Cement and Concrete Research 40 (2010) 85–93

Contents lists available at ScienceDirect

Cement and Concrete Research

j o u r n a l h o m e p a g e : h t t p : / / e e s. e l s ev i e r. c o m / C E M C O N / d e f a u l t . a s p

Influence of aggregate size and volume fraction on shrinkage induced micro-cracking

of concrete and mortar
Peter Grassl a,⁎, Hong S. Wong b, Nick R. Buenfeld b
a
Department of Civil Engineering, University of Glasgow, Glasgow, UK
b
Department of Civil Engineering, Imperial College London, London, UK

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, the influence of aggregate size and volume fraction on shrinkage induced micro-cracking and
Received 31 October 2008 permeability of concrete and mortar was investigated. Nonlinear finite element analyses of model concrete
Accepted 16 September 2009 and mortar specimens with regular and random aggregate arrangements were performed. The aggregate
diameter was varied between 2 and 16 mm. Furthermore, a range of volume fractions between 0.1 and 0.5
Keywords: was studied. The nonlinear analyses were based on a 2D lattice approach in which aggregates were
Microcracking (B)
simplified as monosized cylindrical inclusions. The analysis results were interpreted by means of crack
Interfacial transition zone (B)
Transport properties (C)
length, crack width and change of permeability. The results show that increasing aggregate diameter (at
Shrinkage (C) equal volume fraction) and decreasing volume fraction (at equal aggregate diameter) increase crack width
Aggregate (D) and consequently greatly increases permeability.
Lattice modelling © 2009 Elsevier Ltd. All rights reserved.

1. Introduction durability and service-life. Shrinkage induced micro-cracking has been

investigated recently with a lattice model [25]. However, to the
Drying of cement based composites, such as concrete and mortar, authors' knowledge, the influence of the size of the aggregates on
induces cracking if shrinkage of the constituents is either internally or micro-cracking has not been analysed before.
externally restrained. For example, non-uniform drying leads to a In the present work, shrinkage induced micro-cracking of concrete
moisture gradient, which results in non-uniform shrinkage of the was analysed by means of the nonlinear finite element method. In
specimen. Surface regions shrink faster than the inner bulk material, concrete, stress-free cracks form by a complex nonlinear fracture
which results in surface cracking [2,17]. Additionally, shrinkage might process, during which energy is dissipated in zones of finite size. There
be restrained by aggregates within the composite [14]. Aggregate are three main approaches to describe fracture process zones within
restrained shrinkage can lead to micro-cracking, which strongly the finite element framework. Continuum approaches describe the
influences the transport properties of the material [4,18,28]. However, evolution of cracks as zones of inelastic strains by means of higher-
the evolution of micro-cracks and their dependence on the size and order constitutive models such as integral-type nonlocal models
volume fraction of aggregates is not fully understood yet. In [28] it was [3,23]. In hybrid approaches, cracks are modelled as displacement
observed that permeability increases with increasing aggregate size at discontinuities, which are embedded into the continuum description
a constant aggregate fraction. This result is surprising, since an [7,19,20]. Finally, discrete approaches represent the nonlinear
increase of the aggregate size at a constant aggregate fraction is usually fracture process by means of the failure of discrete elements, such
accompanied by a decrease in the volume of interfacial transition as trusses and beams [13,24]. In recent years, one type of discrete
zones (ITZs), which are known to be more porous than the cement lattice approach based on the Voronoi tessellation has been shown to
paste. One hypothesis is that an increase of aggregate diameter at be very suitable for fracture simulations [5,6]. This lattice approach is
constant aggregate fraction results in an increase of micro-crack width, robust and computationally efficient. With specially designed consti-
which is closely related to permeability [27]. The objective of this work tutive laws, the results obtained with this modelling approach are
was to establish whether this size effect really occurs. This will mesh-independent. In the present study, this lattice approach was
undoubtedly enhance the understanding of the link between micro- used in combination with a damage-plasticity constitutive model,
structure and macro-property, in particular the effect of micro- which was designed to result in mesh-independent responses [10].
cracking on mass transport, which is a critical aspect for predicting The present study is based on several simplifications. Shrinkage is
represented by an Eigenstrain, which was uniformly applied to the
cement matrix only. This is representative of autogenous shrinkage, but
⁎ Corresponding author. Tel.: +44 0141 330 5208; fax: +44 0141 330 4557. does not fully represent transient non-uniform shrinkage due to
E-mail address: grassl@civil.gla.ac.uk (P. Grassl). moisture gradients which may lead to more cracking near the surface

0008-8846/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cemconres.2009.09.012
86 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93

than in the center of a concrete element. Furthermore, the only inclu-

sions considered were aggregates, which were embedded in a uniform
cement paste. Micro-cracking due to other inclusions, e.g. unhydrated
cement and calcium hydroxide crystals, was not considered. Aggregates
were assumed to be separated from the cement paste by interfacial
transition zones (ITZs), which were modelled to be weaker and more
brittle than the cement matrix [16]. The mechanical response of
aggregates was assumed to be elastic. Furthermore, the study was
limited to two-dimensional plane stress analyses, in which aggregates
are idealised as cylindrical inclusions of constant diameter.

