Mesoscale modelling and analysis of damage and fragmentation

of concrete slab under contact detonation

X.Q. Zhou, H. Hao
*
School of Civil & Resource Engineering, The University of Western Australia, 35 Stirling Highway, Crawley WA 6009, Australia
a r t i c l e i n f o
Article history:
Received 12 February 2008
Accepted 12 February 2009
Available online 27 May 2009
Keywords:
Blast loading
Mesoscale model
Numerical simulation
Concrete
Fragment
a b s t r a c t
It is interesting and important for researchers to understand the damage process in order to reliably
predict fragment distribution of concrete material under blast loading. In the present study, a mesoscale
concrete model is developed to simulate the dynamic failure process of a concrete slab under contact
detonation. In the mesoscale model, the concrete material is assumed to consist of two phases, that is,
the high strength coarse aggregates and the low strength mortar matrix, randomly distributed in the
structure components. Each coarse aggregate is assumed to be circular with a random radius in a given
distribution range following the Fullers curve. The mesoscale model together with a dynamic plastic
damage material model is incorporated into the hydrocode AUTODYN. The dynamic damage process of
the concrete slab under contact detonation is numerically simulated. Based on the numerical results, the
fragment size distribution is estimated by an image analysis program. Two different random aggregate
distributions are assumed in the present simulations. Numerical results from the two different cases are
compared, and the results from the mesoscale model are compared with that from the homogeneous
concrete material model. The fragment size distributions obtained from numerical simulations are also
compared with those from the empirical statistic formulae.
2009 Elsevier Ltd. All rights reserved.
1. Introduction
Concrete as a main structural material has been widely used all
over the world. It can be used in not only commercial and resi-
dential buildings, but also in industry and military facilities. Some
concrete structures during their service life might be subjected to
explosive loads. Industrial accidents can be a cause of such events.
Some critical infrastructures, such as embassy buildings, bridges
and government buildings, might be targets of terrorist attack.
Structural damage and personnel injury may be caused by direct
blast loading and the ying secondary fragments from the failed
structural and nonstructural components. It is of interest for
researchers to understand the damage process in order to predict
fragments of concrete material under blast loads. Predicting the
performance of concrete structures to explosive loading through
full-scale tests is often beyond affordability. The majority of blast
models today are mainly based on empirical or semi-empirical
formulae. They tend to overlook the physical behavior of concrete
in the dynamic process in blasting. With the rapid development of
computer technology and the advancement of numerical tech-
niques, it makes the predictions through computer simulation
viable [1,2]. Different numerical methods have already been
reported in the literature to model the damage and fragmentation
of concrete materials under blast loading.
1.1. Concrete material model under blast loading
One of the challenges for reliable numerical simulations is to
develop a proper material model for concrete. To model concrete
damage and fragmentation, its strength criterion is the most
important. Based on static tests, many static strength criteria have
been proposed in the past [36], such as MohrCoulomb Criterion,
DruckerPrager Criterion, WilliamWarnke ve parameter Crite-
rion, Ottosen Criterion, HsiehTingChen Criterion and the unied
twin shear strength model. The dynamic properties of concrete
material are different from its static properties. From dynamic
experimental tests, it has been found that both the tensile strength
and the compressive strength of concrete are highly dependent on
the strain rate, i.e., the strain rate effect, which is usually modelled
by a dynamic increase factor (DIF) to relate the dynamic strength to
the corresponding static strength. The strain rate effect for tensile
and compressive strength is also different [711]. Usually the DIF is
obtained fromexperimental tests. The dynamic strength criterion is
* Corresponding author. Tel.: 61 8 6488 1825; fax: 61 8 6488 1044.
E-mail address: [email protected] (H. Hao).
Contents lists available at ScienceDirect
International Journal of Impact Engineering
j ournal homepage: www. el sevi er. com/ l ocat e/ i j i mpeng
0734-743X/$ see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijimpeng.2009.02.010
International Journal of Impact Engineering 36 (2009) 13151326
also simply obtained by multiplying the static compressive, or
tensile strength, by the respective DIF in practice.
Some material models have been constructed to simulate the
concrete behavior under dynamic loading conditions [1220]. In
1993, Johnson and Holmquist [12] developed a brittle damage
model for concrete [12]. Based on this model, RHT model [14],
Gebbekens model [15], K&C model [16] and modied Drucker
Prgager model [12,17] were developed. Recently, Leppa nen [19]
modied the RHT model by using a different DIF for tension and
a bi-linear crack softening law. All the above-mentioned models
belong to the category of plastic damage model. In addition, some
visco-plastic models were also developed for concrete, for example,
Gatuingt and Pijaudier-Cabot [21] developed a damage visco-
plastic model for concrete. They considered the interaction
between the spherical and deviatoric response. The constitutive
relation for concrete is based on visco-plasticity combined with the
rate-dependant continuum damage. The difference between the
plastic damage model and the visco-plastic damage model is that
the time history effect is considered in the later model. Theoreti-
cally the visco-plastic damage model is more reasonable because it
considers the time dependant plastic ow. However, the visco-
plastic behavior of concrete is very complicated and it is not well
understood yet.
In those models, the concrete material is always assumed to be
isotropic, continuous and homogeneous.
1.2. Mesoscale modelling of concrete
As is well known, concrete is a composite material, produced by
adding the appropriate portions of coarse and ne aggregates,
cement, water and some additives if necessary. Obviously the
concrete material is heterogeneous, and its heterogeneity makes
the behavior of concrete under blast loads rather complicated.
Especially, the heterogeneity of concrete affects the crack pattern
and the fragment size distribution when it is under blast loads.
To analyze the static heterogeneous behavior of concrete, some
mesoscale models for concrete have been developed [2226]. In
most of these mesoscale models, the concrete is assumed con-
sisting of three phases, that is, the coarse aggregates, the mortar
matrix with ne aggregate dissolved in it, and the interfacial
transition zones (ITZ) between the aggregate and the mortar
matrix. Based on the static experimental results, the behavior of
ITZ does affect the mechanical properties of concrete. However, it
is very difcult to obtain the mechanical parameters of ITZ.
Therefore including ITZ in the model introduces some uncer-
tainties. Moreover, considering ITZ in numerical model substan-
tially increases the computational time and computer memory
requirement. For these reasons, in some models the ITZs are not
included in the numerical simulation [22,25], instead, the ideal
bond between the aggregates and the mortar matrix are assumed.
To perform the mesoscopic study of concrete material, both
discrete element methods, such as lattice model [26] and truss
model [27], and continuum nite element methods [2224] have
been used. So far, this kind of mesoscale models has mainly been
applied in static numerical simulations.
The present paper aims to construct a mesoscale heterogeneous
model for concrete material under blast loading. In the mesoscale
model, the concrete material is assumed consisting of two phases,
that is, the high strength coarse aggregate and the low strength
cement paste, randomly distributed in the structure components.
Perfect bond between aggregates and cement paste is assumed. As
a numerical example, the dynamic damage process of a concrete
slab under blast loads studied by other researchers using a different
approach [28] is analysed.
1.3. Concrete dynamic fragmentation
When an explosion occurs, secondary fragments and airborne
debris resulting from the damaged structural components may
cause serious injury and damage. Therefore, it is of interest to know
the fragmentation process of structural components and to predict
the size and velocity distribution of the fragments.
The processes of dynamic fragmentation within a concrete
member are very complicated since discontinuities such as cleavage
cracks and defects with different shapes and orientations are
commonly encountered in concrete material and they have signi-
cant inuence on the failure of concrete. The actual process of
dynamic fragmentation is still not well understood, but some theo-
retical and experimental efforts have provided useful insight
regarding the distribution of fragments. Based on energy and
momentum balance principles, some models have been developed
to predict average fragment size as a function of strain rate and
material toughness [2931]. To determine the distribution of frag-
ment in mass or size, some statistical approaches have been devel-
oped [32,33]. In those approaches, the intrinsic failure process
leading to fragmentation is not modelled. To understand the
mechanisms of fragmentation, some theoretical models have been
suggested to correlate the dynamic fracture and fragmentation
[34,35]. Recently, numerical modelling has been carried out to
simulate the dynamic deformation in the fragmentation process
[3538]. There are mainly three different methods for fragmentation
simulation: 1) Interface elements were incorporated between stan-
dard nite elements to serve as dynamic fracture paths [35]. The
primary drawback of this method is that it cannot give reliable
predictions of the fragment size because the size and shape are
determined by the pre-dened interface; 2) Damage material model
has been put into smooth particle hydrodynamics (SPH method) to
simulate the fragmentation process [28,38]. The fragment distribu-
tion was obtained by checking the radius of the fully damaged
particles. 3) Standard nite element method together with damage
mechanics has been employed to model the dynamic deformation
and to predict the fragment size [36,37]. The fragment size is pre-
dicted by either the energy balance principal or relating the full
damage to the fragmentation. All the previous models assumed
homogenous material properties. In the present study, the later
method is adopted, however, the heterogeneous concrete material
properties are considered. The mesoscale concrete model and
AUTODYN [2] are rstly employed to model the dynamic deforma-
tion of a concrete slab under a contact detonation. The fragment size
distributions and ejection velocities are then predicted by relating
the full damage to the fragmentation. Next, an image analysis
program in MATLAB is used to estimate the fragment size distribu-
tion by analyzing the numerical results based on the damage
mechanics theory and nite element model.
2. Generation of coarse aggregate particles
The mesoscale concrete model requires the generation of
a random aggregate structure in which the shape, size and distribu-
tion of the coarse aggregates closely resemble the real concrete inthe
statistical sense [22]. The coarse aggregates generation method in
the present study is the popular take-and-place method. Firstly, the
size distribution of the coarse aggregate particles is determined by
following a certain given grading curve; and then the aggregate
particles are placed into the mortar matrix one by one at randomly
determinedlocations insuchaway that nooverlapping withparticles
already placed. The similar method was also employed in [22,23,27].
Coarse aggregates are the particles whose diameters are greater
than 4.75 mm. For most concrete, the coarse aggregates represent
4050% of the concrete volume [22]. In the present study, it is
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1316
assumed that the total percentage of the coarse aggregates is 40% of
the concrete. The most common coarse aggregate types are gravels
and crushed stones. The shape of aggregate depends on the
aggregate type. Generally, gravel aggregates have a rounded shape
while crushed stone aggregates have an angular shape [23]. Several
methods to characterize the shape of the coarse aggregates have
been published [23,25]. For simplicity, in this study the shape of the
coarse aggregates is assumed to be circular in two-dimensional
(2D) simulations. The slab is modelled as axisymmetric. It should be
mentioned that the dynamic failure processes in concrete material
are affected by three-dimensional (3D) distribution of the coarse
aggregates. However, due to the computer memory limitation, in
this study only the axisymmetric model is used to do the mesoscale
simulation of the concrete slab.
One of the most popular aggregate distribution models is the
Fullers curve (Fig. 1) [22,25]. It is believed that the Fullers curve
represents a grading of aggregate particles resulting in an optimum
density and strength of the concrete mixture. Fullers curve can be
described by a simple equation [22]
Pd d=d
max

