Engineering Fracture Mechanics 74 (2007) 18–33

[Link]/locate/engfracmech

Influence of internal dilation on the fracture behaviour

of multi-phase materials
E. Schlangen *, E.A.B. Koenders, K. van Breugel
Delft University of Technology, Faculty of Civil Engineering and GeoSciences, Microlab,
P.O. Box 5048, 2600 GA DELFT, The Netherlands

Available online 10 March 2006

Abstract

Numerical modelling on the micro- and meso-level of concrete is presented. The formation of the microstructure,
including the matrix and the interfacial transition zones between aggregate particles is modelled explicitly on the micro-
level. The outcome of these simulations are used in a meso-level model (lattice model) to study the formation of
(micro-)cracking in concrete consisting of particles embedded in a matrix. The different matrix and interface properties
as well as the level of restraining of deformations is included in these simulations. Three concrete mixes with different
water–cement ratio are investigated. A lower w/c shows more autogenous shrinkage, higher eigenstresses and thus a higher
tendency for cracking. From the simulations it is found that eigenstresses caused by autogenous shrinkage reduce the ten-
sile strength of the concrete. However, it is also concluded that the local material properties are very important and influ-
ence the outcome of the simulations to a large extent. Furthermore, the influence of microcracking or weak zones on the
ductility of multi-phase materials is investigated. It is found that the ratio between the strengths of the components in a
material determines the width of the fracture process zone and also the ductility of the material.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Fracture simulation; Autogenous shrinkage; Lattice model; Multi-scale modelling; Eigenstress; Ductility; Microcracks;
Concrete

1. Introduction

Concrete is a multi-phase material and is the most used building material for civil engineering structures. It
is relatively cheap, made of natural ingredients and shows a good performance. Most structures made of con-
crete should last for hundred years or more. During the lifetime the structure faces a broad range of loadings,
varying from mechanical load, environmental changes to internal reactions. The components of the heteroge-
neous material have different properties. The way they react on loading varies too. Variation in stiffness
and strength of the components has influence on the global stiffness and fracture behaviour of the material.
Different thermal expansion coefficients of the components result in internal stresses (eigenstresses) in the

*
Corresponding author. Tel.: +31 15 27 86535; fax: +31 15 27 86383.
E-mail address: [Link]@[Link] (E. Schlangen).

0013-7944/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/[Link].2006.01.033
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 19

material when the global temperature changes. In this respect also heat of hydration during hardening and
temperature changes during a fire are important aspects. Differences in porosity of the components influences
the transport and with that also hygral dilation. Different chemical compositions have influence on internal
reactions taking place inside the material, which can also be a function of ingress of species.
The basic properties of the concrete are already formed during the development of the microstructure when
the concrete hardens. During the lifetime of the structure these properties will change due to impact of various
mechanisms. The development of the microstructure of concrete will be different if comparing the bulk part of
the structure and the material in the cover zone. Wall effects, large particles will not be present at the surface,
and also temperature and moisture variations during hardening will influence the formation of the microstruc-
ture in the cover zone. As a result, the properties of the concrete in the cover zone and the core of a structure
are completely different. The concrete properties are also a function of the water–cement ratio (w/c). Lower
w/c gives higher stiffness and strength. On the other hand, it also leads to more internal shrinkage. In this
paper, one aspect of the formation of the microstructure is investigated, i.e. the internal shrinkage and the
resulting eigenstresses. During the hydration moisture will move through the microstructure, simply because
of variations in packing density close to aggregate particles and matrix. This movement of moisture will lead
to autogenous shrinkage and also will influence the hydration and local mechanical properties of the material.
Modelling of this mechanism on the micro-level is performed with HYMOSTRUC [1].
The outcome of this simulation is used as input in the lattice model developed at Delft University to sim-
ulated fracture [2]. The lattice model acts at the meso-level (Fig. 1). The model is extended in order to be able
to apply internal loading which varies in time and study the development of eigenstresses and progressive dam-
age [3].
The outcome of the meso-model could be used to derive material properties for modelling concrete as a
homogeneous material to analyse building elements on a macro level [4].
When fracturing multi-phase materials like concrete a quasi-brittle behaviour is observed. In continuum
mechanics, often the concrete is schematized as a material showing a softening behaviour. This behaviour
can be measured and fitted with some general function which is implemented in the models as a constitutive
law. However, since the early 90s attention has been focussed on the mechanism behind the softening. Exper-
imental studies have shown that if we zoom in on the material, we can explain why we have a softening behav-
iour. A zone of microcracks is formed and crack bridging, branching and friction effects all have a
contribution to the toughness of concrete. The microstructure of the material has a large influence on the
crack patterns in concrete. Cracks follow the weakest link in the material which is often the interfacial
transition zone (ITZ). Transport of gases, moisture and ions takes place through the pore structure of the con-
crete. Also transport follows the easiest route through the microstructure. Larger porosity in the ITZ and also
cracks will enlarge this transport [5]. It has been observed that also (not interconnected) microcracks in a