2. Experimental results

This section summarises the findings from a recent experimen-

tal study by Wong et al. [28] as background to and motivation for
the modelling work presented in this paper. Concretes and mortars
with a range of aggregate contents, w/c ratio, binder type, curing
period and preconditioning temperature were tested for oxygen dif-
fusivity, oxygen permeability and water sorptivity. Thames Valley
gravel (5–12.7 mm, 2% porosity) and sand (<5 mm, 1% porosity)
were used as coarse and fine aggregates respectively. The main ob-
jective of the study was to examine the relative influences of ITZ
and micro-cracking induced by oven drying on different transport
properties, in light of the inconsistencies in previous investigations
on ITZ and mass transport.
Fig. 1 reproduces the diffusivity and permeability results for
samples cured for 90 days at 20 °C and were subsequently precondi-
tioned at 50 °C, 10% r.h. Note that this represents only a small portion
of the total results presented by Wong et al. [28], but captures the
important aspects of their findings. Interested readers are referred to
the original article for experimental details and in-depth discussion. It
was observed that samples with low w/c ratio and containing silica
fume recorded the lowest transport coefficients for all aggregate
contents and curing ages, as expected. The transport coefficients also
decreased with longer curing age, but increased at higher conditioning
temperature. For mortars, the transport coefficients decreased sig-
nificantly with increase in sand content. At very high sand fractions,
the ITZs should be overlapping, but this did not have any detrimental
effect on the transport properties. The trend remains consistent
regardless of the type of transport property, w/c ratio, binder, curing
Fig. 1. Effect of aggregate volume fraction on a) oxygen diffusivity and b) oxygen
age and even after severe oven drying treatment at 105 °C (results are permeability, of mortars and concretes with different w/c ratios (0.5 & 0.3) and binder
shown in [28]). type. The samples were sealed cured in cling film at 20 °C for the first 90 days and then
The most interesting finding was that concrete has roughly the preconditioned at 50 °C, 10% relative humidity for a further 90 days until constant mass
same diffusivity and sorptivity as its analogous mortar, but signifi- prior to testing. The dotted circle highlights the anomalous trend for concrete samples,
which are represented with squares. (M = mortar; C = concrete; SF = containing 8%
cantly higher permeability (1–2 orders of magnitude), despite having
silica fume).
about a third less ITZ (see Fig. 1). This striking feature is also con-
sistent across different w/c ratios, binder types, curing ages and
conditioning regimes, increasing the confidence that it is not caused gates per unit volume to arrest the cracks. Indeed, backscattered
by experimental errors or a series of coincidences. Furthermore, the electron microscopy and image analysis on a few samples found that
same sample was used for all three transport tests, and the perme- this may be the case [28]. However, experimental results are often
ability test was performed in between diffusivity and sorptivity, which affected by many varying parameters that are difficult to isolate and
discounts the possibility that the samples were damaged during per- control. Therefore, the present numerical study was carried out to
meability testing. However, the higher permeability in concretes confirm the trends observed in the experiments and to increase the
cannot be attributed to the ITZs, because it would have a larger in- understanding of the underlying mechanisms.
fluence on the mortars if this was the case.
Wong et al. [28] suggested that the higher permeability is due to 3. Modelling approach
more micro-cracks forming in concrete compared to in the analogous
mortar, when both are subjected to the same drying condition. This is In the present work, shrinkage induced cracking was described by
because aggregates in concrete have larger size distribution, which at means of a lattice approach [5] combined with a damage-plasticity
the same volume fraction, equates to lower surface area to bond with constitutive law [10]. Here, the modelling approach is briefly reviewed.
the paste. Thus, any shrinkage (or expansion) in the paste would Nodes are placed randomly in the specimen constrained by a minimum
produce larger stresses at the interface and more micro-cracking distance dm, i.e. the smaller dm is, the smaller the average element
compared to the analogous mortar. Since diffusivity and sorptivity are length (Fig. 2a). Based on these randomly placed nodes, the spatial
less sensitive to micro-cracking than permeability, these properties arrangement of lattice elements is determined by a Voronoi tessellation.
are not as affected. Micro-cracks in concretes also have lower specific The cross-sections of the lattice elements, which connect the nodes
lengths and tortuosity, because there are fewer obstructing aggre- (Voronoi sides), are the edges of the Voronoi polygons (Fig. 2a). Each
 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93 87

