Noname manuscript No.
(will be inserted by the editor)
Tensile failure of micro-concrete: from mechanical tests to FE
meso-model with the help of x-ray tomography
Olga Stamati · Emmanuel Roubin · Edward Andò · Yann Malecot
Received: date / Accepted: date
Abstract In this work, concrete is studied at meso-
scale (aggregates, macro-pores and mortar matrix), where
the local failure mechanisms are known to drive the
macroscopic behaviour of the material. In order to high-
light the impact of the mechanical and morphological
properties of each phase (along with their interfaces),
micro-concrete specimens are prepared with rather small
dimensions compared to the size of the heterogeneities.
X-ray tomography is used to reliably obtain the mor-
phology of the heterogeneous meso-structure, which is
then given as an input to a 3D FE meso-model with
enhanced discontinuities. A uniaxial tensile numerical
simulation is performed as a first application. To vali-
date the numerical model, a uniaxial tensile test of the
same micro-concrete specimen is performed inside the
x-ray scanner and the in-situ evolution of the micro-
structure is followed. Thus, both a direct validation of
the model and a valuable insight of the 3D fracture
mechanisms while the load progresses are obtained. Af-
ter identification of the numerical parameters, compar-
ison of experimental and numerical results reveals the
capability of the meso-model to reproduce the actual
material response (in terms of macroscopic strength,
Young’s modulus and fracture patterns), with the ex-
plicit representation of the meso-scale heterogeneities
being its key feature. To further challenge the meso-
model, a new morphology coming from an x-ray scan of
another characteristic micro-concrete specimen is intro-
duced and its macroscopic behaviour is computed with-
out a priori numerical identification. Starting from an
x-ray scan in meso-scale, it is shown that the 3D meso-
O. Stamati
Univ. Grenoble Alpes, CNRS, Grenoble INP, 3SR, F-38000
Grenoble, France
E-mail: olga.stamati@3sr-grenoble.fr
model is capable to predict the macroscopic behaviour
and the failure patterns of the material.
Keywords meso-scale FE modelling · micro-concrete ·
tensile failure · local failure mechanisms · morphological
description · x-ray tomography · in-situ evolution ·
Digital Volume Correlation
1 Introduction
Being a manufacturing material widely used in the world,
the mechanical behaviour of concrete has been investi-
gated over decades. However, its failure mechanisms are
complex phenomena and still constitute a very active
area of research. The quasi-brittle behaviour observed
at the macro-scale is strongly influenced by the mor-
phology and the properties of the material constituents
and their mutual interactions over a large range of scales:
from nano- to meso-scale, referring to nm (hydrated ce-
ment scale) and cm (largest aggregates scale), respec-
tively (2; 38; 19).
At the meso-scale, concrete can be viewed as a three
phase geomaterial, constituted of aggregates and macro-
pores embedded within a mortar matrix (mix of ce-
ment paste and fine sand). Macro-porosity refers to the
pores being much larger than the capillary voids voids
(50 nm). It corresponds to both the entrained air voids
(usually ranging from 50 to 500 µm) and the entrapped
air voids (almost as large as the largest aggregates,
depending on the concrete workability). The capillary
porosity, referred to as micro-pores, plays an impor-
tant part in dessication, shrinkage and creep, whereas
macro-pores are more influential in determining the strength
and permeability of the material (14).
The presence of both aggregates and macro-pores in
the concrete mix, with various sizes and shapes, results
2 Olga Stamati et al.
to an heterogeneous stress field at meso-scale, even un-
der a uniform loading, leading to stress concentrations
and initial micro-cracking around the weakest regions
(usually the interfaces). With the increase of the load-
ing, these micro-cracks growth and coalescence leads
to a critical macro-crack resulting to the failure of the
material.
Consequently, the meso-scale is a scale of interest
to study the local failure mechanisms of concrete and
therefore, in recent years, many numerical meso-models
have been developed. These models, by explicitly repre-
senting the meso-structure (particularly in 3D), could
be an effective and practical alternative to experimen-
tal studies. They can be used to reveal the influence of
the meso-scale heterogeneities on the global response
by easily changing global descriptors, such as phase vol-
ume fractions, aggregates and macro-pores size distri-
butions, aggregates shape, phase or mechanical proper-
ties of each phase. A multi-scale approach could also be
achieved by using the meso-model as a constitutive law
at the macro-scale.
These numerical approaches can fit within the con-
text of continuum Finite Element models (FEM) (31;
12; 7; 25), Discrete Element models (DEM) (13; 5; 28)
or Finite Element models enhanced with discontinuities
(3; 24), which is the approach followed in this work. No
matter the selected approach, two main aspects have
to be considered in these models; the morphological de-
scription of the meso-structure and how to account for
the quasi-brittle behaviour of the material.
Concerning the former, the most common method is
the use of morphological models, such as sphere packing
(5; 13; 35) or excursions of correlated Random Fields
(23). However, these methods have limitations regard-
ing both the representativeness of the generated mor-
phology and the possibility of a quantitative validation
of the models with experimental results. To overcome
this, in recent years, real meso-morphologies are ob-
tained by taking advantage of recent advances in non-
destructive 3D imaging combined with image analysis.
This is the approach, in particular the use of x-ray to-
mography (20; 21; 16; 22), followed in this work.
Once the meso-structure morphology is reliably ob-
tained, the question that follows is how to account for
its complex heterogeneous aspects (explicit representa-
tion of aggregates and macro-pores) and how to rep-
resent the quasi-brittle behaviour of the material. In
the context of the E-FEM (Embedded Finite Element
Method), adopted in this study, two sets of local dis-
continuities are introduced inside each element: in the
strain field (accounting for the difference in the material
properties between the phases) on the one hand, and in
the displacement field (accounting for micro-cracking),
on the other hand.
Having introduced the basic characteristics of the
meso-model, the question that then arises is the valida-
tion and the predictive ability of the adopted numerical
approach. Therefore, in this work, the development of
a suitable experimental set-up and the direct compar-
ison of the numerical outcomes with the experimental
results is done.
In short, advanced numerical and experimental tech-
niques are used in this paper to get a better understand-
ing of the local failure mechanisms of concrete at meso-
scale. The main originality comes from conducting in-