Fig. 1. Schematic representation of micro- meso- and macro-level modelling.

20 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

material increase the diffusivity, especially when the material is loaded in tension and the microcracks are
open [6].
In this study, the influence of microcracks or weak zones in a material is investigated by simulating tensile
tests with distributions of strengths. In these simulations, the above-mentioned lattice model is used.

2. Towards modelling microstructure

The microstructure of a cement paste consists of a network of hardening cement particles that are glued
together by a variety of hydration products. Modelling this kind of structure formation requires a broad scale
of mathematical formulations and numerical techniques. At Delft University of Technology, a numerical sim-
ulation model has been developed, called HYMOSTRUC [1], which has the potential to calculate the hydra-
tion process of cement-based systems. This model takes into account the formation of the inter-particle
contacts in the hardening cement paste matrix. In the model, hydrating cement particles are considered to
grow gradually in outward direction. In the HYMOSTRUC model, on contact with water an arbitrary cement
particle starts to dissolve under formation of reaction products. These products are formed partly inside and
partly outside the original surface of the cement particle. The outward growing hydration product makes con-
tact with neighbouring particles. Ongoing hydration of the particles goes along with ongoing expansion and
even embedding of neighbouring particles. The embedding of other particles causes an extra expansion and, as
a consequence, embedding of even more particles. An algorithm has been developed for the expansion mech-
anism of continuous expansion and embedding of particles through which the stereological aspect of micro-
structural development can be simulated.
The original modelling work was based on a homogeneous distribution of the particles in cement paste. In
later work, the model has been extended to include randomly stacked particles in a predefined envelope shape,
representing an initial cement paste system. The hydration algorithm of the extended HYMOSTRUC 3D
model [7] is now applied to the randomly stacked particles, calculating the growth of the particles as hydration
proceeds. This forms the basis for the so-called ribbon model. The extended model also provides information
about the capillary pore structure that remains after hydration of the cement particles through the inverse
space that remains from a hydrating cement paste volume. This pore structure is the basis for the moisture
transport calculations that determined the movement of capillary water through the hardening cement paste
system during hardening.

3. The ribbon model

The ribbon model is a three-dimensional model for simulating cement paste between two aggregate parti-
cles. Going from a concrete level to a modelling level (Fig. 2), the hardened cement paste between two aggre-
gates can be represented by a tubular volume, or ribbon. The way of simulating the microstructure in this
ribbon is based on a cement paste volume in which spherical cement grains are stacked randomly in a tubular
volume.
The initial distribution set up this way is, in fact, a representation of the initial stage of the mixed water–
cement emulsion (cement particles in the aqueous phase). The tubular volume is then subdivided in slices, each
of them having a certain water–cement ratio, obtained by dividing the amount of water present in the slice by
the amount of cement present in that slice. For each slice, the hydration process can be simulated with the
HYMOSTRUC software, starting with the initial water–cement ratio of the slice in view.
The generated initial microstructure can be characterized by its geometrical properties, like water–cement
ratio or paste density. This reflects the distribution of the cement grains over the ribbon width, while taking
into account the packing restrictions at the regions close to the aggregate surface. The HYMOSTRUC
simulations for each of the subsequent slices of the ribbon provide information about the evolution of the
microstructure and the associated moisture movements in the subsequent slices. This might influence the
micro-structural properties like porosity, permeability, etc. and influence the movement of moisture during
hardening.
The movement of capillary water through the hardening cement paste system is explicitly modelled within
the HYMOSTRUC 3D model [8]. As hydration proceeds, water is chemically and physically bounded to the
E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 21

Fig. 2. Overview of the ribbon model [8].