are modelled by placing nodes strategically, so that cross-sections of

lattice elements form the boundary between the phases [29].
The constitutive model of the lattice elements is a combination
of plasticity and damage-mechanics [10], which relates stresses to
strains. The strains ε = (εn, εs)T are determined from the displacement
jump uc = (un, us)T at mid-point C in Eq. (1) as

uc
ε= ð4Þ
h

These strains are related to the nominal stress σ = (σn, σs)T as

σ = ð1−ωÞDe ðε−εp −εs Þ = ð1−ωÞ 

σ ð5Þ

where ω is the damage variable, De is the elastic stiffness, εp = (εpn,εps)T

is the plastic strain, εs = (εs, 0)T is the shrinkage Eigenstrain and σ ̅ =
(σ n̅ , σ s̅ )T is the effective stress. The elastic stiffness is
 
E 0
De = ð6Þ
0 γE

where E and γ are model parameters controlling both the Young's

modulus and Poisson's ratio of the material [11]. For a regular lattice and
plane stress, Poisson's ratio ν is

1−γ
ν= ð7Þ
3+γ

The plasticity part of the model is based on the effective stress σ ̅ and
consists of an elliptic yield surface, an associated flow rule, an evolution
law for the hardening parameter and loading and unloading conditions.
A detailed description of the components of the plasticity model is
presented in [10]. The initial yield surface of the plasticity model is
determined by the tensile strength ft, the shear strength sft and the
compressive strength cft. The evolution of the yield surface during
hardening is controlled by the model parameter μ , which is defined
as the ratio of permanent and reversible inelastic displacements. The
Fig. 2. Discretisation: (a) lattice elements (solid lines) and cross-sections (dashed lines)
damage part is formulated so that linear stress inelastic displacement
obtained from the Delaunay triangulation and dual Voronoi tessellation, respectively.
(b) Degrees of freedom u1, v1, ϕ1, u2, v2 and ϕ2 of the lattice element of length h and laws for pure tension and compression are obtained, which are
cross-section width l in the local coordinate system. The point C at which the interface characterised by the fracture energies Gft and Gfc, respectively.
model is evaluated is in the center of the polygon facet at a distance e from the center of The constitutive response of the interface model is demonstrated
the lattice element.
in Fig. 3 by the stress–strain response for fluctuating normal strains
for μ = 1 and μ = 0. The normal strain is increased to point A (A′).
Then the strain is reduced to point B (B′) and again increased to point
node has three degrees of freedom, two translations and one rotation, C (C′). For μ → 0, a pure damage-mechanics response is obtained,
shown in the local coordinate system in Fig. 2b. The degrees of freedom which is characterised by reversible inelastic strains. For μ = 1, on the
ue = {u1, v1, ϕ1, u2, v2, ϕ2}T of two nodes of a lattice element are related other hand, a pure plasticity model is obtained. The unloading is
to the displacement discontinuities uc = {uc, vc}T at the mid-point C of
the cross-section by

uc = Bue ð1Þ

where


−1 0 e 1 0 −e
B=  ð2Þ
0 −1 −h = 2 0 1 −h = 2

The variable h is the element length and e is the eccentricity,

defined as the distance between the mid-points of the lattice element
and the corresponding cross-section, respectively (Fig. 2b). The ele-
ment stiffness in the local coordinate system is

l T
Ke = B DB ð3Þ
h

where l is the length of the cross-section (Fig. 2b) and D is the Fig. 3. Stress–strain response for fluctuating normal strains for μ = 1 (solid line) and
constitutive matrix relating stresses to strains. Different material phases μ = 0 (dashed line).
88 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93

elastic and the compressive strength is reached sooner than for μ = 0.

The equivalent crack opening w̃c is defined by the equivalent inelastic
displacement, which for the present damage-plasticity model is
defined as

w̃c = ∥wc ∥ ð8Þ

where

wc = hðεp + ωðε−εp ÞÞ ð9Þ

The inelastic displacement vector wc is composed of a permanent

and reversible part, defined as hεp and hω(ε − εp), respectively.
The present lattice model has been implemented within the frame-
work of the nonlinear finite element method, with an incremental–
iterative solution strategy. A modified Newton–Raphson iteration with
the secant stiffness is used.