situ direct tensile tests on micro-concrete specimens of
realistic composition (including cement, sand, aggre-
gates and water). The evolution of the fracture pro-
cesses as the load progresses is followed, by mapping the
relative deformations between consecutive x-ray 3D im-
ages. The construction of a three-phase 3D meso-model
based on the morphology obtained from the initial x-ray
scan and the direct experiment-modelling comparison,
at meso-scale, during the loading are also rare in liter-
ature ((16; 37; 10)).
The detailed outline of the article is the following:
First, micro-concrete specimens are prepared with rather
small dimensions compared to the size of the hetero-
geneities (Section 2.1). X-ray tomography is used to
obtain a realistic repartition of the phases (aggregates,
macro-pores and mortar matrix) (Section 2.2). The re-
sulted segmented x-ray 3D image is then used as an in-
put to a 3D FE meso-model and a uniaxial tensile test
is simulated (Section 3). For validating the numerical
model, a suitable experimental set-up, compatible with
the x-ray apparatus, is developed (Section 4). Thus, by
following the in-situ micro-structure evolution both a
direct validation of the numerical model and an insight
of the 3D fracture mechanisms during loading and up
to failure are achieved. After parameter identification,
comparison of experimental and numerical results show
that the meso-model is capable to reproduce both the
material response and the cracking patterns (Section
5). For a further validation of the 3D meso-model, an-
other micro-concrete specimen with a new morphology
is tested and the model is used to predict its macro-
scopic behaviour without new identification (Section 6).
2 Studied material: scale and morphology
2.1 Characteristic sizes of the specimens
To investigate the local failure mechanisms in concrete
at the meso-scale, specimens are required with signif-
icant heterogeneities which will give insights into the
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography
impact of both the mechanical and the morphological
properties of each phase (aggregates, macro-pores, mor-
tar matrix) along with their respective interfaces on the
macroscopic response of the material.
A number of factors affect the choice of the specimen
size:
– The specimen must be of a sufficient size with re-
spect to the largest heterogeneities so that it has
some mechanical resistance and some relevance to
real concrete.
– The use of x-ray tomography to obtain an image of
the micro-structure representation, implies a trade-
off between field-of-view (i.e., specimen size) and
spatial resolution (i.e., the smallest heterogeneity
that can be identified). In this case, the smallest
heterogeneities are the smallest macro-pores (size
of about 100 µm). This encourages a pixel size of
at least 20 µm px−1 in order to be able to describe
these pores coarsely thus limiting the field of view
to around 30 mm.
– The specimens are mechanically loaded in-situ in
the x-ray scanner, meaning that their dimensions
and failure force must be in agreement with the
limitations of the mechanical loading system. The
current limitation of the loading system used herein
is 500 N in tension (and 10 kN in compression) for
20 mm in diameter.
– The FE mesh that explicitly represents the hetero-
geneities of the material imposes additional con-
strains to the specimens size, due to computational
limitations. Particularly, the characteristic element
size in the FE mesh should be as small as the small-
est heterogeneities.
– Another characteristic scale to be chosen is the size
of the largest heterogeneities (i.e., size of largest ag-
gregates). These aggregates have to be large enough
compared to the size of the specimen to give some
direct effect on the macroscopic behaviour so as to
easily test the predictive ability of the meso-model.
– The points above point towards a small specimen
size, meaning that a micro-concrete is automatically
selected. In preliminary tests performed, it was not
possible to core a specimen smaller than 10 mm.
Making a trade-off between the difference constraints,
cylindrical specimens are chosen of 11 mm in diameter
and 23 mm height. Such a size, considering the tensile
concrete strength (about 3.5 MPa) means a tensile fail-
ure load that should not exceed 350 N which is com-
patible with the system mentioned above. The largest
aggregates are chosen with a characteristic size of 3 mm
(about the quarter of the specimen diameter) which is
both quite large to hope some discrepancy between dif-
3
ferent specimens but also small enough to have a be-
haviour that is still representative of concrete.
2.2 Specimens preparation and scanning
A regular concrete mix is prepared, using ordinary Port-
land cement, aggregates, fine sand (D=1800 µm) and
water. The mix proportion is 1:3:3.8:0.6, by weight of
cement: sand: aggregates: water. The aggregates used
are rolled and siliceous (chemical composition: SiO2 >
97.3%) coming from Mios (France), with a maximum
size of 3 mm. The material is mixed with a benchtop
rotary mixer, casted into blocs of about 10 cm3 in vol-
ume and cured in wet conditions for seven days. The
cylindrical specimens are extracted from the material
bloc using a 11 mm inside diameter diamond coring bit.
The extracted cylindrical core is then cut to a nominal
24 mm length with a diamond wire and both surfaces
are rectified, resulting to the specimen to be tested.
The voltage and current of the x-ray source are set
to 125 kV and 80 µA, respectively, and the generated x-
ray beam is polychromatic of a conical shape. With this
scanning apparatus, the voxel size is 13 µm achieving a
spatial resolution that captures efficiently smallest con-
sidered heterogeneities of the specimen (about 0.1 mm).
Projections are acquired in 1120 different angular posi-
tions between 0◦ and 360◦ , as the specimen is rotated
around a vertical axis and six images of each angle are
averaged in order to reduce the noise. By assembling
the 2D radiographs using a filtered back projection al-
gorithm available in the XAct software provided by RX-
Solutions (Annecy, France), the reconstructed 3D im-
age of the scanned specimen is obtained, representing a
field of x-ray attenuation coefficients, as shown in Fig.
1a.
The three phases of the generated meso-structure
can be recognized, with macro-pores appearing black
(lower x-ray attenuation), while coarse aggregates and
mortar matrix (mix of finer aggregates and sand) shar-
ing shades of grey (higher x-ray attenuation). This grey
scale 3D image has necessarily to pass through a series
of filters, as described in details here (27), to be trans-
formed into a segmented discrete image where each
phase is identified, as shown in Fig. 1b.
The obtained repartition of phases can now be used
as the input morphology of a FE meso-model, as de-
scribed in details in the following section.
4 Olga Stamati et al.
Fig. 1: Reconstructed 3D image coming from the x-ray scan of the micro-concrete specimen in the initial configu-
ration: (a): grey scale image, (b): segmented image