22 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 3. Autogenous shrinkage for mix with w/c = 0.3 at three stages of hydration: 20, 100 and 1000 h.

cement while forming hydration products that contribute to the development of a load bearing microstructure.
Capillary water is consumed and taken out of the system, causing unbalances in the capillary surface tensions.
These unbalances are forcing capillary water to move from the water-rich areas (e.g. ITZ) towards the water
lacking areas in the central parts of the cement paste ribbon matrix. This process redistributes the capillary
water in the paste, while influencing the materials properties development.
In addition to this, the volumetric changes of the hardening cement paste system will be affected as well.
Using the phenomenological relation of Bangham [8], the shrinkage of the ribbon paste could be quantified.
The autogenous shrinkage for the mix with w/c is equal to 0.3 is plotted at three stages in the hydration pro-
cess in Fig. 3. The results of the simulations for the three mixes (w/c = 0.3, 0.4 and 0.5) are presented in Fig. 4,
where the autogenous shrinkage (a) and the development of E-modulus (b) is plotted after 20 h of hydration.
The mix with the lower w/c shows the highest shrinkage. From Fig. 3, it can be observed that at the ITZ the
shrinkage is larger, leading to larger micro-structural deformation close to the aggregates. The denser packing
in the central bulk paste is causing a slight increase of the shrinkage in that particular region. It appeared that
a denser initial particle structure is causing higher internal pore pressures, causing larger potential deforma-
tions, but that on the other hand, these deformations are opposed by the higher elastic modulus (Fig. 4).

4. Modelling fracture on a meso-level

Heterogeneous materials have complicated fracture mechanisms, which are related to their microstructure.
The use of linear elastic fracture mechanics to analytically describe these mechanisms is very hard, since the
fracture patterns consist of a main crack, with various branches, secondary cracks and microcracks. To gain
some more insight into the problem the use of numerical tools is a good option.
Studies in theoretical physics indicated that lattice type models can be quite successful for the simulation of
fracture processes in heterogeneous materials, see, for instance [9]. These models were adopted for specific
applications such as to simulate fracture in concrete, see, for instance, [2,10,11]. In these models, a material
is discretized as a lattice consisting of small beam (or spring) elements that can transfer forces (see Fig. 5).
The simulation of fracture is realized by performing a linear elastic analysis of the lattice under loading and
removing (or partially removing [11]) an element from the mesh that exceeds a certain threshold, for example,
strength, strain or energy. The results that are obtained from simulations with lattice models, however, depend
strongly on the used fracture criterion and the chosen element and/or mesh type. To obtain realistic results, it
is important that the relation between the implemented disorder in the model and the heterogeneity of the
material that is to be simulated is as close as possible. Heterogeneity of the concrete is implemented following
various methods [12]:
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 23

Fig. 4. Autogenous shrinkage (a) and development of E-modulus (b) after 20 h of hydration for three mixes, with different w/c.

Fig. 5. Lattice of beam elements (a), definition of forces and degrees of freedom (b), stress–strain relation of beam element (c).

• using a distribution of strengths and stiffness of the elements in the lattice,

• making use of a generated or digital image of a microstructure of the concrete and implementing different
properties to elements representing the different phases in the microstructure,
• using a random orientation of the elements in the lattice.
24 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 6. Stress deformation curve of uni-axial tensile test of concrete, experiment and simulation with lattice model [12].

Details on the elastic equations as well as the fracture procedure of the model are explained in [13]. In this
reference also a description of the fast conjugate gradient solver implement in the Delft Lattice Model can be
found.
In the present paper, a fracture criterion based on normal force in the beams only is adopted. Various frac-
ture criteria are used in the past, all having pros and cons [14]. General trend of all the simulations (in which a
brittle fracture law of the elements is adopted) with the lattice models applied to concrete is up to now that the
crack patterns look very realistic, but the load–displacement response is not correct. In tension, the behaviour
is too brittle and in compression the peak loads are too high and the behaviour is too ductile. An example of
such a result, a simulation of a uni-axial tensile test, is shown in Fig. 6.
Different causes and possible solutions for this too brittle behaviour found in simulations of tensile fracture
are discussed in various papers:

• In the generated (and scanned digital images of) microstructures of concrete the small particles are omitted.
This results in less microcracking at the weak interface between particles and matrix. Smaller sizes of the
beam elements are needed to overcome this problem [12].
• Most of the lattice simulations are performed in 2D, whereas the real fracture in the material is a 3D pro-
cess. 3D lattice simulations are performed [15,16] but show mostly no real improvements, because the size
of the elements is then increased due to computational limitations.
• The fracture law, which decides which element to break, can be altered. The fracture law can consist of dif-
ferent combinations of normal force, bending moments and or shear forces. This only leads to minor
changes in the response [13].
• To create a more global ductile response also a ductile or softening behaviour on the local scale of the ele-
ments can be implemented [10,11,22]. However, doing this the simplicity of the model is mostly lost. Non-
linear iterative steps are needed instead of sequential linear elastic analyses. An exception is the solution
procedure described in [22] which makes use of piecewise linear models to solve the problem.
• In a heterogeneous material like concrete, eigenstresses are present due to variation in properties of the
components. Thermal or hygral variations lead to differences in shrinkage strains which results in eigen-
stresses. This also has an effect of the global material response [17].
• Implementation of a closing pressure could lead to more ductile response. If a crack is formed, this crack
will not close completely simply by removing the load. A closing pressure has to be applied to close the
crack. Implementation of this closing pressure especially will lead to better results is simulations of com-
pressive failure and alternating or fatigue loading [18].
In previous application of the lattice model mostly fracture of materials was studied due to an external
loading or deformation on a specimen. Now attention is focused on the simulation of eigenstresses that
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 25

develop in the material as a result of autogenous shrinkage. The simulation is performed in time. The time
steps, development of E-modulus and shrinkage are taken from the ribbon model described in the previous
section. The meshes (regular triangular lattices) shown in Fig. 7 are used in the simulations. The 2D meshes
have a total width and height of 600 lm. They consist of circular aggregates with diameters between 40 and
200 lm embedded in a matrix. The size of the beam elements in the mesh is 8 lm. The average distance
between the particles, the width of the matrix, is 100 lm. This corresponds with the size of the ribbon in
the ribbon model. The ITZ (arbitrarily taken a width of 15 lm) is plotted in a different colour just to distin-
guish between aggregate and matrix. In reality the material (matrix and the two ITZs) between aggregates is
modelled as a continuous matrix as explained below.
The aggregate is given a stiffness of 70 GPa and a strength of 10 MPa. The matrix is given the stiffness
according to the E-modulus in the various slices obtained from the ribbon model, shown in Fig. 4. The ele-
ments close to the particles get a lower stiffness and the elements far away from the particles a higher stiffness.
In every time step the stiffness is increased. Relaxation is taken into account by assuming that 50% of the stress

Fig. 7. Lattice meshes used; periodic boundary lattice (a), free boundaries (b).
26 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 8. Cracking due to autogenous shrinkage in full restrained lattices (periodic boundaries); w/c = 0.3 (a), w/c = 0.4 (b), w/c = 0.5 (c).
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 27

that occurs in the elements relaxes. This value is taken arbitrarily. However, experiments presented in [20]
indicate that this is the right range. Time effects should be taken into account. Further research is required
to study this effect on a local scale. The tensile strength (ft) is taken proportional to the stiffness values
(ft = E/10,000).
In the simulation of autogenous shrinkage, the loading that is applied on the mesh is the internal shrinkage
of the matrix in every time step that followed from the ribbon model. After applying the shrinkage strain on
the matrix elements in the mesh, the nodal displacements and the corresponding internal forces are redistrib-
uted to satisfy the equilibrium relations [13]. For all the beams the ratio of stress/strength is calculated. The
beam element that has the highest ratio, and in addition a ratio higher than 1, is removed form the mesh. Next,
the deformations are allowed to relax again, after which the next beam is removed. This procedure is repeated
until all beams have a ratio of stress/strength lower than 1. Then the timestep is increased and the stiffness,
strength and strain of all the beams are adjusted. This procedure is repeated for all time steps. In addition
to the simulation of the eigenstresses explained above also simulations of tensile tests are carried out. These
are both on material including and excluding the eigenstresses from the moisture transport during hydration.
Additional tensile test simulations are performed to investigate the influence of weak zones on the ductility of
a heterogeneous material.