4. Nonlinear finite element analysis of shrinkage

induced micro-cracking

Shrinkage induced micro-cracking was analysed by means of the

nonlinear finite element approach described above. The elements
representing the cement paste were subjected to an incrementally
applied uniform shrinkage strain up to εs = 0.5 % (Eq. (5)). This value
was chosen for the simulation to represent a relatively severe
shrinkage of the neat cement paste on first-drying to low humidities.
For example, Helmuth and Turk [12] reported first-drying shrinkage
values of 0.35–0.70% for neat cement pastes with w/c ratios of 0.3–0.6,
cured for 6–30 months and dried at 47% r.h. Further drying to 7% r.h.
produced additional irreversible shrinkage of 0.1%. In another study,
Fu et al. [8] measured shrinkage Eigenstrain values of 0.2–1.3%
by oven drying neat cement pastes (w/c ratios 0.25–0.70, cured for
6 months) at increasing temperature up to 110 °C. The influence of
aggregate volume fraction and aggregate diameter was studied. Ag-
gregate volume fractions ρ = 0.5, 0.3 and 0.1 were modelled. Fur-
thermore, four different aggregate diameters (ϕ = 16, 8, 4, and 2 mm)
Fig. 4. (a) Geometry of the test specimen of length L. Aggregate size ϕ and volume ratio
were used. The geometry of the specimen analysed for all volume
ρ are varied. Specimen length L is determined according to Eq. (10). (b) Lattice for
fractions and aggregate sizes is shown in Fig. 4a. The length L of the numerical analysis for ρ = 0.3 and ϕ = 16 mm.
specimen in Fig. 4a is
sffiffiffiffiffiffiffiffiffi 10 s of μm [26]. Therefore, the stiffness of these lattice elements was
πϕ2 approximated as
L= ð10Þ
ρ
 −1
1 1
EI = + ð11Þ
Thus, the smaller the aggregate diameter ϕ and the greater the 2Ea 2Em
volume ratio ρ, the smaller is the specimen length L. The smallest
separation distance between aggregate particles (sa − ϕ), i.e. mini- where Ea and Em are the Young's modulus of aggregate and mortar
mum width of the cement paste decreases with increasing aggregate respectively. Consequently, the Young's modulus of these elements
volume fraction at constant aggregate size, as would be expected is independent of the element size. The strength of these lattice
in the case of real mortars with increasing aggregate fraction. At elements is determined by the strength of the ITZ, which is the
constant aggregate fraction, the separation between aggregate par- weakest link. Here, the strength and fracture energy ratio of cement
ticles increases with increasing aggregate size, similar to the case of paste and ITZ was chosen as 2. This ratio is an approximate value for
a mortar compared to concrete at the same aggregate content. The samples with a relatively weak ITZ. For instance, Hsu and Slate [16]
ratio of the aggregate diameter ϕ and the minimum distance dm for observed that the average tensile bond strength of the paste–
the nodes of the lattice was chosen as 64/3. We will show later that aggregate interface varied from 50–75% of the paste tensile strength,
varying this ratio has little effect on the simulation results. Thus, the depending on the aggregate type, surface roughness and w/c ratio
same detail of discretisation was applied, independently of the size of (0.265–0.36). The strength ratio, however, was observed to be
aggregates. For a diameter of ϕ = 4 mm for instance, the minimum independent of curing age (3–90 days).
distance corresponds to dm = 0.1875 mm. The material parameters
for the constitutive model were chosen according to Table 1 [9]. Ag- Table 1
gregates were modelled elastically. Lattice elements crossing the Model parameters.
boundary between aggregates and cement paste represent the
Element E [GPa] γ ft [MPa] s c Gft [J/mm2] Gfc [J/mm2] μ
average response of the interfacial transition zones and the two
adjacent material phases. In all the analyses, the length of the lattice Cement paste 40 0.33 6.5 2 20 100 100,000 0
Interface 57.1 0.33 3.25 2 20 50 50,000 0
elements is significantly greater than the width of the highly non-
Aggregate 100 0.33 – – – – – –
uniform interfacial transition zones, which is usually in the range of
 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93 89