3 FE meso-model description
From a numerical point of view, the main challenges to
model the quasi-brittle behaviour of materials such as
concrete at the meso-scale are:
– explicitly take the morphology into account,
– model the local discrete failure mechanisms.
In this study, a FE model that tackles both issues within
a framework of embedded kinematics enhancement is
used. Only the main ingredients of the model are pre-
sented here; for a detailed description see (24).
3.1 Local kinematics enhancement
3.1.1 Accounting for the morphology
A first challenge for a meso model consists in taking
into account the morphology of the meso-structure.
The most commonly used method to address this
issue, referred to as adapted mesh method, roughly con-
sists in: first, a surface meshing of the interfaces and
then a volumetric meshing of each phase. This results
in very accurate representation of the morphology, re-
quiring though complex and time costly algorithms (see
(26)). Furthermore, particular attention has to be paid
on the interface modelling (zero thickness elements or
cohesive bands for example).
A lighter framework, referred to as non-adapted mesh
method (15), is used here. First of all a FE mesh (made
of 4-noded tetrahedra) is generated regardless of any
morphology (i.e., with randomly and evenly distributed
nodes in space) (2b). Then the morphology resulting
from the segmentation above (2a) is projected onto this
mesh. This consists in inspecting each element and iden-
tifying what phase of the morphology each node falls
into. Based on this identification, two kinds of elements
can emerge: elements where all the nodes belong to the
same phase and elements where nodes belong to two dif-
ferent phases. The first kind are classical FE where the
material properties of the phase is defined. The second
kind needs to account for two phases and the interface
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography
between them. In this work, this is handled by a kine-
matics enhancement, called weak discontinuities. This
kinematic enhancement accounts for the material dis-
continuity by introducing a jump in the strain field (29),
accounting for the interface. See (24) for more details.
In the examined case, the result of the morphologi-
cal projection leads to two different kinds of elements:
those that have all their nodes within the same phase
(macro-pores, aggregates or mortar matrix) and those
that have nodes in two different phases. It is the later
ones that are enhanced by a weak discontinuity to ac-
count for the differences in the material properties of
the two phases. This can be written for an element as:
ε = Bd + Gw [|ε|] (1)
where the first term is the classical FE strain field (deriva-
tive of the shape functions B and nodal displacements
d) and the second term corresponds to the weak discon-
tinuity. [|ε|] is a vector of new unknowns, i.e. the weak
discontinuities, and Gw an operator that represents the
interface geometry. This geometry accounts for the vol-
ume fraction of each phase along with the orientation
of their respective interface, both being computed dur-
ing the projection, regardless of any mechanical state
(details here (24)).
As shown in Fig. 2, the projection leads to five sets
of FE, three with no weak discontinuities (inside ag-
gregates, mortar matrix or macro-pores) and two with
weak discontinuities accounting for aggregates/mortar
and pores/mortar along with their respective interfaces.
3.1.2 Accounting for the local failure mechanisms
The local failure mechanism considered is micro-cracking,
being the key phenomenon, with the coalescence and
bridging of these micro-cracks resulting to the macro-
scopic failure of the specimen. In order to account for it,
another set of local discontinuities is introduced. Simi-
larly with the weak discontinuities, the FE kinematics
are enhanced, this time, by a jump within the displace-
ment field, thus representing the discontinuous aspect
of a crack. This new discontinuity is referred to as strong
discontinuity and can be written for one element as:
ε = Bd + Gs [|u|] (2)
where the first term is still the classical FE strain and
the second one corresponds to the strong discontinu-
ity. [|u|] is directly the crack opening, a vector of new
unknowns, and Gs an operator that represents the po-
sition and the orientation of the crack.
5
3.1.3 Double kinematics enhancement
The two enhancements presented above can be com-
bined together leading to a formulation within an ele-
ment written as:
ε = Bd + Gw [|ε|] + Gs [|u|] (3)
| {z } | {z }
weak disc. strong disc.
This permits a material interface (weak discontinuity)
to fail (strong discontinuity), thus modelling debond-
ing. Fig. 3 shows the three kinds of discontinuity set-
tings that can occur inside an element.
It is assumed here, that if both discontinuities are
present inside an element, the strong discontinuity ori-
entation is taken to be the same as the weak one, thus
making the crack open following the material geometry.
3.2 Phenomenology and resolution scheme
The double kinematics enhancement brings two addi-
tional sets of unknowns in the mechanical problem, one
being the weak discontinuity vectors [|ε|] and the other
the crack-openings [|u|]. Thus the behavior law can be
written as a function of them. It is assumed here that
the relationship between stress and strain remains elas-
tic (no damage within the material), leading to the fol-
lowing behaviour law:
σ(d, [|ε|], [|u|]) = Cε = C (Bd + Gw [|ε|] + Gs [|u|]) (4)
where C contains the elastic material properties of each
phases.
It is important to note here that the stress soften-
ing does not come from material degradation (through
modification of C), but through kinematics considera-
tions, i.e., a crack opening (modification of [|u|]). The
energy is thus dissipated through the discontinuity which
plays the role of softening regularisation.
The phenomenology of the failure mechanism fol-
lows the Discrete Strong Discontinuity Approach (18),
which links the crack opening with the stress state through
a traction-separation law. This discrete approach is ac-
tivated when a certain stress state within an element
is reached; the tensile strength σy . This is expressed by
the following localisation criterion:
Φl = T · n − σy (5)
where n is the discontinuity orientation and T = σ · n
is the traction vector at the discontinuity surface. As
mentioned above, in case of weak discontinuities, n is
considered to be the geometrical orientation (material
interface). In case of no weak discontinuities, n is con-
sidered to be the direction of the largest principal stress.
6 Olga Stamati et al.
Fig. 2: Non-adapted mesh method: (a) the heterogeneous morphology (segmented 3D image) is projected into a
FE mesh (c) leading to 5 sets of FE
(a) (b) (c)
Fig. 3: Adapted from (11), element with: (a) one weak
discontinuity, (b) one strong discontinuity and (c) both
weak and strong discontinuity
In opposition to rotating crack models, here, when the
localisation occurs (Φl = 0), the orientation is recored
and remains fixed.
Then, the following criterion drives the opening:
 