n
(1)
where P(d) is the cumulative percentage passing a sieve with
aperture diameter d, d
max
is the maximum size of aggregate
particle and n is the exponent of the equation (n 0.450.70). In
practical concrete construction, the typical maximum size of
aggregate is about 32 mm to obtain high quality concrete mix.
Therefore, in the present study, d
max
is assumed to be 32 mm, and
n is taken as 0.5.
From Fig. 1, it can be found that the percentage for the coarse
aggregates, i.e., aggregates with size larger than 4.75 mm, is about
61.5% of the total aggregates. The total volume of the coarse
aggregates is assumed to be 40% of the whole concrete volume.
The grading curve is divided into several segments. In the present
study, three aggregate diameter ranges are selected, i.e., 4.75
10 mm, 1020 mm and 2032 mm. According to Fullers curve, the
percentages for the three aggregate ranges are 17.37%, 23.17%, and
20.96%, respectively, which correspond to 11.3%, 15.07% and 13.63%
of the total concrete volume. The aggregate distribution process
starts with the grading segment containing the largest size
particle, that is, 2032 mm. Firstly, random position for the centre
of the circular aggregate is determined; next the diameter within
the grading segment (2032 mm) is randomly decided; then
overlapping between any two aggregates and every aggregate with
the slab boundary are checked. If overlapping between two
aggregates or part of an aggregate is outside of the slab boundary,
the next simulated aggregate size and position is used. Once the
total area of the coarse aggregates in the size segment reaches the
designed percentage, the iteration is stopped and the next grading
segment is placed. An example of a generated 2D aggregate
particle distribution is shown in Fig. 2. Similar method can also be
used to construct a 3D mesoscale model. A 3D example is shown in
Fig. 3.
Fig. 3. An example of a generated 3D aggregate distribution for a volume of
200 200 200 mm
3
.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.125 0.25 0.5 1 2 4 8 16 32
Fuller's curve
Size (mm)
T
o
t
a
l