With μ = 0.001, the constitutive model response is close to a perfect where A = L2 is the area of the specimen, li is the width of the cross-
elasto-damage response (Fig. 3), which is characterised by the absence section (Fig 2) of lattice element i, and nc is the number of cracked
of permanent displacements. For an aggregate diameter of ϕ = 16 mm lattice elements. Furthermore, the average crack width is
and a volume fraction of ρ = 0.3, the crack patterns and the deformed
mesh are shown in Fig. 5a and b. Black lines mark cross-sections of nc n
 = 1 ∑ ∑i c= 1 li w̃ci
lattice elements, for which the maximum crack width is reached at this w c l w̃ = ð13Þ
lc A i = 1 i ci n
∑i c= 1 li
stage of the analysis. This crack stage is defined as active. On the other
hand, grey lines represent lattice elements, which reached their
where w̃ci is the crack width of element i (Eq. (8)). The influence
maximum crack widths at an earlier stage of the analysis. The crack
of the aggregate diameter and volume fraction on the specific crack
opening in these elements is either reducing, or increasing again after an
length and average crack width are shown in Fig. 6a and b,
earlier reduction. The simulated widths of the active and passive cracks
respectively. The specific crack length in Fig. 6a increases with
range between 0 and 70 μm. Some bond cracks between the aggregate
decreasing aggregate size at constant volume fraction. An increase of
and paste can be observed, however most cracks appear to originate at
volume fraction at constant aggregate size results in a decrease of the
the aggregate surface and propagate towards the matrix. With the
specific crack length. Furthermore, the average crack width (Fig. 6b)
deformed mesh in Fig. 5b, it is illustrated that the deformations are
decreases with decreasing aggregate size at constant volume
localised in a few cracks, which connect the aggregates in a regular
fraction. Again, an increase of the volume fraction at constant
square pattern. Qualitatively very similar crack patterns were obtained
aggregate size results in a decrease of the crack width. The crack
for the other aggregate diameters and volume fractions. This pattern of
width is controlled by the spacing of the dominant cracks (Fig. 5b),
cracking seems consistent to that observed by Hsu [15] in ‘2D’ model
which is a function of aggregate diameter and volume fraction. The
samples made of sandstone discs arranged in a square grid and filled
greater the spacing between cracks, the greater is the average crack
with paste that is subsequently subjected to drying shrinkage. De-
width.
pending on the separation between the aggregates, cracks were seen to
Crack width is closely related to transport properties of concrete, in
occur at the interface (i.e. bond cracks), near the shortest distance and
particular in the case of flow under a pressure gradient. Assuming that
diagonally at the largest distance between aggregates.
the paste matrix is dense so that flow occurs predominantly through the
Length and width of micro-cracks for different volume fractions
and aggregate diameters were compared. The specific crack length lc is

1 nc
lc = ∑l ð12Þ
A i=1 i

Fig. 5. (a) Crack patterns obtained for ρ = 0.3 and ϕ = 16 mm at a shrinkage strain in Fig. 6. (a) Specific crack length lc in Eq. (12) versus aggregate diameter ϕ for volume
the matrix of εs = 0.5 %. Black lines indicate active cracks, which increase at this stage of fractions of ρ = 0.1, 0.3 and 0.5 at a shrinkage strain of εs = 0.5 %. (b) Average crack
analysis. Grey lines represent inactive cracks, which do not increase at this stage of width wc̅ in Eq. (13) versus aggregate diameter ϕ for volume fractions of ρ = 0.1, 0.3
analysis. (b) Deformations (magnified by a factor of 50) at the same shrinkage strain. and 0.5 at a shrinkage strain of εs = 0.5 %.
90 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93

cracks, the permeability kc in the out-of-plane direction due to cracking 5. Discussion

can be described by the cube of the crack width as
The edge length L of the specimen (Eq. (10)) depends on the
aggregate diameter ϕ and the volume fraction ρ. Thus, the specimen
ξ nc 3
kc = ∑ li w̃ci ð14Þ area for which crack length, crack opening and hydraulic conductivity
A i=1
were evaluated, differed depending on the micro-structure consid-
ered. Comparison of the results for the different micro-structures is
where ξ is a material constant [1,27]. The influence of aggregate only valid if the average properties are independent of the specimen
diameter and volume fraction on kc is shown in Fig. 7a on a log–log area. Here it is demonstrated that this is the case. It is assumed that
scale for ξ = 1. Note that what is of interest here is not the actual value the square crack pattern, shown in Fig. 5b in the form of cracks
of the estimated permeability, but the change in permeability caused connecting the aggregate particles with a spacing sa, is independent of
by varying either the aggregate volume fraction or particle size. At a the number of aggregate particles considered. Thus, the number of
constant aggregate volume fraction, increasing the aggregate diam- dominant cracks in one direction of the specimen is nc = L/sa. For
eter from 2 to 16 mm caused approximately a 2.5 orders of magnitude instance, for the crack patterns shown in Fig. 5b, nc = 2, whereas for a
increase in permeability. At constant aggregate diameter, increasing specimen of double the size it would be 4, since sa is related to the
the aggregate volume fraction from 0.1 to 0.5 produced about 1 order material structure and is independent of the specimen length L. With
of magnitude decrease in permeability. The aggregate diameter this assumption, the specific crack length lc in Eq. (12), which is
influences the permeability strongly, since the crack width, which defined as the total length of cracks divided by the specimen area, is
increases with increasing aggregate diameter (Fig. 6b), enters Eq. (14)