σy
Φo = T · n − σy exp − [u] (6)
Gf
where Gf is the fracture energy of the phase and [u] =
k[|u|]k is the norm of the crack opening vector, which
leads to a local brittle behaviour. Since this law provides
information only on the norm [u] of the crack vector
[|u|], it is assumed that the orientation of this vector is
the same as n, thus corresponding to a mode I opening.
Following the Embedded FE formulation (17; 33;
24), the equations added by the presence of these en-
hanced parts of the strain field are solved at the el-
ement level, by performing a local resolution. Then a
static condensation (34) allows the system to keep the
same size as without enhancements (i.e., there are no
additional global unknowns). This specific feature of
the embedded formulation allows for constant systems
sizes no matter the number of cracks or the complexity
of the geometry.
The presented framework is implemented in the FE
code FEAP (30) and the implemented quasi-Newton
BFGS resolution scheme coupled with an iterative solver
(Conjugate Gradient) is used to solve the non-linear set
of equations of the global system.
3.3 A first application of a uniaxial tensile test
Now that the basic characteristics of the model have
been introduced, a first application to the micro-concrete
specimen follows. The free software GMSH (6), based
on a 3D Delaunay triangulation of randomly positioned
nodes, is used to create the FE mesh. In the studied
case, this is a cylinder of the dimensions of the micro-
concrete specimen consisting of 4-node tetrahedra ele-
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography 7