p
e
r
c
e
n
t
a
g
e

Fig. 1. Fullers grading curve.
Fig. 2. An example of a generated 2D aggregate distribution for an area of
300 300 mm
2
.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1317
3. Material model for concrete
AUTODYN[2] is a hydrocode. In a hydrocode, the material model
contains two parts: the equation of state (EOS) which is used to
determine the hydro pressure in terms of the local density and the
local energy, and the strength criterion which controls the yield
strength according to the rst stress invariant I
1
and the second
invariant of the stress deviator tensor J
2
. Different material models
are adopted here for the aggregate and the mortar matrix because
they have different mechanical behaviors.
3.1. Material model for mortar matrix
Typically the mortar matrix has a porosity of 10% and it has
a complex non-linear compression behavior, thus it can be consid-
ered as a porous material. The equation of state used to model the
mortar matrix is a piecewise-linear porous model, in which the
plastic compaction path is dened as a piece-wise linear path from
which unloading and reloading can occur along an elastic line as
shown generically in Fig. 4. Similar material model was constructed
in our previous work to model the concrete behavior [12].
In this model, the material is assumed to be compacted from its
state of initial density r
0
along an elastic path dened by the
differential equation
dp
dr
c
2
init
(2)
until the pressure reaches the plastic yield stress, which is dened
by the value of the pressure in the rst (r; p) pairs. C
init
is the initial
sound speed. Subsequent loading takes place along the plastic
compaction path until the material is fully compacted, at which
point further compression takes place according to the linear
relationship
p c
2
s
r r
s
(3)
where c
s
is the solid sound speed, r
s
is the solid density.
Along the plastic compaction path, unloading and reloading
modulus need to be determined. In the present model, the corre-
sponding bulk sound speed is interpolated between c
init
and c
s
,
c c
init
c
s
c
init

_
a a
0
1 a
0
_
(4)
where a r
ref
=r is the current porosity, and a
0
r
ref
=r
0
is the
initial porosity. Eq. (4) is dened for the unloading and reloading
path, while the plastic compaction path is a piece-wise line deter-
mined by the points 2, 3, 4, 5 in Fig. 4. Due to a lack of experimental
results, the density-pressure pairs in this study are constructed
based on available experimental and reasonably assumed data. The
densitypressure pairs for these points are: 2.34 10
3
kg/m
3
,
25 MPa; 2.35 10
3
kg/m
3
, 28 MPa; 2.4 10
3
kg/m
3
, 70 MPa; and
2.5 10
3
kg/m
3
, 250 MPa.
Once the hydrodynamic pressure reaches the limit p
min
(shown
in Fig. 4), tensile failure occurs. A piece-wise DruckerPrager model
[12] is used here to model the strength behavior of the mortar
matrix (Fig. 5). The equivalent yield strength can be determined by
F
p

J
2
_
b
i
p a
id
0i 1wN (5)
where J
2
(1/2)s
ij
s
ij
, is the second invariant of the stress deviator
tensor s
ij
, p is the hydrostatic pressure and p (1/3)I
1
, in which I
1
is the rst invariant of the stress tensor s
ij
, a
id
a
i
(1D), D is
a damage scalar. In the present study, four sets (N4) of experi-
mental data are used to determine the parameters, that is: (1) cut
off hydro-tensile strength f
ttt
f
ttt
s
1
s
2
s
3
; (2) uniaxial
tensile strength f
t
f
t
s
1
; s
2
s
3
0; (3) uniaxial compres-
sive strength f
c
s
1
s
2
0; s
3
f
c
; (4) conned compressive
strength (I
1
10

3
p
f
c
;