in its cube.  
2nc ðL−nc ϕÞ L2 s2 1− 2ϕ
sa 2 2ϕ
In Fig. 5a many inactive cracks (grey lines) are visible, whereas the lc = =
a
= 1− ð15Þ
displacements are localised in only a few zones as illustrated by the A L2 sa sa
deformed specimen in Fig. 5b. These inactive cracks have small widths
(<1 μm), but the evaluation of crack length in Eq. (12) and crack where L − ncϕ is the length of one crack. The specific crack length lc
width in Eq. (13) would be strongly influenced by this initial cracking. depends only on the parameters sa and ϕ, but not on the specimen
To avoid misinterpretation, only lattice elements with crack openings length L. An increase of L, at constant ϕ and sa, does not change lc if
greater than 1 μm were considered in the evaluation of Eqs. (12) to the influence of the specimen boundaries, which is studied later, is
(14). negligible. Equivalent observations can be made for the average crack
width wc̅ and kc.
The damage-plasticity constitutive model used for the finite element
analyses was designed to result in crack openings, which are inde-
pendent of the size of the lattice elements [10]. In the present study, a
mesh-size independent description of the crack opening is important,
since the analysis for varying aggregate diameters for the aggregate
arrangement in Fig. 4a was performed with the same diameter mesh-
size ratio of 64/3. Consequently, the smaller the aggregate diameter, the
smaller is the length of the lattice elements used. To investigate a
possible influence of element size on crack openings, two additional
analyses with ϕ to dm ratios of 32/3 and 128/3 were performed for a
volume fraction ρ = 0.3 and a constant aggregate diameter ϕ = 4 mm.
The influence of element size on permeability (Eq. (14)) with ξ = 1 is
shown in Fig. 7b on log–log scale. The evaluation of permeability is
almost independent of the element size. The change in the simulated
permeability due to element size is small and negligible compared to the
change in permeability due to varying aggregate size and volume
fraction (Fig. 7a). As mentioned above, the results of the analyses were
evaluated for lattice elements with crack openings greater than 1 μm.
This threshold was applied to avoid misinterpretation due to distributed
initial cracking. The influence of the crack opening threshold on crack
length (Eq. (12)) and permeability (Eq. (14)) with ξ = 1 is shown in
Fig. 8a and b, respectively. The specific crack length is strongly
influenced by the crack threshold. On the other hand, crack permeability
is almost independent of the threshold, since lattice elements with small
crack openings contribute little to the overall crack permeability.
The meso-structure used in this study is a strong idealisation of
concrete and results in regular square crack patterns (Fig. 5b), which
differ considerably from the random crack patterns observed in
experiments [28]. Therefore, additional analyses were carried out to
investigate the influence of random aggregate arrangements for a
volume fraction of ρ = 0.3 and two aggregate diameters of ϕ = 4 and
8 mm. The specimen length was chosen as L = 50 mm. Three analyses
for each aggregate diameter were carried out to determine the mean
of the random results. The same material properties as for the regular
Fig. 7. (a) Crack permeability kc (Eq. (14)) versus aggregate diameter ϕ for volume
analyses were used. The crack patterns for two random analyses for
fractions of ρ = 0.1, 0.3 and 0.5 at a shrinkage strain of εs = 0.5 %. (b) Minimum spacing aggregate diameters ϕ = 4 and 8 mm are shown in Fig. 9a and b,
of lattice nodes versus permeability kc for ρ = 0.3 and ϕ = 4 mm. respectively. The specific crack length lc, the average crack width w̅c.
 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93 91

Fig. 9. Crack patterns for ρ = 0.3 for (a) ϕ = 4 mm and (b) ϕ = 8 mm at a shrinkage
strain in the matrix of εs = 0.5 %. Black lines indicate active cracks, which increase at this
stage of analysis. Grey lines represent inactive cracks, which do not increase at this
stage of analysis.

