Engineering Fracture Mechanics 130 (2014) 5364

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

3D simulation of dependence of mechanical properties

of porous ceramics on porosity
A.Yu. Smolin a,b,, N.V. Roman b, I.S. Konovalenko a, G.M. Eremina a,b, S.P. Buyakova a,b,
S.G. Psakhie a,b
a
b

Institute of Strength Physics and Materials Science, SB Russian Academy of Science, pr. Academichesky 2/4, Tomsk 634021, Russia
Tomsk State University, pr. Lenina 36, Tomsk 634050, Russia

a r t i c l e

i n f o

Article history:
Available online 13 April 2014
Keywords:
Porous ceramics
3D modeling
Discrete approach
MCA method

a b s t r a c t
3D computer simulation of mechanical behavior of a brittle porous material under uniaxial
compression is considered. The movable cellular automaton method, which is a representative of particle methods in solid mechanics, is used for computation. In an initial
structure the automata are positioned in fcc packing. The pores are set up explicitly by
removing single automata from the initial structure. The computational results show that
dependence of strength and elastic properties of the modeled material on porosity below
percolation threshold (material with closed pores) differs from the dependence for porosity
above the threshold (permeable material). The results obtained are in close agreement
with available experimental data.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Let us consider a porous material in which all pores have the same size. If porosity of the material is small, each pore is
closed and isolated form the others. With porosity increasing the number of pores becomes greater and the distance among
pores decreases. When porosity reaches percolation threshold the pores are not closed anymore, connect with the others in a
spanning cluster and consequently form a new morphology. It is well known that dependence of strength and elastic properties of porous materials on porosity is determined by the pore morphology (this problem has a long history [1,2] and is
very interesting up till now [38]). Thus, such a dependence should undergoes a change when the percolation threshold
is exceeded [9].
Notice that experimental investigation of the problem is very difcult, because of stochastic nature of the pore size and
morphology in real materials. For example, it is extremely hard to fabricate samples with various porosity values and the
same porous structure, for example the same pore size. Of course, the modern technology like selective laser sintering allows
producing very complex 3D structures but only for special materials, mainly polymers. At the same time potentialities of the
modern computational mechanics allow to study in detail not only deformation of complex heterogeneous materials but also
their fracture.
The most of research in computational mechanics are performed using nite element method. But for modeling severe
distortion and failure of a material, converting nite elements into particles is more convenient. For example the authors
Corresponding author at: Institute of Strength Physics and Materials Science, SB Russian Academy of Science, pr. Academichesky 2/4, Tomsk 634021,
Russia. Tel.: +7 9059 91 05 42; fax: +7 3822 49 25 76.
E-mail address: asmolin@ispms.tsc.ru (A.Yu. Smolin).
http://dx.doi.org/10.1016/j.engfracmech.2014.04.001
0013-7944/ 2014 Elsevier Ltd. All rights reserved.

54

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

Nomenclature
C
di
Eeff
E0
~
Fi
~
F pair
ij

~
FX
i
Gi
hij
bJ i

porosity
size of the ith automaton
effective elastic modulus of the sample
elastic modulus of the material
total force acting on the ith automaton
pair part of the force acting between the automata i and j
volume-dependent part of the force ~
Fi
shear modulus of the material of the ith automaton
overlap in the pair of automata
moment of inertia of the ith automaton

~
K rot
ij

torque caused by relative rotation of automata in the pair ij

Ki
~lshear

bulk modulus of the material of the ith automaton

ij

mi
~ ij
M
~
nij
Pj
qij
~
Ri
r ij
Sij
t
~
t ij
~
vi
Vi
~
cij

gij
~
hi
nij

rci
ri;ab
rint
~
sij
~ ij
x
i; j
a; b
D

relative tangential displacement in the pair ij

mass of the ith automaton
total torque acting between the automata i and j
unit vector directed from the ith automaton to the jth one
pressure in the volume of the jth automaton
distance from the center of the ith automaton to the point of its interaction with the jth one
position of the ith automaton
distance between the automata i and j
area of interaction of the automata i and j
time
unit vector of the direction of tangential part of the force ~
F pair
ij
translation velocity of the ith automaton
volume of the ith automaton
shear strain in the pair ij
value of specic force of central interaction in the pair ij
rotation vector of the ith automaton
normal strain of the ith automaton in the direction of the jth one
strength of material of the ith automaton
components of the average stress tensor in the ith automaton
stress intensity (von Mises stress)
specic force of tangential interaction in the pair ij
rotational velocity of the pair ij as a whole (rigid body)
subscripts denoting automaton number
subscripts denoting components of tensors and vectors
denotes increment over one time step