ments. In order to model the morphological information

of the smallest heterogeneities of the 3D segmented im-
age, the edge length of each tetrahedron is chosen to be
0.1 mm generating a mesh of 6.3 × 106 4-node tetrahe-
dra elements and 3.4×106 degrees of freedom, as shown
in Fig.2b.
As explained in the previous subsection, a linear-
elastic continuum relationship links strain and stress
fields outside the discontinuity, while for the disconti-
nuity surface a basic single traction and brittle softening
criterion is used. Thus, for each of the three phases (ag-
gregates, mortar matrix, macro-pores) two elastic and Fig. 4: Stress-strain curve of a uniaxial tensile numerical
two failure parameters are required: the Young’s modu- simulation along with the displacement field in the axial
lus E and the Poisson ratio ν, and the yield stress (σy ) direction at the end of the simulation
and the fracture energy (Gf ), respectively. Regarding
the elements with shared material properties (enhanced
with a weak discontinuity), no specific parameters are
scopic displacement) are illustrated on Fig. 4 along with
required in the elastic regime, whereas specific failure
the displacement field in the axial direction at the end
parameters are defined. Equal failure properties with
of the simulation. The discontinuity localisation in the
the mortar matrix are assigned to these elements, still
displacements in a roughly horizontal “plane” (perpen-
leading to a weaker behaviour of the interface, since the
dicular to the loading direction) forms a single macro-
difference of the elastic properties between the phases
crack, typical for a mode I failure in tension. Even
results to a higher stress concentration at the interfaces.
though the characteristic quasi-brittle behaviour of the
Here, aggregates are considered elastic, i.e., no crack
material has been indeed numerically reproduced, the
initiation and propagation allowed inside these elements.
computed tensile strength and Young’s modulus, as well
Regarding the macro-pores, the physically required neg-
as the reliability of the failure patterns is questionable.
ligible (zero) stiffness, would potentially cause numer-
For this reason, the actual mechanical response of the
ical problems and further mechanical influences, due
scanned micro-concrete specimen under uniaxial ten-
to the high contrast in material properties among the
sion will be experimentally examined, as explained in
phases. Therefore, elements that after the projection
details in the following section.
are identified as macro-pores, have been removed from
the final FE mesh, creating some cavities in the mesh.
On the inside of these cavities, a free surface bound-
ary condition is applied, consequently these nodes can-
not carry any load which does not attract any spuri-
ous stress concentrations. In the case where nodes are
both in a macro-pore and a solid phase, the choice 4 In-situ tensile test
has been made to consider the full element as part of
the solid phase. A sensibility analysis (not presented Using x-ray tomography on both intact (before load-
here) has been made showing a negligible effect of this ing) and damaged (after unloading) materials is quite
choice over considering weak discontinuities with a very valuable for obtaining the internal micro-structure be-
soft material that would represent macro-pores. Taken fore and after failure. However, scanning the internal
from bibliography, the assigned material properties are: micro-structure during loading can reveal important in-
Ea = 70 GPa, Em = 12 GPa, νa = νm = 0.2 and formation about the actual evolution of the fracturing
σym = 6 MPa, Gfm = 0.1 Jm−2 . process (4; 36; 16). Specific experimental facilities are
A first mechanical application of a uniaxial tensile required for these in-situ studies, though, making them
test is then performed. The minimum constrains for less commonly used. In this work, a suitable experimen-
blocking the rigid body motion are applied and a dis- tal set-up compatible with the x-ray scanner of Labo-
placement control computation is performed along the ratoire 3SR is developed, allowing the micro-concrete
vertical axis of the cylindrical mesh, while the perpen- specimen to be scanned while it is under load. This
dicular directions are stress free. permits both a direct validation of the 3D E-FE meso-
The macroscopic normal stress as a function of the model presented in previous section and an insight into
macroscopic axial strain (coming from the imposed macro- the 3D fracture mechanisms while the load increases.
8 Olga Stamati et al.
(a)
Fig. 5: In-situ tensile test (a): Sketch of the designed load frame (b):In-situ x-ray facility at Laboratoire 3SR
4.1 In-situ experimental set-up
Fig. 5a shows a sketch of the loading cell, along with
its key features and Fig. 5b shows a photo of the load-
ing cell inside the in-situ x-ray facility. The load frame
is built into a low x-ray absorption plexiglass cell de-
signed for triaxial testing of rocks with 7 MPa confining
pressure. The two surfaces of the specimen are glued in
two specifically designed aluminum caps with a high
strength epoxy. The shape of the caps is designed (see
Fig. 5a) to ensure good centering and to avoid any stress
concentration or failure on the axial faces of the speci-
men. In order to ensure that the applied load remains
purely axial, each aluminum cap is screwed into a ball-
joint, allowing free rotation in the two planes normal to
the loading direction and eliminating any bending mo-
ment due to potential defects in parallelism. The bot-
tom ball-joint is screwed into a ram, while the top one
remains fixed into a ball-joint holder as shown in the
sketch. Once the specimen is mounted into the loading
cell, a uniaxial tensile load can be applied to its bot-
tom surface through the ram, while an LVDT sensor is
used to measure the displacement. Since the shape of
the x-ray beam is conical, a displacement of the loading
cell closer to the source, makes the specimen enlarged
on the detector (due to geometric magnification) and
the desired spatial resolution (13 µm) can be achieved.
Once the set up is ready, the in-situ experiment can be
conducted.
(b)
4.2 Experimental stress-strain curve and macro-crack
identification
The first tomographic scan is performed prior to any
load application, once the specimen is mounted in the
loading cell and positioned at the desired distance from
the source that is kept fixed during the whole test. The
meso-structure obtained from this scan (see Fig. 1) is
the one used in the FE model presented in Section 3. Af-
ter the first scan is completed, a tensile load is applied
by moving the ram downwards with a fixed displace-
ment rate till a load value of 150 N is reached. At this
stage, the displacement is stopped and a second scan is
performed. The specimen is then loaded, with the same
strain rate, up to 250 N when a third scan is carried
out. The initial objective is to have at least: a scan in
the initial configuration, two others at two intermedi-
ate load values before the peak and a fourth one after
failure. Since the exact tensile strength of the specimen
is not known a priori, these loading steps were cho-
sen approximately by using information concerning the
concrete composition, the dimensions of the specimen
and previous mechanical tests done on similar micro-
concrete specimens.
Fig. 6 shows the experimental macroscopic stress-
strain curve, after subtracting the strain corresponding
to the elastic stiffness of the loading cell and assum-
ing a level of zero stress in the specimen after the ma-
terial failure. Note that due to the use of a polycar-
bonate cell, the loading system is not stiff enough to
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography
Fig. 6: Macroscopic response of in-situ tensile test. The
loading points where the scans were conducted are in-
dicated, along with the reconstructed 3D image of the
micro-concrete specimen
capture the post-peak. The loading points where the
tomographic scans were carried out are indicated along
with the reconstructed 3D images of the specimen. No
obvious micro-structural change is visible in the 3D im-
ages between the initial and the two following scans,
whereas a clear macro-crack, normal to the direction
of loading, appears in the one after failure. While the
specimen size is admittedly small compared to its het-
erogeneities, the measured tensile strength is 3.2 MPa
and the macroscopic Young’s modulus is 32 GPa, values
that are very close to the ordinary concrete from which
the micro-concrete composition is derived.
Conducting a post-peak scan gives valuable infor-
mation about the damaged internal micro-structure.
By isolating the voids and keeping the largest connect-
ing component in the post-peak 3D image, the exper-
imental macro-crack can be extracted and is shown in
Fig. 7. A typical macro-crack for fracture mode I is
formed, crossing the cylindrical specimen. It is clear
that the shape and location of both aggregates and
macro-pores form its morphology. The crack is formed
around the aggregates, sometimes propagating through
macro-pores and rarely only through a single aggregate.
Note that, at the scale observed, in the post-peak scan,
only the failure macro-crack can be extracted. How-
ever, micro-cracks of too small thickness to be identi-
fied from a single scan and zones with localised defor-
mation surely exist in the specimen before the failure.
To identify these, it is necessary to proceed to DVC
measurements as explained in details in the following
subsection.
4.3 Following the micro-structural evolution
In addition to the presented macroscopic measures, hav-
ing performed multiple successive scans under progres-
9
(a) (b)
Fig. 7: Post-peak x-ray scan: (a): segmented 3D image
with the macro-crack (b): skeletonised extracted macro-
crack of 1 pixel width
sive loading is ideally suited for following the micro-
structural evolution while the micro-concrete specimen
deforms. Digital Volume Correlation (DVC), is in par-
ticular, a suitable technique for obtaining the full 3D