Fig. 8. Influence of crack opening threshold for ρ = 0.3 and ϕ = 4 mm. (a) Specific crack
length versus crack opening threshold. (b) Crack permeability kc (Eq. (14)) versus crack
opening threshold. to shrinkage from the induced drying rather than damage from
sample preparation because these cracks were impregnated with
epoxy and hence ‘preserved’, prior to grinding and polishing. The
crack pattern bears some resemblance to the simulations shown
and the change of permeability kc obtained by means of the random in Fig. 9, such as its random orientation, the partial bond cracks
analyses are presented in Figs. 6a and b, and 7a (triangular symbols) appearing around some aggregate particles and the matrix cracks that
together with the results obtained from the analysis with regular propagate through the paste and very often bridging several ag-
aggregate arrangements. Although the resulting crack pattern differs gregate particles. However, the simulations did not produce cracks
considerably, as would be expected, the specific crack length, average that propagate through the aggregates. Also, the simulated cracking
crack width and permeability results for random and regular aggre- appears to be more severe than that observed on the mortars dried at
gate arrangements are in good agreement. Consequently, the regular 50 °C, 10% r.h. A possible reason for this is a mismatch in the model
arrangement seems appropriate to investigate the influence of aggre- parameters used for the simulation.
gate diameter on shrinkage induced cracking. Wong et al. [28] found that increasing aggregate fraction of the
It may be useful to relate the findings of this study to that of the mortar from 10% to 50% resulted in a decrease in measured perme-
experimental results by Wong et al. [28]. A direct comparison is not ability by about an order of magnitude (Fig. 1b). Given the same drying
possible because of the approximations used in this study, although treatment, concretes (with max. aggregate size 12.7 mm) have sig-
certain trends seem to be captured by our simulations. Using image nificantly higher permeability, about 1.5 orders of magnitude, than
analysis, Wong et al. [28] found that the concretes have more micro- mortars (max. aggregate size 5 mm) at the same aggregate content.
cracks, and that the micro-cracks have lower specific lengths, compared Interestingly, this change in permeability due to the aggregate content
to analogous mortars subjected to the same drying conditions. Fig. 10 and size effect is close to the simulated values in Fig. 7. This obviously
shows crack patterns observed on a mortar containing 40% vol. sand (0.5 may be a fortuitous result since several important features were not
w/c ratio, 90-day cured) after drying at 50 °C, 10% r.h. using back- captured by the model, for instance, the contribution of capillary pores,
scattered electron microscopy. Note that the samples used for BSE the effect of tortuosity and connectivity, shape and size distribution.
microscopy were dried in the same way as those for transport testing so Some may have an opposite effect or a minor contribution to transport
that the observed micro-structure is representative. properties, but clearly more work needs to be carried out. Nevertheless,
Examination of samples dried at 105 °C found more severe the results of this study show that the aggregate content and particle
cracking, and significantly higher permeability [28]. The micro-cracks size have a significant influence on the formation of microcracking
have widths of about 0.5–10 μm, although finer cracks were seen at induced by aggregate restrained shrinkage, which in turn affects the
higher magnifications. The observed micro-cracks are most likely due permeability of the composite.
92 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93

Fig. 10. Unsegmented (top row) and segmented (bottom row) BSE images of a 90-day mortar (w/c 0.5, 40% vol. sand) that was conditioned at 50 °C, 10%r.h. showing the micro-
cracks, pores, aggregates and cement paste. The image on the left is a composed image produced by aligning 3 × 3 images captured at 100×, to give a larger field of view
(3100 × 2400 μm). The image on the right was obtained at 200×. Pores with aspect ration <2 were filtered to highlight the micro-cracks.