5. Analyses with lattice model

5.1. Simulation of shrinkage

Simulation of autogenous shrinkage is carried out on the lattice with periodic boundary conditions. This
case represents the situation in which deformations of the material are fully restrained (like in a large structure
with external restraining).
Applying the shrinkage of the matrix to the mesh results in cracking as shown in Fig. 8. Due to the low
stiffness of the elements in the ITZ close to the particles, at this spot localization of the deformations takes
place.
Eigenstresses are the result of restrained deformations. The strength of the elements determines the forma-
tion of cracks. In the case of the low w/c (w/c = 0.3) first crack localization in the ITZ occurs already in the
early time steps (after 2 h of hydration). After about 50 h of hydration localized cracks in the matrix form,
which splits the mesh in several parts (Fig. 8a). In the simulation of the mortar with w/c = 0.4 cracking of
the ITZ starts at a later stage, after 12 h and localization of cracking in the matrix is also found after 50 h
(Fig. 8b). For the mix with w/c = 0.5 (Fig. 8c) after 1000 h of hydration only some cracking in the ITZ occurs
and no failure of the matrix is found. Experimental evidence of cracking of concrete with low w/c caused by
autogenous shrinkage can be found in many references, see, for instance [19].

Fig. 9. Load–deformation curves in tension after 20 h of hydration for three mixes.

28 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 10. Simulation of tensile fracture; pre-peak (a) at peak load (b); post-peak (c).
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 29

5.2. Simulation of tensile loading

On a mesh with free boundary conditions a tensile loading is applied. The mesh is stretched in the horizon-
tal direction. Fracture is simulated according to the cracking procedure in [13]. Simulations are carried out for
three mixes (w/c ratio equal to 0.3, 0.4 and 0.5) and at three ages (20, 100 and 630 h of hydration).
The results are plotted in Figs. 9–11. In Fig. 9 the load–deformation curves obtained after 20 h of hydration
for the three mixes are shown. The total deformation of the mesh is plotted in the graphs, which results in
snap-backs in the descending branch.
Fig. 10 gives the fractured meshes at pre-peak, peak load and at the end of the softening curve for the mix
with w/c = 0.3 and at 20 h of hydration. However, the cracking in the other simulations, with different w/c and
at different ages show comparable crack patterns.
Fig. 11 shows the development of the tensile strength in time for the three mixes. In the graph, the peak
loads obtained at the three hydrations times (20, 100 and 630 h) are plotted.

5.3. Influence of eigenstress on strength

Autogenous shrinkage of a homogeneous material that is not externally restrained does not lead to stresses,
since the material can deform freely. However, concrete is heterogeneous and as a result internal stresses will
develop. Stiff aggregates restrain the shrinking matrix. The aggregates will be compressed [20,21]. In the
matrix, however, also tensile stresses will be develop as a result of the autogenous shrinkage. The tensile stres-
ses (eigenstresses) found with the lattice model in the three mixes after 20 h of hydration are shown in Fig. 12.
The light colours represent tensile and the dark colours compressive stresses.
The tensile eigenstresses introduce micro cracking in the ITZ in case of the low w/c (w/c = 0.3). Due to these
microcracks in combination with the eigenstresses the tensile strength is reduced by 15% in case of w/c is equal
to 0.3 (see Fig. 13). This is found from a tensile test simulation which is performed on the mesh after the eigen-
stresses due to autogenous shrinkage (Fig. 12a) are calculated. For higher w/c the reduction in tensile strength is
less. For w/c is equal to 0.4 only a small change and for w/c is 0.5 no change in tensile strength is found. Note
that these simulations are only performed after 20 h of hydration. At other time steps the results can be differ-
ent. Shrinkage, stiffness and strength change with ongoing hydration. The change is not proportional for all the
elements located in the different slices of the ribbon between aggregates. Furthermore, the difference between
matrix and aggregates changes, because the properties of the aggregates do not change with hydration.

5.4. Simulation with distribution of strength

To investigate the influence of low strength zones (or microcrack/eigenstresses) on the ductility of concrete,
simulations on a lattice under tensile loading are performed. The boundaries where the load is applied are

Fig. 11. Development of tensile strength in time for three mixes.