2J
2
_
6f
c
), estimated from [39], where I
1
is the rst stress invariant; J
2
is the second deviatoric stress
invariant.). Once the yield surface is reached, the material has
Table 2
Material parameters for aggregate.
Density r
0A
(kg/m
3
) 2.750 10
3
Bulk modulus K
A
(GPa) 35.7
Shear modulus G
A
(GPa) 17.4
Damage parameters a
t
, a
c
0.5
Tensile damage threshold 3
st0A
4. 10
4
Compressive damage threshold 3
sc0A
4. 10
3
Tensile strength f
tA
(MPa) 10
Compressive strength f
cA
(MPa) 150
Cut-off tensile strength f
tttA
(MPa) 5
Table 1
Material parameters for mortar matrix.
Initial densityr
0M
(kg/m
3
) 2.33 10
3
Solid density r
sM
(kg/m
3
) 2.450 10
3
Initial sound speed C
0M
(m/s) 2.20 10
3
Solid sound speed C
s
(m/s) 2.80 10
3
Initial compaction pressure p
eM
(MPa) 24.0
Solid compaction pressure p
sM
(MPa) 250
Shear modulus (GPa) 8.3
Damage parameters a
t
, a
c
0.5
Tensile damage threshold 3
st0M
3.0 10
4
Compressive damage threshold 3
sc0M
3.0 10
3
Tensile strength f
tM
(MPa) 4
Compressive strength f
cM
(MPa) 48
Cut-off tensile strength f
tttM
(MPa) 2
p
J
2
0
Undamaged material D=0
Damaged material 0<D<D
max
Residual strength DD
max
Fig. 5. Damage-dependent piece-wise DruckerPrager strength criterion.
Solid sound speed c
s
1
2
3
4
5
Pressure P
Pmin

0

s Density
Initial sound speed C
init
Plastic compaction path
Elastic unloading/loading paths
Fig. 4. Pressure and density curve for piecewise linear porous equation of state.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1318
permanent plastic strain. The details on the plastic ow treatment
in AUTODYNcan be found in [2]. The damage scalar Dis determined
by Mazars damage model [40].
In the damage model, the two scalars, namely, D
t
and D
c
, which
correspond, respectively, to the damage in tension and compres-
sion of concrete, are dened as
D
t
_
~3
t
_
1 e
a
t
_
~3
t 3
t0
__
3
t0
D
c
_
~3
c
_
1 e
a
c~3
c3
c0=3
c0
(6)
where a
t
and a
c
are the damage parameters that depend on the
material properties, the range for them are from 0 to 1. For more
details of the determination of the parameters, please refer to ref
[41]. In this study they are taken as 0.5, while 3
to
and 3
co
are the
threshold strains in the uniaxial tensile and compressive states. ~3
t
and~3
c
are the equivalent tensile and compressive strains, dened as
~3
t

i 1;3
_
3

i
_
2

~3
c

i 1;3
_
3

i
_
2

(7)
where 3

i
is the positive principal strain. The means it vanishes if
it is negative. 3

i
is the negative principal strain, and the means it
vanishes if it is positive.
The cumulative damage scalar can be determined by combining
the tensile and compressive damage
D A
t
D
t
A
c
D
c
;
_
D
t
> 0;
_
D
c
> 0 and A
t
A
c
1 (8)
the weights A
t
and A
c
are dened by the following expressions [40],
A
t

i 1;3
H
i
_
3

i
_
3

i
3

i
_
~3
2
; A
c

i 1;3
H
i
_
3

i
_
3

i
3

i
_
~3
2
(9)
where ~3

i1;3
3

i
3

i

2
_
is the effective strain. H
i
[x] 0 when
x <0 and H
i
[x] x when x 0. It can be veried that in uniaxial
tension, A
t
1, A
c
0, DD
t
, and vice versa in compression.
Material parameters for mortar matrix are listed in Table 1.
The dynamic yield strength surface is amplied from static
surface by considering the strain rate effect. Typically the
compressive (tensile) strength is multiplied by a compressive
(tensile) dynamic increase factor (DIF). In the model, the DIFs are
from the CEB recommendation for concrete [42], which are
obtained from many experimental test data on concrete and
cement mortar.
The compressive DIF recommended by CEB is,
CDIF
f
cd
f
cs

_
_ 3
d
_ 3
cs
_
1:026a
for _ 3
d
30s
1
(10a)
CDIF
f
cd
f
cs
g
_
_ 3
d
_1
3
for _ 3
d
> 30s
1
(10b)
where f
cd
is the dynamic compressive strength at the strain rate _ 3
d
(in the range of 30 10
6
1000 s
1
), _ 3
cs
30 10
6
s
1
,
log g 6:156a 0:49; a 5 3f
cu
=4
1
, f
cs
is the static
compressive strength, and f
cu
is the static cube compressive
strength (in MPa).
Axis of symmetry
High explosive
Outflow
boundary
Outflow boundary
Concrete
Air
Euler-
Lagrange
interface
Fig. 9. Axisymmetric numerical model.
High explosive
1200
103
3
2
0

7
5
Fig. 8. Test setup (mm).
Compressive DIF
0
1
2
3
4
5
0.0001 0.01 1 100 10000
Rouabhi
LI
Cai
LI XB
Lok
Present model
Fig. 7. Compressive DIF of rock.
Tensile DIF
0
1
2
3
4
5
0.000001 0.0001 0.01 1 100 10000
Rouabhi
Cho
Wang
Cai
Lok
Present model
Fig. 6. Tensile DIF of rock.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1319
The CEB recommendation for the tensile DIF is [42],
TDIF
f
td
f
ts