6. Conclusions References
[1] C.M. Aldea, M. Ghandehari, S.P. Shah, A. Karr, Estimation of water flow through
In the present work the influence of aggregate size and volume cracked concrete under load, ACI Materials Journal 970 (5) (2000) 567–575.
fraction on shrinkage induced micro-cracking was studied numeri- [2] Z.P. Bažant, W.J. Raftshol, Effect of cracking in drying and shrinkage specimens,
cally by means of the nonlinear finite element method. The work Cement and Concrete Research 120 (2) (1982) 209–226.
[3] Z.P. Bažant, M. Jirásek, Nonlocal integral formulations of plasticity and damage:
resulted in the following conclusions: survey of progress, Journal of Engineering Mechanics 128 (2002) 1119.
[4] J. Bisschop, J.G.M. van Mier, Effect of aggregates on drying shrinkage microcracking
• The length of micro-cracks decreases with increasing aggregate in cement-based composites, Materials and Structures 350 (8) (2002) 453–461.
[5] J.E. Bolander, S. Saito, Fracture analysis using spring networks with random
diameter and increasing volume fraction. geometry, Engineering Fracture Mechanics 61 (1998) 569–591.
• The average crack width increases with increasing aggregate [6] J.E. Bolander, N. Sukumar, Irregular lattice model for quasistatic crack propagation,
diameter and decreasing volume fraction. Physical Review B 71 (2005).
[7] G.T. Camacho, M. Ortiz, Computational modelling of impact damage in brittle
• The permeability, which is related to the cube of the crack width,
materials, International Journal of Solids and Structures 33 (1996) 2899–2938.
increases with increasing aggregate diameter and decreasing [8] Y. Fu, P. Gu, P. Xie, J.J. Beaudoin, Development of Eigenstress due to drying shrinkage
volume fraction in hardened Portland cement pastes: thermomechanical analysis, Cement and
Concrete Research 24 (1994) 1085–1091.
[9] P. Grassl, Meso-scale analysis of concrete splitting tests, Internal report
Thus, the aggregate diameter and volume fraction influence the GUCE2008PG02, Department of Civil Engineering, University of Glasgow, Glasgow,
formation of micro-cracks induced by aggregate restrained shrinkage, 2008.
which is in agreement with experimental observations in [28] [10] P. Grassl, R. Rempling, A damage-plasticity interface approach to the meso-scale
modelling of concrete subjected to cyclic compressive loading, Engineering Fracture
(Section 2). However, the representation of aggregates as circular Mechanics 75 (2008) 4804–4818.
inclusions and the use of one aggregate size oversimplifies the micro- [11] D.V. Griffiths, G.G.W. Mustoe, Modelling of elastic continua using a grillage of
structure of concrete, which does not allow a direct comparison with structural elements based on discrete element concepts, International Journal for
Numerical Methods in Engineering 50 (2001) 1759–1775.
experimental results. In future work, the present modelling approach [12] R.J. Helmuth, D.H. Turk, The reversible and irreversible drying shrinkage of
will be extended to 3D and the use of realistic aggregate grading hardened Portland cement tricalcium silicate pastes, Journal of Research and
curves will be included by applying multi-scale analysis approaches. Development Laboratory of Portland Cement Association 9 (1967) 8–21.
[13] H.J. Herrmann, A. Hansen, S. Roux, Fracture of disordered, elastic lattices in two
dimensions, Physical Review B 390 (1) (1989) 637–648.
[14] D.W. Hobbs, Influence of aggregate restraint on the shrinkage of concrete, ACI
Acknowledgements Journal Proceedings 710 (9) (1974).
[15] T.T.C. Hsu, Mathematical analysis of shrinkage stresses in a model of hardened
The simulations were performed with the object-oriented finite concrete, ACI Journal 600 (3) (1963) 371–390.
[16] T.T.C. Hsu, F.O. Slate, Tensile bond strength between aggregate and cement paste
element package OOFEM [21,22] extended by the present authors.
or mortar, ACI Journal Proceedings 600 (4) (1963).
HSW and NRB acknowledge the financial support from the Engineer- [17] C.L. Hwang, J.F. Young, Drying shrinkage of Portland cement pastes I. microcrack-
ing and Physical Sciences Research Council, UK. ing during drying, Cement and Concrete Research 14 (1984) 585–594.
 P. Grassl et al. / Cement and Concrete Research 40 (2010) 85–93 93

[18] A.D. Jensen, S. Chatterji, State of the art report on microcracking and lifetime of [26] K.L. Scrivener, A.K. Crumbie, P. Laugesen, The Interfacial Transition Zone (ITZ)
concrete—part I, Materials and Structures 290 (1) (1996) 3–8. between cement paste and aggregate in concrete, Interface Science 120 (4) (2004)
[19] M. Jirásek, Comparative study on finite elements with embedded cracks, 411–421.
Computer Methods in Applied Mechanics and Engineering 188 (2000) 307–330. [27] P.A. Witherspoon, J.S.Y. Wang, K. Iawai, J.E. Galw, Validity of cubic law for fluid
[20] N. Moës, T. Belytschko, Extended finite element method for cohesive crack flow in a deformable rock fracture, Water Resources Research 160 (6) (1980)
growth, Engineering Fracture Mechanics 69 (2002) 813–833. 1016–1024.
[21] B. Patzák, Object oriented finite element modeling, Acta Polytechnica 39 (1999) [28] H.S. Wong, M. Zobel, N.R. Buenfeld, R.W. Zimmerman, The influence of the
99–113. interfacial transition zone and microcracking on the diffusiviy, permeability and
[22] B. Patzák, Z. Bittnar, Design of object oriented finite element code, Advances in sorptivity of cement-based materials after drying, Magazine of Concrete Research 61
Engineering Software 32 (2001) 759–767. (2009) 571–589.
[23] G. Pijaudier-Cabot, Z.P. Bažant, Nonlocal damage theory, Journal of Engineering [29] M. Yip, J. Mohle, J.E. Bolander, Automated modeling of three-dimensional structural
Mechanics, ASCE 113 (1987) 1512–1533. components using irregular lattices, Computer-Aided Civil and Infrastructure
[24] E. Schlangen, J.G.M. van Mier, Simple lattice model for numerical simulation of Engineering 200 (6) (2005) 393–407.
fracture of concrete materials and structures, Materials and Structures 25 (1992)
534–542.
[25] E. Schlangen, EAB Koenders, K. van Breugel, Influence of internal dilation on the
fracture behaviour of multi-phase materials, Engineering Fracture Mechanics 740
(1–2) (2007) 18–33.