displacement field between a reference and a deformed
state of the specimen from which the 3D strain field can
then be calculated (9). Moreover, cracks with widths
less than the voxel size (too narrow to be visible on the
x-ray 3D images) can also be detected by DVC, giving
valuable insight into the initiation and propagation of
micro-cracks.
In this work, a software called spam (the authors
are core developers) is used for the DVC analysis of the
examined micro-concrete specimen. A “local” DVC ap-
proach is used, in which the region of interest in the
reference configuration is subdivided into a set of inde-
pendent subvolumes (zones of interest). Correlation be-
tween the two configurations (two pairs of 3D images)
is done by calculating a linear transformation opera-
tor that gradually deforms each reference subvolume
so as to best match the 3D volume of the deformed
configuration. Here, a fine 3D regular grid with a node
spacing of 20 voxels (i.e., 0.26 mm) in each direction is
created to capture the small micro-structure changes.
This grid imposes subvolumes that are far too small to
capture features within large aggregates and so subvol-
umes that fall completely inside aggregates are not suc-
cessfully correlated. However, aggregates are identified
using the procedure in (27) (which are recognised due
to to their low variance i.e., lack of features) and large
aggregates are correlated using the “Discrete DIC” op-
tion in spam (in the style of (8; 1)), meaning that a
single linear transformation operator is measured for
10
each aggregate. The displacement of each point of the
grid (center of each subvolume) that has successfully
correlated (i.e., mainly points that fall in the mortar
matrix and partially between the phases) is indepen-
dently calculated resulting to a 3D displacement field.
By combining this field with the displacements calcu-
lated separately for each large aggregate the final 3D
displacement field is obtained. The strain field in then
calculated using TomoWarp2 (32) following the large
strains hypothesis, with the displacements of the large
aggregates being excluded from the calculation.
Fig. 8: Central vertical slice of the displacement and
strain fields in the axial direction computed with DVC.
A combination of a “local” (for matrix and small aggre-
gates) and a “Discrete” (for large aggregates) DVC ap-
proach is followed for the calculation of displacements.
The strain field derives from the displacement field with
the discplacements of large aggregates being excluded
from the calculation (grey colour in the strain field cor-
responds to the position of the large aggregates). The
dashed line represents the macro-crack that appears af-
ter failure
It should be noted here, that the transformation
operator is a linear operator, unable to capture large
strains linked to the total fracture of the material. In
other words, a correlation between any 3D image and
the final (post-peak) one is not possible. Fig. 8 presents
the central vertical slices of the reconstructed 3D im-
ages of the first 3 scans along with the central vertical
Olga Stamati et al.
slices of the calculated displacement and strain fields
in the axial direction. Note that between the first two
scans, there is a big displacement step (rigid body mo-
tion of the specimen) in the range of 300 µm, while the
range of the relative displacements corresponding to
the micro-structural evolution of the material is only
4 µm. From the computed 3D fields it is shown that the
response of the material is not purely elastic-brittle.
Even for the first correlated pair of 3D images (at a
load 50% of the maximum failure load), an heteroge-
neous displacement field appears, pointing out strain
localizations and crack initiations. An observation that
justifies that performing an in-situ tensile test gave in-
deed valuable information about the evolution of the
fracture processes of the material.
5 Identification of numerical parameters
Since the actual material response is known, the numer-
ical parameters used in the meso-model can be iden-
tified to better fit the numerical response to the ex-
perimental results. Examining the 3D image coming
from the post-peak x-ray scan, it seems that some of
the aggregates did break. This means that crack ini-
tiation and propagation should be allowed inside these
elements but with much higher failure parameters com-
pared to the mortar matrix. A parametric analysis is
performed by assuming the same value of Poisson ratio
νa = νm = 0.2 and keeping a constant ratio between:
Ea : Em , σya : σym and Gfa : Gfm . The calibrated ma-
terial properties are summarised in Table 1. The com-
parison of the experimental and numerical stress-strain
curves is shown in Fig. 9 with a good agreement both in
terms of macroscopic strength and in terms of Young’s
modulus.
Table 1: Calibrated parameters of each phase and their
interfaces
Phase E [GPa] ν σy [MPa] Gf [Jm−2 ]
aggr. 70 0.2 45 0.5
mortar 14 0.2 9 0.1
pores - - - -
aggr./mortar - - 9 0.1
Fig. 10 illustrates the evolution of the displacement
fields in the direction of loading during the compu-
tation. At the beginning of the loading the displace-
ment field is rather homogeneous, while at the end,
the discontinuity in the displacement field in a roughly
horizontal plane (perpendicular to the loading direc-
tion) indicates the presence of a single macro-crack.
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography
3.5
3.0
2.5
stress [M P a]
2.0
1.5
1.0
0.5
0.0
0.00 0.05 0.10 0.15
strain [×10−3] [−]
Fig. 9: Comparison of experimental-numerical macro-
scopic stress-strain curves
Fig 11 shows the elements for which a strong disconti-
nuity has been activated, which could be associated to
a visualisation of the crack patterns predicted by the
numerical model. Its evolution with the loading gives
an idea about the failure process during the numeri-
cal computation. Note that the magnitude of the jump
in the displacement field directly corresponds to the
crack opening. At the beginning of the loading, due to
the difference in elastic moduli between the phases, a
stress concentration around the weakest points (around
aggregates and macro-pores) appears, which leads to
cracks initiation at these points that will propagate in
the perpendicular direction (mode I). One element be-
ing locally under failure, certainly, does not correspond
to any specific macroscopic feature. However, when a
large number of them appears, then it can be said that
a macroscopic response does emerge. This is exactly
what is shown in Fig. 11, where, due to the quasi-brittle
behaviour of the material, these micro-cracks quickly
bridge into a final critical macro-crack that leads finally
to the macroscopic failure of the material.
A qualitative comparison of the crack patterns in
the end of the loading obtained experimentally and nu-
merically (see Fig. 12) shows a very good agreement
between the two responses. The ability of the numer-
ical meso-model to predict the macro-crack location
highlights the influence of the heterogeneities on the
failure process of concrete. The shape and location of
aggregates and macro-pores are two of the key param-
eters that form the geometry of the macro-crack which
propagates by bridging the interfacial micro-cracks. It
should be noted here, that at the observed scale, one
single critical macro-crack has been extracted from the
post-peak 3D image, whereas in the numerical model
there are certainly regions with localised deformation
outside the macro-crack.
11
6 Numerical prediction
numerical
experimental
The explicit representation of the meso-scale hetero-
geneities seems to be the key feature of the presented
meso-model that allows a realistic prediction of the ma-
terial’s macroscopic response. For strengthening this as-
sumption and further challenging the numerical model,
an x-ray scan of another micro-concrete specimen is
performed and a new heterogeneous morphology is in-
0.20 0.25 0.30
troduced to the meso-model. This time, the numerical
response of the second specimen is computed without
any new identification, only based on the material pa-
rameters that were calibrated from the first test. Sim-
ilarly with the first specimen, an in-situ tensile test is
conducted.
Fig. 13 shows the comparison between the experi-
mental and the numerical macroscopic stress-strain curves
for the second micro-concrete specimen. A very good
agreement both in terms of macroscopic strength and
in terms of Young’s modulus is observed. A good qual-
itative comparison of the crack patterns at the end of
the loading is also shown on Fig. 14, again highlight-
ing the impact of the meso-scale heterogeneities on the
formation of the cracks.
The presented validated numerical results show that
starting from an x-ray scan of a micro-concrete speci-
men, the meso-model is able to predict both the macro-
scopic response of the material and roughly the location
of the macro-crack after failure, offering thus interesting
potentials for future applications.
7 Conclusions and perspectives
In this work, a combination of numerical and experi-
mental techniques revealed the impact of the meso-scale
heterogeneities on small specimens of concrete on their
local failure mechanisms under uniaxial tension. The
use of x-ray tomography combined with a mechanical
loading system imposed certain size constrains on the
examined specimens. Small, cylindrical, micro-concrete
specimens (11 mm diameter and 23 mm height) com-
pared to the size of the largest heterogeneities (macro-
pores reaching 2 mm and aggregates reaching 3 mm)
were prepared. These morphologies directly give valu-
able insight into the impact of the mechanical and mor-
phological properties of each phase (aggregates, macro-
pores, mortar matrix), along with their respective in-
terfaces, on the macroscopic response of the material.
An x-ray scan of the initial undamaged specimen
was conducted to reliably extract the morphology of
the meso-structure. This morphology was then given
as an input to a 3D meso-model with enhanced dis-
continuities and a uniaxial tensile test was simulated.
12 Olga Stamati et al.