30 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 12. Eigenstresses in matrix after 20 h of hydration; w/c = 0.3 (a), w/c = 0.4 (b), w/c = 0.5 (c); light colours = tensile stress, dark
colours = compressive stress.
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 31

Fig. 13. Load–deformation curves for tensile test with and without eigenstress for w/c = 0.3 after 20 h of hydration.

non-rotating. For all the simulation a regular triangular mesh is used. As input for the simulation the follow-
ing values are used for the beams: Young’s modulus E = 30000, thickness b = 1, height h = 1, length l = 1.
The strength of the elements in the lattice is varied. In the lattice, elements with two strengths are randomly
placed; 25% of the elements have a low strength, and 75% of the elements have a high strength. The value of
25% is taken to be below the percolation threshold value. The low strength has always a value equal to 1. For
the high strength the values are equal to 2, 3, 3.25, 4 and 5 respectively.
If the simulations with the variation in high strength values are compared, some interesting observations
can be made. First of all the maximum load that can be put on the lattice is the same for all simulations (except
for the highest strength). This means that the maximum load is determined by the low strength elements. The
ductility of the global response increases if the high strength value increases (see Fig. 14). The amount of
microcracks in the network, before localization starts, also increases with increasing high strength value
(see Fig. 15). This means that the ratio between the strengths in the lattice determines the width of the zone
with microcracks (fracture process zone) and also the ductility of the global response. In the case of high
strength value equal to 5, first all the low strength elements crack, after which the maximum load on the net-
work has to be increased again to get failure. For the latter some resemblance with (fibre) reinforced concrete
can be made, where first cracking of the concrete takes place after which the reinforcement will fail (due to pull
out, plasticity or breaking of the reinforcement).

100

2
75
3
3.25
Load

50 4
5

25

0
0 0.001 0.002 0.003 0.004 0.005
Deformation

Fig. 14. Load–deformation curves of simulation of tensile fracture in which the strength ratio of the elements is varied.
32 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33

Fig. 15. Cracked and deformed lattices after simulation of tensile fracture with variation of strength ratio; Top: (from left to right)
strength ratio is 2.0, 3.0 and 3.25; Bottom: (from left to right) strength ratio is 4.0 and 5.0.

6. Discussion and conclusions

In the paper, simulations on two levels of concrete are presented. The outcome of a lower level is used as
input at a higher level. Modelling the complete process of formation of microstructure, building up of eigen-
stresses, formation of local damage and structural performance in a single model is still not possible due to
computational limitations.
Furthermore, it is also not practical to model the complete mechanism with a single model. Zooming in on
a level and studying the processes in more detail will increase the insight in the mechanism taking place.
The formation of the microstructure with HYMOSTRUC and the movement of moisture with the ribbon
model resulted in local information of stiffness and strain in ITZ and matrix. Simulations for three mixes with
different w/c are performed. Lower w/c gives higher autogenous shrinkage, but also higher stiffness and
strength.
This information is used as input in the lattice model to study deformations, eigenstresses and damage. The
simulations with the lattice model have shown that autogenous shrinkage leads to cracking already at an early
stage of the hydration for mixes with a low water–cement ratio. A requirement for this to happen is the full
restraining of deformations. In the case of free boundaries, however, internally a restraining will be present
due to the stiff particles embedded in the matrix, which leads to (tensile) eigenstresses in the matrix. Further-
more, some eigenstresses in the material will always be present. This can either be externally or internally by
temperature or moisture gradients or expansion mechanisms due to chemical reactions in the material. The
tensile strength is reduced by these eigenstresses and thus the material has a lower load bearing capacity.
Cracks patterns as obtained in this study are of course found in many experimental investigations. How-
ever, the exact shape of the crack patterns (ITZ or localized matrix cracks) depends on the amount of eigen-
stress and the local properties, such as strength, stiffness and relaxation. The determination of local properties
needs to be worked out. Nano-indentation seems to be a promising approach for this. This technique com-
bined with fracture tests and inverse modelling can help determining local material properties of mortars
and concrete. The results obtained from the simulations presented in this paper are qualitative results. If reli-
able local properties are available it is possible to obtain quantitative results with the kind of modelling
 E. Schlangen et al. / Engineering Fracture Mechanics 74 (2007) 18–33 33

presented in this paper, which opens the way to better predict fracture mechanisms and get more insight in
durability issues.
In the simulations it is further assumed that bonds which break in the mesh will always form a crack. If
these cracks are formed at an early stage of hydration, there is a chance that due to ongoing hydration and
formation of microstructure these cracks will heal. The strength, stiffness but also the amount of shrinkage
at these locations is then unknown and needs further research.
From the simulations of the uni-axial tensile loading with distributed strengths in the lattice it is found that
the ratio between the strengths determines the width of the fracture process zone and also the ductility of the
material. In the simulations with two strength values in the lattice the lowest strength determines the maxi-
mum load. In real materials (e.g. concrete) the strength values will not be constant but follow a certain dis-
tribution. Taking this into account in simulations will definitely result in more smooth load–displacement
responses.