_
_ 3
d
_ 3
ts
_
1:016d
for _ 3
d
30s
1
(11a)
TDIF
f
td
f
ts
b
_
_ 3
d
_ 3
ts
_
1=3
for _ 3
d
> 30s
1
(11b)
where f
td
is the dynamic tensile strength at the strain rate _ 3
d
(in the
range of 3 10
6
300 s
1
), f
ts
is the static tensile strength at the
strain rate _ 3
ts
_ 3
ts
3 10
6
s
1
, and log b 7:11d 2:33, in
which d 1=10 6f
0
c
=f
0
co
; f
0
co
10MPa, f
0
c
is the static uniaxial
compressive strength (in MPa).
3.2. Material model for coarse aggregate
The aggregate is assumed to suffer brittle failure with
a minimum deformation. Therefore the simplest linear equation of
state is adopted to calculate the pressure
p Km (12)
where p is the pressure, m (r/r
0
)1, and K is the material bulk
modulus.
Because rock behaves similarly as concrete under dynamic
loading, the same strength criterion for the mortar matrix is
adopted here to model the aggregates, only the material constants
are different. The parameters for the aggregates are given inTable 2.
Fig. 11. Density distribution (a) case I (b) case II.
Fig. 10. Aggregate distribution (a) case I (b) case II.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1320
The DIFs for the coarse aggregates are based on some test results
on rock materials. Some research works have been done to study the
strain rate effect on both the tensile and the compressive strength of
the rock material. Cho et al. [43] investigated the dynamic tensile
strength of Indian granite and Tage tuff, by using experimental
approach based on Hopkinsons effect combined with the spalling
phenomena. Wang et al. [44] used split Hopkinson pressure bar
(SHPB) to impact a attened Brazilian disc of marble for testing
dynamic tensile strength of rock material. Cai et al. [45] conducted
boththe tensile andthe compressionSHPBtests tostudythe dynamic
behavior of Meuse/HauteMarne argillite. Lok et al. [46] used a sha-
ped striker bar in a large diameter SHPB tests to obtain the dynamic
tensile andcompressionstrengthof granite. Li et al. [47] alsoreported
some compression test results by using the same equipment. Li et al.
[48] obtained the dynamic compressive strength for different rock
materials by using a rock dynamic testing systemwhich is driven by
hydraulic and air. Based on the above-mentioned test results, the
tensile and compressive DIFs are obtained as follows,
TDIF 0:0225 log _ 3 1:12 _ 3 0:1s
1
(13a)
TDIF 0:7325
_
log _ 3
_
2
1:235
_
log _ 3
_
1:6 0:1s
1
_ 3 50s
1
(13b)
CDIF 0:0225log _ 3 1:12 _ 3 10s
1
(14a)
CDIF 0:2713
_
log _ 3
_
2
0:3563
_
log _ 3
_
1:2275 10s
1
_ 3 2000s
1
(14b)
Comparisons of the above empirical formulae with the available
test results are shown in Figs. 6 and 7. Rouabhi et al. [49] performed
numerical simulations to construct the material constitutive model
for rock, the numerically derived tensile and compressive dynamic
increase factors are also shown in Figs. 6 and 7 for comparison. It is
found that the DIFs fromthe numerical model are lower than those
obtained from test results.
3.3. Material model for high explosive and air
High explosives are typically modelled by using the Jones
WilkinsLee (JWL) equation of state, which models the pressure
generated by chemical energy in an explosion. It can be written in
the form,
p C
1
_
1
u
r
1
v
_
e
r1v
C
2
_
1
u
r
2
v
_
e
r2v