Fig. 10: Evolution of displacement fields in the axial direction during the numerical simulation

Fig. 11: Evolution of crack patterns during the numerical simulation

3.5
numerical
experimental
3.0

2.5
stress [M P a]

2.0

1.5

1.0

0.5

0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30

strain [×10−3] [−]

Fig. 13: Comparison of experimental-numerical macro-

Fig. 12: Comparison between: (a): experimental macro-
crack (coming from the segmented 3D image of the
post-peak x-ray scan) and (b): numerical crack patterns
(at the end of simulation)
Meanwhile, a uniaxial tensile test was conducted by
progressively loading the specimen until failure while
scopic stress-strain curves for the second micro-concrete
specimen
scanning, allowing a direct validation of the numeri-
cal response. After identification of the numerical pa-
rameters, comparison of experimental and numerical
results revealed the capability of the meso-model to
reproduce the actual material response (in terms of
macroscopic strength, Young’s modulus and fracture
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography
Fig. 14: Comparison between: (a) experimental macro-
crack (coming from the segmented 3D image of the
post-peak x-ray scan) and (b) numerical crack patterns
(at the end of simulation) for the second micro-concrete
specimen
pattern), with the explicit representation of the meso-
scale heterogeneities being its key feature.
While the specimen size was admittedly small com-
pared to its heterogeneities, the measured tensile strength
and macroscopic Young’s modulus were very close to
those of the ordinary concrete from which the micro-
concrete composition was derived. A similar critical macro- 4. De Wolski, S., Bolander, J. et Landis, E.
crack to the experimental one was predicted with the
numerical model and it is thus clear that the shape
and location of both aggregates and macro-pores drove
its geometry. Meanwhile, with the DVC technique, the
relative deformations between consecutive x-ray 3D im-
ages were mapped. Zones of localised deformations, cor-
responding to the macro-crack location, were revealed
at only 50% of the maximum failure load. These zones
are not identifiable by examining each x-ray 3D im-
age separately, thus justifying the use of x-ray scanning
during the tensile test.
The meso-model was further challenged by intro-
ducing a new heterogeneous morphology obtained by
scanning another micro-concrete specimen and comput-
ing its numerical response without any new identifica-
tion. Starting from an x-ray scan of a micro-concrete
specimen, it was shown that the model was able to pre-
dict both the macroscopic response of the specimen and
the location of the macro-crack after failure.
The numerical results presented in this article, di-
rectly compared to corresponding experiments, suggest
interesting future potential applications of the meso-
model. Through a parametric analysis of different mor-
phologies and material properties, information about
the quantitative and qualitative influence of each phase
13
on the global response of the material can be obtained.
Moreover, the fact that the proposed approach may ac-
curately model the characteristic features of the quasi-
brittle behaviour of concrete under tensile loading, al-
lows the future consideration of other types of load-
ing with more complex failure patterns (compressive or
bending loading for instance).
Conflict of Interest: The authors declare that
they have no conflict of interest.
References
1. Andò, E., Hall, S. A., Viggiani, G., Desrues,
J. et Bsuelle, P. (2012). Experimental microme-
chanics: grain-scale observation of sand deformation.
Gotechnique Letters, 2(3):107–112.
2. Basheer, L., Basheer, P. et Long, A. (2005). In-
fluence of coarse aggregate on the permeation, dura-
bility and the microstructure characteristics of ordi-
nary portland cement concrete. Construction and
Building Materials, 19(9):682–690.
3. Benkemoun, N., Hautefeuille, M., Colliat, J.-
B. et Ibrahimbegovic, A. (2010). Failure of het-
erogeneous materials: 3d meso-scale fe models with
embedded discontinuities. International journal for
numerical methods in engineering, 82(13):1671–1688.
(2014). An in-situ x-ray microtomography study of
split cylinder fracture in cement-based materials. Ex-
perimental Mechanics, 54(7):1227–1235.
5. Dupray, F., Malecot, Y., Daudeville, L. et
Buzaud, E. (2009). A mesoscopic model for the be-
haviour of concrete under high confinement. Interna-
tional Journal for Numerical and Analytical Methods
in Geomechanics, 33(11):1407–1423.
6. Geuzaine, C. et Remacle, J.-F. (2009). Gmsh: A
3-d finite element mesh generator with built-in pre-
and post-processing facilities. International journal
for numerical methods in engineering, 79(11):1309–
1331.
7. Gitman, I., Askes, H. et Sluys, L. (2008).
Coupled-volume multi-scale modelling of quasi-brittle
material. European Journal of Mechanics-A/Solids,
27(3):302–327.
8. Hall, S., Bornert, M., Desrues, J., Pannier, Y.,
Lenoir, N., Viggiani, G. et Bsuelle, P. (2010).
Discrete and continuum experimental study of lo-
calised deformation in hostun sand under triaxial
compression using x-ray ct and 3d digital image cor-
relation. Gotechnique, 60(5):315–322.
9. Hild, F. et Roux, S. (2006). Digital image corre-
lation: from displacement measurement to identifica-
14
tion of elastic properties–a review. Strain, 42(2):69–
80.
10. Huang, Y., Yang, Z., Ren, W., Liu, G. et Zhang,
C. (2015). 3d meso-scale fracture modelling and vali-
dation of concrete based on in-situ x-ray computed to-
mography images using damage plasticity model. In-