References

[1] van Breugel K. Simulation of hydration and formation of structure in hardening cement-based materials, PhD thesis. Delft University
of Technology, Delft, The Netherlands, 1991.
[2] Schlangen E, van Mier JGM. Experimental and numerical analysis of micromechanisms of fracture of cement-based composites. Cem
Concr Compos 1992;14:105–18.
[3] Schlangen E, Koenders EAB, van Breugel K. Multi-scale modelling of crack formation in the concrete cover zone. In: Lange DA
et al, editors. Advances in cement and concrete, Proc ECI conference, Copper Mountain, Co, 10–14 August 2003. p. 281–91.
[4] van Beek A, Schlangen E. Simulating the effect of shrinkage on concrete structures. In: Baroghel-Bouny V, Aı̈tcin P-C, editors.
Shrinkage 2000, vol. 17, Rilem PRO 2000. p. 477–92.
[5] Jacobsen S, Marchand J, Boisvert L. Effect of cracking on chloride transport in OPC concrete. Cem Concr Res 1996;26(6):869–81.
[6] François R, Arliguie G. Effect of microcracking and cracking on the development of corrosion in reinforced concrete members. Mag
Concr Res 1999;51(2):143–50.
[7] Koenders EAB. Simulation of volume changes in hardening cement-based materials, PhD thesis. Delft University of Technology,
Delft, The Netherlands, 1997.
[8] Koenders EAB, Ye G, van Breugel, K. Numerical simulation of moisture flow through hardening cementitious cover. In: Finger FA,
editor. Proc. IBAUSIL, vol. 2, 24–27 September, Weimar, 2003. p. 917–26.
[9] Herrmann HJ, Roux S, editors. Statistical models for the fracture of disordered media. North Holland, Amsterdam: Elsevier; 1990.
[10] Bolander JE, Saito S. Fracture analysis using spring networks with random geometry. Engng Fract Mech 1998;61(5–6):569–91.
[11] Ince R, Arslan A, Karihaloo BL. Lattice modelling of size effect in concrete strength. Engng Fract Mech 2003;70:2307–20.
[12] Schlangen E. Experimental and numerical analysis of fracture processes in concrete, PhD thesis. Delft University of Technology, The
Netherlands, 1993. p. 121.
[13] Schlangen E, Garboczi EJ. Fracture simulations of concrete using lattice models: computational aspects. Engng Fract Mech 1997;
57(2/3):319–32.
[14] Mier JGM van. Fracture processes of concrete. New directions in civil engineering. Boca Raton, FL: CRC Press; 1997.
[15] Schlangen E, van Mier JGM. Fracture simulations in concrete and rock using a random lattice. In: Siriwardane HJ, Zaman MM,
editors. Computer methods and advances in geomechanics. Rotterdam: Balkema; 1994. p. 1641–6.
[16] Lilliu G, Van Mier JGM. 3D lattice type fracture model for concrete. Engng Fract Mech 2003;70(7/8):927–41.
[17] Schlangen E, Koenders EAB, van Breugel K. Formation of eigenstress and cracks due to autogenous shrinkage. In: Li VC, Leung
CKY, Willam KJ, Billington SL, editors. Fracture mechanics of concrete structures 2004. p. 447–54.
[18] Schlangen E. Influence of microcracking on ductility of multi-phase materials, ICF XI, Turin, Italy, 20–25 March, 2005.
[19] Baroghel-Bouny V, Aı̈tcin P-C, editors. Shrinkage 2000, Rilem PRO 17, 2000.
[20] Dela B, Stang H. Two dimensional analysis of crack formation around aggregates in high-shrinkage cement paste. Engng Fract Mech
2000;65:149–64.
[21] Stang H, Ostergaard L. Cracking and fracture in early age concrete. In: Lange DA et al., editors. Advances in cement and concrete.
Proc. ECI conference, Copper Mountain, Co, 10–14 August 2003. p. 257–68.
[22] Jirásek M, Bažant ZP. Macroscopic fracture characteristics of random particle systems. Int J Fract 1995;69:201–28.