ue
v
(15)
where p hydrostatic pressure; v specic volume; e specic
internal energy; and C
1
, r
1
, C
2
, r
2
and u are material constants. The
values of the constants for many common explosives have been
determined from dynamic experiments and are available in
AUTODYN [2]. In the present simulation, C
1
, r
1
, C
2
, r
2
, and u are
assumed as, 3.7377 10
5
MPa, 4.15, 3.7471 10
3
MPa, 0.9, 0.35,
respectively.
Air is modelled by the ideal gas equation of state, which is one of
the simplest forms of equation of state. The pressure is related to
the energy by
p g 1re (16)
where g is a constant, r is the air density and e is the specic
internal energy. In the simulation, the standard constants of air
from the AUTODYN material library are utilized, that is, the air
density r 1.225 kg/m
3
and g 1.4. The air initial internal energy is
assumed to be 2.068 10
5
kJ/kg.
4. Numerical simulation
In this study, a concrete slab tested under blast loading by other
researchers [28] is modelled. In numerical simulation, the concrete
slab is modelled by a Lagrange subgrid, in which the coordinates
move with the material; while the air and high explosive are
Fig. 12. Damage distribution (a) case I (b) case II.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1321
modelled by Euler subgrid, in which the grid is xed and material
ows through it. At the Euler-Lagrange interface, interaction is
considered. The Lagrange subgrid imposes a geometric constraint
to the Euler subgrid while the Euler subgrid provides a pressure
boundary to the Lagrange subgrid.
4.1. Test setup
The dimension of the tested slab was 1.2 1.2 0.32 m
3
and the
concrete had a static compressive strength of 48 MPa. This slab was
loaded by an explosive cone of TNT and Composition B, and the
equivalent charge weight was about 350 g. The inner cone con-
sisted of TNT and the outer thin cone of composition B. The
dimensions are shown in Fig. 8. This slab was tested by other
researchers [28]. It is employed here to validate the proposed
numerical model.
4.2. Numerical model
The slab is approximately modelled as axisymmetric in this
study. It should be mentioned that the dynamic fragmentation
process may be affected by the 3D aggregate distribution. The 3D
Fig. 13. Cracked mesh and aggregate distribution.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1322
effect analysis will be carried out in our future research work with
improved computational power. The 2D axisymmetric numerical
model is shown in Fig. 9. Two random aggregate distributions are
shown in Fig. 10. The blue circles are the coarse aggregates and the
white area is the mortar matrix. The density distributions for the
two cases are shown in Fig. 11. In the numerical simulation, quad-
rilateral elements are used. Therefore the circular geometry of the
aggregate is only approximately modelled by quadrilateral
elements. The element size in the 2D simulation is 2 2 mm. The
mesh size was determined by convergence test in the homoge-
neous model and proved yielding reliable predictions of the blast
testing results [12].
4.3. Numerical results and fragment analysis
Typical numerical results on the damage process of the two
different aggregate distribution cases are shown in Fig. 12. Fromthe
Fig., the damage of the slab caused by stress wave propagation can
be clearly seen. At 0.05 and 0.1 ms, the compressive stress wave has
not reached the bottomof the concrete slab, the concrete damage is
associated with concrete crushing caused by the compressive
stress; at 0.15 ms, the stress wave reaches the free boundary
(bottomof the slab), a strong tensile stress wave is generated owing
to the reection of the compressive stress wave, and the tensile
stress is high enough to cause many cracks near the bottom surface
of the concrete slab; at 0.2 ms, the cracks extend further. From the
gure, it can be found that both the aggregate and the mortar
matrix are fully damaged in the high compression zone, which is
the upper crater directly caused by the contact detonation.
However, in the bottom spalling zone, the damage is caused owing
to the tensile failure. Basically the damage (tensile crack) occurs in
the mortar matrix around the aggregates. Therefore, the aggregate
distribution in this part highly affects the tensile damage.
In order to further study the fragment distribution, erosion
technique is adopted. In this simulation the highly distorted
elements and the elements with high tensile strains are eroded. The
cracks are estimated after the elements are eroded. It should be
mentioned that the fragment size less than the element size of
2 mm cannot be obtained in the present numerical simulation due
to the limitation of the erosion technique. The cracked meshes and
the distributions of the aggregates for the two cases are shown in
Fig. 13. Comparison of the two cases shows that the aggregate
distribution does affect the crack position and the fragment
distribution. The cracks basically occur at the boundary of the
coarse aggregate, therefore the size and the position of the coarse
aggregates in the spalling area inuence the crack generation and
propagation. For comparison, the result from the homogeneous
model is also obtained and shown in Fig. 14. In the homogeneous
model, the same material properties as those of the mortar in
the mesoscale model are adopted. It should be mentioned that the
actual strengths of the present homogeneous model and the
heterogeneous model are not exactly the same, however, our
previous research shows that the compressive and the tensile
strength of the mesoscale model are only slightly higher than the
strength of the mortar matrix. The experimental results are shown
in Fig. 15. Comparison of Figs. 1315 shows that all the three models
(two mesoscale model and one homogeneous model) successfully
predicted the perforation of the slab. However, the crater sizes
obtained from different models differ a lot. Comparison of the
craters of test results and the numerical results are listed in Table 3.
It can be found that the numerical results fromthe case I agree well
with the test results. The mesoscale model II overestimates both the
top and the bottom craters, whereas the homogeneous model
overestimates the top crater size but underestimates the bottom
crater size. The results indicate that the concrete slab damage can
be reliably predicted if a proper mesoscale model is used. However,
the results very much depend on the aggregate distributions. Since
the aggregate distribution is random in real concrete structures,
probabilistic analysis is needed to determine the mean and varia-
tion of the concrete structural damage levels. This will be our future
research topic.
Fig. 14. Cracked mesh (homogeneous model).
Fig. 15. Test results [28] (a) top view (b) bottom view.
Table 3
Comparison of the numerical results and the experimental results.
Upper crater
diameter (mm)
Bottom crater
diameter (mm)
Experimental results 51 62
Numerical results (case I) 49.5 61.5
Numerical results (case II) 71.0 71.0
Numerical results (homogeneous) 70.5 52.5
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1323
From the numerical simulations, the ejecting velocity of the
fragments can also be obtained. The highest ejecting velocities of
the fragments from the bottom surface of the concrete are 20.6,
24.0 and 19.8 m/s, for case I, case II and the homogeneous model,
respectively. It should be noted that these ejecting velocities are
associated with the fragments with size larger than 2 mm. For the
fragments smaller than 2 mm, the ejecting velocity might be
higher, but as discussed above, the present numerical model is not
capable of generating fragments smaller than 2 mm because of the
element size used in the model. The results from the numerical
simulation also depend on the erosion criterion used. The erosion
criterion used here is a combination of the effective strain and the
damage value. If the tensile damage value is higher than 0.99 and
the effective strain is higher than 0.2, or if the effective strain is
higher than 2.0, the element is assumed to be fully damaged and
eroded from the model. The erosion criterion adopted here is
partially based on ref [2], partially based on numerical trial. Nor-
mally the erosion is mainly based on effective strain, only the highly
distorted elements are deleted to avoid numerical difculty, but in
the present study, tensile damage is also considered to model the
possible tensile cracks.
The results from the numerical simulation in Figs. 13 and 14 are
then treated by an image analysis program (a toolbox in MATLAB)
to predict the fragment size distribution. Fig. 16 shows the frag-
ments from the bottom spalling (estimated from Figs. 13 and 14). It
should be mentioned that the fragments in the upper crater is not
analysed in the present study because a large number of the
elements are eroded due to large element distortion in this region.
In the image analysis program, the gures are loaded rst by using
the function imread, then the imcomplement function in MAT-
LAB is used to take the complement of the image; next, the
bwmorph function is used to skeletonize the results, nally each
fragment boundary is found and the fragment area is obtained.
After the treatment by the image analysis program, the fragment
distributions can be seen more clearly in Fig. 17. This gure is then
used to extract the area distribution. For each fragment, the size is
estimated as the square root of the fragment area. The size distri-
butions for the different cases are shown in Fig. 18. In the gure, the
vertical axis corresponds to the mass percentage. In the mass
calculation, it is assumed that the fragment size in the third
direction is the same as that in the 2D plane, i.e., the square root of
the fragment area. From Fig. 18, it can be found that the homoge-
neous model predicts more small size fragments than the meso-