ternational Journal of Solids and Structures, 67:340–
352.
11. Jirsek, M. (2000). Comparative study on fi-
nite elements with embedded discontinuities. Com-
puter Methods in Applied Mechanics and Engineering,
188(1):307 – 330.
12. Kim, S.-M. et Al-Rub, R. K. A. (2011). Meso-
scale computational modeling of the plastic-damage
response of cementitious composites. Cement and
Concrete Research, 41(3):339–358.
13. Lilliu, G. et van Mier, J. G. (2003). 3d lattice
type fracture model for concrete. Engineering Frac-
ture Mechanics, 70(7):927–941.
14. Mehta, P. K. (1986). Concrete. structure, proper-
ties and materials.
15. Moës, N., Cloirec, M., Cartraud, P. et
Remacle, J.-F. (2003). A computational approach
to handle complex microstructure geometries. Com-
puter Methods in Applied Mechanics and Engineering,
192(28):3163 – 3177. Multiscale Computational Me-
chanics for Materials and Structures.
16. Nitka, M. et Tejchman, J. (2018). A three-
dimensional meso-scale approach to concrete fracture
based on combined dem with x-ray µct images. Ce-
ment and Concrete Research, 107:11–29.
17. Oliver, J. (1996). Modelling strong discontinu-
ities in solid mechanics via strain softening constitu-
tive equations. International Journal for Numerical
Methods in Engineering.
18. Oliver, J. (2000). On the discrete constitutive
models induced by strong discontinuity kinematics
and continuum constitutive equations. International
Journal of Solids and Structures, 37(48):7207 – 7229.
19. Piotrowska, E., Malecot, Y. et Ke, Y. (2014).
Experimental investigation of the effect of coarse ag-
gregate shape and composition on concrete triaxial
behavior. Mechanics of Materials, 79:45–57.
20. Poinard, C., Piotrowska, E., Malecot, Y.,
Daudeville, L. et Landis, E. N. (2012a). Com-
pression triaxial behavior of concrete: the role of the
mesostructure by analysis of X-ray tomographic im-
ages. European Journal of Environmental and Civil
Engineering.
21. Poinard, C., Piotrowska, E., Malecot, Y.,
Daudeville, L. et Landis, E. N. (2012b). Com-
pression triaxial behavior of concrete: the role of the
mesostructure by analysis of x-ray tomographic im-
Olga Stamati et al.
ages. European Journal of Environmental and Civil
Engineering, 16(sup1):s115–s136.
22. Ren, W., Yang, Z., Sharma, R., Zhang, C. et
Withers, P. J. (2015). Two-dimensional x-ray ct im-
age based meso-scale fracture modelling of concrete.
Engineering Fracture Mechanics, 133:24–39.
23. Roubin, E., Colliat, J.-B. et Benkemoun, N.
(2015a). Meso-scale modeling of concrete: A morpho-
logical description based on excursion sets of random
fields. Computational Materials Science, 102(Supple-
ment C):183 – 195.
24. Roubin, E., Vallade, A., Benkemoun, N. et
Colliat, J.-B. (2015b). Multi-scale failure of hetero-
geneous materials: A double kinematics enhancement
for Embedded Finite Element Method. International
Journal of Solids and Structures.
25. Shahbeyk, S., Hosseini, M. et Yaghoobi, M.
(2011). Mesoscale finite element prediction of con-
crete failure. Computational Materials Science, 50(7):
1973–1990.
26. Sonon, B., François, B. et Massart, T. J.
(2015). An advanced approach for the generation of
complex cellular material representative volume ele-
ments using distance fields and level sets. Computa-
tional Mechanics, 56(2):221–242.
27. Stamati, O., Roubin, E., And, E. et Malecot,
Y. (2018). Phase segmentation of concrete x-ray to-
mographic images at meso-scale: Validation with neu-
tron tomography. Cement and Concrete Composites,
88:8 – 16.
28. Suchorzewski, J., Tejchman, J. et Nitka, M.
(2017). Discrete element method simulations of frac-
ture in concrete under uniaxial compression based on
its real internal structure. International Journal of
Damage Mechanics, page 1056789517690915.
29. Sukumar, N., Chopp, D. L., Mos, N. et Belyt-
shko, T. (2001). Modelling holes and inclusions by
level sets in the extended finite element method. Com-
puter Methods in Applied Mechanics and Engineering.
30. Taylor, R. L. (1987). FEAP-ein finite ele-
ment analysis programm. Ing.-Gemeinschaft Klee &
Wrigges.
31. Tejchman, J. et al. (2010). Calculations of frac-
ture process zones on meso-scale in notched concrete
beams subjected to three-point bending. European
Journal of Mechanics-A/Solids, 29(4):746–760.
32. Tudisco, E., And, E., Cailletaud, R. et Hall,
S. A. (2017). Tomowarp2: A local digital volume cor-
relation code. SoftwareX, 6:267 – 270.
33. Wells, G. et Sluys, L. (2001). Three-dimensional
embedded discontinuity model for brittle fracture. In-
ternational Journal of Solids and Structures, 38(5):
897–913.
Tensile failure of micro-concrete: from mechanical tests to FE meso-model with the help of x-ray tomography 15

34. Wilson, E. L. (1974). The static condensation al-

gorithm. International Journal for Numerical Meth-
ods in Engineering, 8(1):198–203.
35. Wriggers, P. et Moftah, S. (2006). Mesoscale
models for concrete: Homogenisation and damage be-
haviour. Finite elements in analysis and design,
42(7):623–636.
36. Yang, Z., Ren, W., Sharma, R., McDonald, S.,
Mostafavi, M., Vertyagina, Y. et Marrow, T.
(2017). In-situ x-ray computed tomography charac-
terisation of 3d fracture evolution and image-based
numerical homogenisation of concrete. Cement and
Concrete Composites, 75:74–83.
37. Yu, Q., Liu, H., Yang, T. et Liu, H. (2018). 3d
numerical study on fracture process of concrete with
different itz properties using x-ray computerized to-
mography. International Journal of Solids and Struc-
tures.
38. Zingg, L., Briffaut, M., Baroth, J. et Male-
cot, Y. (2016). Influence of cement matrix porosity
on the triaxial behaviour of concrete. Cement and
concrete research, 80:52–59.