Solids and Structures, 2004, 41(3-4): 801-
826.
[4] Yang,Z.J., Fully automatic modelling of mixed-mode crack propagation using scaled boundary nite element
method. Engineering Fracture Mechanics, 2006, 73(12): 1711-1731.
[5] Yang,Z.J. and Deeks,A.J., Fully-automatic modelling of cohesive crack growth using a nite element-scaled
boundary nite element coupled method. Engineering Fracture Mechanics, 2007, 74(16): 2547-2573.
[6] Yang,Z.J., Su,X.T., Chen,J.F. and Liu,G.H., Monte Carlo simulation of complex cohesive fracture in random
heterogeneous quasi-brittle materials. International Journal of Solids and Structures, 2009, 46(17): 3222-
3234.
[7] Barenblatt,G.I., The formation of equilibrium cracks during brittle fracture: general ideas and hypothesis,
axially symmetric cracks. Applied Mathematics and Mechanics, 1959, 23: 622-636.
[8] Dugdale,D.S., Yielding of steel sheets containing slits. Journal of Mechanics of Physics and Solids, 1960,
8: 100-104.
[9] Hillerborg,A., Modeer,M. and Petersson,P., Analysis of crack formation and crack growth in concrete by
means of fracture mechanics and nite elements. Cement and Concrete Research, 1976, 6: 773-782.
[10] Wawrzynek,P.A. and Ingraea,A.R., An interactive approach to local remeshing around a propagating
crack. Finite Elements in Analysis and Design, 1989, 5: 87-96.
[11] Xie,M. and Gerstle,W.H., Energy-Based Cohesive Crack Propagation Modeling. Journal of Engineering
Mechanics-ASCE, 1995, 121(12): 1349-1358.
[12] Yang,Z.J. and Xu,X.F., A heterogeneous cohesive model for quasi-brittle materials considering spatially
varying random fracture properties. Computer Methods in Applied Mechanics and Engineering, 2008, 197:
4027-4039.
282 ACTA MECHANICA SOLIDA SINICA 2010
[13] Xu,X.P. and Needleman,A., Numerical simulations of fast crack growth in brittle solids. Journal of the
Mechanics and Physics of Solids, 1994, 42(9): 1397-1434.
[14] Lopez,C., Carol,I. and Aguado,A., Meso-structural study of concrete fracture using interface elements. I:
numerical model and tensile behavior. Materials and Structures, 2008, 41(3): 583-599.
[15] Lopez,C., Carol,I. and Aguado,A., Meso-structural study of concrete fracture using interface elements. II:
compression, biaxial and Brazilian test. Materials and Structures, 2008, 41(3): 601-620.
[16] Camacho,G.T. and Ortiz,M., Computational modelling of impact damage in brittle materials. International
Journal of Solids and Structures, 1996, 33: 2899-2938.
[17] Pandol,A. and Ortiz,M., An ecient adaptive procedure for three-dimensional fragmentation simulations.
Engineering with Computers, 2002, 18: 148-159.
[18] Melenk,J.M. and Babuska,I., The partition of unity nite element method: Basic theory and applications.
Computer Methods in Applied Mechanics and Engineering, 1996, 139: 289-314.
[19] Gasser,T.C. and Holzapfel,G.A., Modeling 3D crack propagation in unreinforced concrete using PUFEM.
Computer Methods in Applied Mechanics and Engineering, 2005, 194: 2859-2896.
[20] Moes,N., Dolbow,J. and Belytschko,T., A nite element method for crack growth without remeshing. In-
ternational Journal for Numerical Methods in Engineering, 1999, 46: 131-150.
[21] Dolbow,J.E., An Extended Finite Element Method with Discontinuous Enrichment for Applied Mechanics.
Ph.D. dissertation, Northwestern University, 1999.
[22] Areias,P.M.A. and Belytschko,T., Analysis of three-dimensional crack initiation and propagation using the
extended nite element method. International Journal for Numerical Methods in Engineering, 2005, 63:
760-788.
[23] Bordas,S.P, Rabczuk,T. and Zi,G., Three-dimensional crack initiation, propagation, branching and junc-
tion in non-linear materials by an extended meshfree method without asymptotic enrichment. Engineering
Fracture Mechanics, 2008, 75: 943-960.
[24] Sun,H.T. and Wang,Y.H., The meshless virtual boundary method and its applications to 2D elasticity
problems. Acta Mechanica Solida Sinica, 2007, 20(1): 30-40.
[25] Wolf,J.P. and Song,C.M., Finite-element Modelling of Unbounded Media, Wiley, Chichester, 1996.
[26] Ooi,E.T. and Yang,Z.J., Modelling multiple cohesive crack propagation using a nite element-scaled bound-
ary nite element coupled method. Engineering Analysis with Boundary Elements, 2009, 33: 915-929.
[27] Abaqus 6.7 User Documentation, Dessault Systems, 2007.
[28] Matlab R2008a Users Guide, MathWorks, 2008.
[29] Trunk,B., Einuss der Bauteilgroesse auf die Bruchenergie von Beton. Aedicatio Publishers, 2000 (in
German).
[30] Feist,C., Numerical Simulations of Localization Eects in Plain Concrete. Ph.D. dissertation, University
Innsbruck, 2003.
[31] Brokenshire,D.R., A Study of Torsion Fracture Tests. Ph.D. Dissertation, Cardi University, 1996.
[32] Jeerson,A.D., Barr,B., Bennett,T. and Hee,S., Three dimensional nite element simulations of fracture
tests using the Craft concrete model. Computers and Concrete, 2004, 1: 261-284.
[33] Gasser,T.C. and Holzapfel,G.A., 3D Crack propagation in unreinforced concrete: A two-step algorithm for
tracking 3D crack paths. Computer Methods in Applied Mechanics and Engineering, 2006, 195: 5198-5219.
[34] Rots,J.G., Computational Modelling of Concrete Fracture. Ph.D. Dissertation, Delft University of Tech-
nology, 1988.
[35] Du,J., Yon,J.H., Hawkins,N.M., Arakawa,K. and Kobayashi,A.S., Fracture process zone for concrete for
dynamic loading. ACI Materials Journal, 1992, 89: 252-258.
[36] Belytschko,T., Organ,D. and Gerlach,C., Element-free galerkin methods for dynamic fracture in concrete.
Computer Methods in Applied Mechanics and Engineering, 2000, 187: 385-399.
[37] Yon,J.H., Hawkins,N.M. and Kobayashi,A.S., Strain-rate sensitivity of concrete mechanical properties. ACI
Materials Journal, 1992, 89: 146-153.