scale model. It can also be found that the fragment size
distributions are different for the two aggregate distribution cases,
indicating again the inuence of the aggregate distribution on the
concrete slab damage to blast loads.
According to a statistical approach [33], the cumulative mass of
fragments with mass less than or equal to m is,
M
c
m M
_
1 e
m=m
a
_
(17)
accordingly, the cumulative size of fragments with sizes less than or
equal to s is,
M
c
s M
_
1 e
s=sa
3 _
(18)
where M is the total fragment mass, m
a
is the average fragment
mass, and s
a
is the average fragment size.
For comparison, the results based on statistical empirical
formulae (Eq. (18)) are also shownin Fig. 18, inwhich Asize 30, 40, 50
denote the average fragment size are 30, 40 and50 mm, respectively.
From this gure, it can be found that the numerical simulations
predict that the average fragment size is in the range of 3050 mm,
which is in the same order of the biggest coarse aggregate. It also
indicates that the numerical results match the empirical statistical
predictions when the fragment size is smaller than the average
fragment size, however, the numerical results and the statistical
predictions differ when the fragment size is larger than the average
fragment size. Nonetheless the numerical results of fragment
distribution are in the same range of the statistical prediction.
Unfortunately, no experimental result on fragment size distribution
is available from the experimental test analysed in this study.
Fig. 16. Fragment distribution (estimated from the numerical results, bottom spalling crater) (a) case I (b) case II (c) homogeneous model.
Fig. 17. Fragment distribution (treated by image analysis program).
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1324
5. Conclusions
Inthe present paper, a two-phasemesoscalemodel is developedto
simulate the dynamic damage and fragmentation of a concrete slab
under blast loading. The distribution of the coarse aggregates is
assumed to follow the Fullers curve. Two different aggregate distri-
bution cases are generated to simulate the example concrete slab. It is
found that the aggregate distribution signicantly affects the crack
and fragment distribution. The cracks in general occur in the mortar
matrix around the coarse aggregates. The fragment size distributions
obtained from the two mesoscale models are compared with that
obtained from the homogeneous model and the results from the
statistical empirical formula. It is found that the estimated fragment
size distributions from the two mesoscale models and the homoge-
neous model are all comparable with the statistical predictions based
on the empirical formula. The fragment size is in the range between
0 and about 6080 mm. The mean size of the fragment is in the same
order as the biggest coarseaggregate. The present study demonstrates
a practical method to predict the fragment size distribution using the
image analysis method and numerical simulations.
3D mesoscale numerical simulation will be carried out in our
future study. Experimental tests will also be carried out to obtain
fragment size distribution and ejecting velocity to calibrate the
present numerical model.
Acknowledgements
The authors would like to thank the Australian Research Council
for nancial support under grant No. DP0774061 to carry out this
research work.
References
[1] LS-DYNA, Keyword users manual. Livermore, California, USA; Livermore
Software Technology Corporation; 2006.
[2] Autodyn, Century dynamics, Theory manual. Concord, California, USA:
Century Dynamics; 2005
[3] Pinto PE. RC elements under cyclic loading, state of the art report. London:
Thomas Telford; 1996.
[4] Chen ACT, Chen WF. Constitutive relations for concrete. Journal of Engineering
Mechanics Division, ASCE 1975;101(EM4):46581.
[5] Kotsovos MD, Newman JB. Generalised stress-strain relations for concrete.
Journal of Engineering Mechanics Divisions, ASCE 1978;104(EM4):84556.
[6] Ottesen NS. A failure criteria for concrete. Journal of the engineering
Mechanics Division, ASCE 1977;103(EM4):52735.
[7] Bischoff PH, Perry SH. Compressive behaviour of concrete at high strain rate.
Materials and Structures 1991;24:42550.
[8] Fu HC, Erki MA, Seckin M. Review of effects on loading rate on concrete in
compression. Journal of Structural Engineering, ASCE 1991;117(12):364559.
[9] Malvar LJ, Ross CA. Review of strain rate effects for concrete in tension. ACI
Materials Journal 1998;95(M73):7359.
[10] Eibl J, Schmidt-Hurtienne B. Strain-rate_sensitive constitutive law for
concrete. Journal of Engineering Mechanics 1999;125(12):141120.
[11] Schuler H, Mayrhofer C, Thoma K. Spall experiments for the measurement of
the tensile strength and fracture energy of concrete at high strain rates.
International Journal of Impact Engineering 2006;32:162550.
[12] Zhou, X.Q., Hao, H. &Deeks, A.J., Modellingdynamicdamageof concreteslabunder
blast loading. In: Proceedings of the 6th International Conference on Shock &
impact Loads on Structures. Perth, Australia: 79, December, 2005, p. 703710.
[13] Holmquist, T.J. and Johnson, G.R. A Computational Constitutive Model for
Concrete Subjected to Large Strains, High Strain Rates, and High Pressures. In:
Proceedings of the 14th International Symposium on Ballistics. Quecbec,
Canada: 1993. p. 591600.
[14] Riedel W., Thoma K., Hiermaier S. and Schmolinske E. Penetrating of rein-
forced concrete by BETA-B-500 numerical analysis using a new macroscopic
concrete model for hydrocodes. In: Proceedings of 9th International Sympo-
sium IEMS, Berlin: 1999. p. 315322.
[15] Gebbeken N, Ruppert M. A new material model for concrete in high-dynamic
hydrocode simulations. Archive of Applied Mechanics 2000;70:46378.
[16] Malvar LJ, Crawford JE, Wesevich JW, Simons D. A plasticity concrete
material model for DYNA3D. International Journal of Impact Engineering
1997;19:84773.
[17] Katayama M, Itoh M, Tamura S, Beppu M, Ohno T. Numerical analysis method
for the RC and geological structures subjected to extreme loading by energetic
materials. International Journal of Impact Engineering 2007;34(9):154661.
[18] Rabczuk T, Eibl J. Simulation of high velocity concrete fragmentation using
SPH/MLSPH. International Journal of Numerical Methods in Engineering
2003;56:141244.
[19] Leppa nen J. Concrete subjected to projectile and fragment impacts: modeling
of crack softening and strain rate dependency in tension. International Journal
of Impact Engineering 2006;32:182841.
[20] Clayton JD. A model for deformation and fragmentation in crushable brittle
solids. International Journal of Impact Engineering 2008;35:26989.
[21] Gatuingt F, Pijaudier-Cabot G. Coupled damage and plasticity modelling in
transient dynamic analysis of concrete. International Journal for Numerical
and Analytical Methods in Geomechanics 2002;26:124.
[22] Wriggers P, Moftah SO. Mesoscale models for concrete: homogenisation and
damage behaviour. Finite Elements in Analysis and Design 2006;42:62336.
[23] Wang ZM, Kwan AKH, Chan HC. Mesoscopic study of concrete I: generation of
random aggregate structure and nite element mesh. Computers and Struc-
tures 1999;70(533):544.
[24] Kwan AKH, Wang ZM, Chan HC. Mesoscopic study of concrete II: nonlinear
nite element analysis. Computers and Structures 1999;70(545):556.
[25] Ha fner S, Eckardt S, Luther T, Ko nke C. Mesoscale modelling of concrete:
Geometry and numerics. Computers and Structures 2006;84:45061.
[26] VanMier JGM, VanVliet MRA. Inuence of microstructure of concrete onsize/scale
effects in the tensile fracture. Engineering Fracture Mechanics 2003;70:2281306.
[27] Bazant ZP, Tabbara MR, Kazemi MT, Pijaudier-Cabot G. Random particle model
for fracture of aggregate or bre composites. Journal of Engineering
Mechanics, ASCE 1990;116:1686705.
[28] Rabczuk T, Eibl J. Modelling dynamic failure of concrete with meshfree
methods. International Journal of Impact Engineering 2006;32:187897.
[29] Grady DE. The spall strength of condensed fragmentation. Journal of the
Mechanics and Physics of Solids 1988;36:35384.
[30] Yew CH, Taylor PA. A thermodynamic theory of dynamic fragmentation.
International Journal of Impact Engineering 1994;15:38594.
[31] Zhang YQ, Lu Y, Hao H. Analysis of fragment size and ejection velocity at high
strain rate. International Journal of Mechanical Sciences 2004;46:2734.
[32] Grady DE, Kipp ME. Geometric statistics in dynamic fragmentation. Journal of
Applied Physics 1985;58:121022.
[33] Grady DE. Particle size statistics in dynamic fragmentation. Journal of Applied
Physics 1990;68:6099105.
[34] Grady DE, Kipp ME. Continuum modeling of explosive fracture in oil shale.
International Journal of Rock Mechanics and Mining Science & Geomechanics
Abstracts 1980;17:14757.
[35] Espinosia HD, Zavattieri PD, Dwivedi SK. A nite deformation continuum/
discrete model for the description of fragmentation and damage in brittle
material. Journal of the Mechanics and Physics of solids 1998;46:190942.
[36] Liu LQ, Katsabanis PD. Development of a continuum damage model for
blasting analysis. International Journal of Rock Mechanics and Mining Science
1997;34(2):21731.
[37] Zhang YQ, Hao H, Lu Y. Anisotropic dynamic and fragmentation of rock
materials under explosive loading. International Journal of Engineering
Science 2003;41:91729.
[38] Rabczuk T, Eibl J, Stempniewski L. Numerical analysis of high speed concrete
fragmentation using a meshfree Lagrangian method. Engineering Fracture
Mechanics 2004;71:54756.
[39] ChenWF. Plasticity inreinforced concrete. NewYork: McGrawHill; 1982. p. 474.
[40] Mazars J. A description of micro- and macroscale damage of concrete struc-
tures. Engineering Fracture Mechanics 1986;25(5/6):72937.
[41] Ma GW, Hao H, Zhou YX. Modeling of wave propagation induced by under-
ground explosion. Computers and Geotechnics 1998;22:283303.
0
10
20
30
40
50
60
70
80
90
100
0 20 40 60 80 100 120
fragment size (mm)
p
e
r
c
e
n
t
a
g
e

p
a
s
s
i
n
g

(
%
)
Case I
Case II
Homo
Asize 30
Asize 40
Asize 50
Fig. 18. Fragment size distribution.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1325
[42] Comite Euro-International du Beton. CEBFIP model code 1990. Trowbridge,
Wiltshire, UK: Redwood Books; 1993.
[43] Cho SH, Ogata Y, Kaneko K. Strain-rate dependency of the dynamic tensile
strength of rock. International Journal of Rock Mechanics & Mining Science
2003;40:76377.
[44] Wang QZ, Li W, Song XL. A method for testing strength and elastic
modulus of rock materials using SHPB. Pure and Applied Geophysics
2006;163:1091100.
[45] Cai M, Kaiser PK, Sourineni F, Su K. A study on the dynamic behaviour of the
Meuse/HauteMarne argillite. Physics and Chemistry of the Earth
2007;32:90716.
[46] Lok TS, Zhao PJ, Li XB and Lim CH. Dynamic stress-strain response of granite
from split Hopkinson Pressure Bar tests. In: Proceedings of the 5th Asia-Pacic
Conference on Shock & Impact Loads on Structures. November 1214, 2003,
Changsha, Hunan, China: p. 27786.
[47] Li XB, Lok TS, Zhao J. Dynamic characteristics of granite subjected to
intermediate loading rate. Rock Mechanics and Rock Engineering
2005;38(1):2139.
[48] Li HB, Zhao J, Li JR, Liu YQ and Zhou QC. Experimental studies on the strength
of different rock types under dynamic compression. International Journal of
Rock Mechanics and Mining Science. 2004, 41(3). Paper 1A 12 SINOR-
OCK2004 Symposium.
[49] Rouabhi A, Tijani M, Moser P, Goetz D. Continuum modelling of dynamic
behaviour and fragmentation of quasi-brittle materials: application to rock
fragmentation by blasting. International Journal for Numerical and Analytical
Methods in Geomechanics 2005;29:72949.
X.Q. Zhou, H. Hao / International Journal of Impact Engineering 36 (2009) 13151326 1326