of work [10] use a so called meshless (or mesh free) method for this purpose (namely smoothed particle hydrodynamics
(SPH) and generalized particle algorithm (GPA)). To be precise meshless methods are not pure particle ones. They just use
scattered nodes associated with centers of nite volumes to discretize continuum mechanics equations in space, i.e. they
are really based on continuum approach. Concerning the problem of modeling porous media it should be noted that at least
35 meshless particles are required to represent a pore wall. A true particle method can describe the wall using one particle.
Thus, the most advantage in modeling fracture of heterogeneous materials seems to belong to particle method. That is
why the purpose of this study was to develop a computational model of porous ceramics based on particle approach and
to study elastic and strength properties of the modeled material in wide range of porosity.
2. Method of movable cellular automata
Method of movable cellular automata (MCA) [1116] is a new efcient numerical method in particle mechanics that is
different from methods in the traditional continuum mechanics. It should be noted that the particle methods started from
well-known molecular dynamics (MD) method which was developed to study materials at the atomic level. At the same
time, the possibilities of atomic description of the behavior of a solid on spatial and temporal scales that are of interest
for engineering application are severely limited. This motivated the development of MD-like methods for meso- and

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

55

macroscopic description of material (particle methods) in which structural elements are of nite size (unlike atoms, which
are point masses) and hence interact only with the immediate surroundings.
The most well-known representative of this group of methods is discrete element method (DEM) [17,18]. DEM is now
widely used to study deformation of granular (loose) and weakly bound media, in particular rheological peculiarities of these
systems, their fracture and mixing [19,20]. At the same time, until recently applicability of DEM to study mechanical phenomena in consolidated media has been restricted mainly to brittle porous materials [18,21] due to insufciently advanced
mathematical interaction models of discrete elements. In particular, the majority of discrete element models use pairwise
(two-particle) interaction potentials or forces. This simplication can entail a series of articial effects in the response of
a particle ensemble like anisotropy, unregulated macroscopic values of shear modulus or Poissons ratio and so on
[18,21]. Among the problems arisen from these articial effects there is correct simulation of irreversible strain accumulation (plasticity) of materials.
Our research shows that many basic problems of DEM, including those with description of consolidated solids at various
scales, can be solved by using many-particle interaction. In [15] a generalized approach to construction of many-particle
interaction forces for discrete elements is described which is based on interaction force notation similar to the interatomic
potential notation for embedded atom method [22]. Namely, to construct a many-particle interaction model, the expression
for the force acting on a discrete element i from the surrounding N i particles should be written in the form:

~
Fi

Ni
X
~
F pair
F Xi :
~
ij
j1

This force is represented as a superposition of the pair components ~

dependent on the spatial position or displacement of
F pair
ij
the element i relative to the neighbor j and the volume-dependent component ~
FX
i related to collective effects of the
surroundings. This many-particle approach for notation of interaction forces is an essential part of the movable cellular
automaton method.
Within the frame of MCA, it is assumed that any material is composed by a certain amount of elementary objects of nite
size (automata) which interact among each other and can move from one location to another, thereby simulating a real
deformation process. The automaton motion is governed by the NewtonEuler equations:

8
Ni
X
>
2~
>
~
>
F pair
F Xi
m d Ri
~
>
ij
>
< i dt2
j1
Ni
>
X
>
>
b d2~hi
~ ij
>
M
>
: J i dt2

j1

Ri ; ~
where ~
hi ; mi and bJ i are the location vector, rotation vector, mass and moment of inertia of ith automaton respectively,
pair
~
FX
F ij is the interaction force of the pair of ith and jth automata, ~
i is the volume-dependent force acting on ith automaton
and depending on the interaction of its neighbors with the remaining automata. In the latter equation,
~pair
~rot
~ ij q ~
M
ij nij  F ij K ij , here qij is the distance from the center of ith automaton to the point of its interaction (contact)
Rj  ~
Ri =rij is the unit vector directed from the center of ith automaton to the jth one and r ij is the
with jth automaton, ~
nij ~
distance between automata centers (Fig. 1), ~
K rot
ij is the torque caused by relative rotation of automata in the pair (see below).
Note, that automata of the pair may represent the parts of different bodies or of one consolidated body. Therefore its
interaction is not always really contact one. That is why we put the word contact in quotation marks. More of that, as
it shown in Fig. 1 size of the automaton is characterized by one parameter di , but it does not mean that the shape of the
automaton is sphere. When we compute the volume of an automaton, its shape is determined by area of its contacts with
neighbors. For example, if we use initial fcc packing then the automata are shaped like a rhombic dodecahedron, but if we

Fig. 1. Schematic of determination of the spatial parameters of a pair of the movable cellular automata i and j.

56

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

use cubic packing then the automata are cube-shaped. But when we compute interautomation forces or torques it is
enough to use only one size parameter di and value of the contact area Sij . In this case we can imagine the automaton like
a sphere.
For locally isotropic media the volume-dependent component can be expressed in terms of the pressure P j in the volume
of the neighboring automaton j as follows [15,16]:
Ni
X
~
F Xi A Pj Sij~
nij
j1

where Sij is the area of interaction surface of automata i and j and A is a material parameter.
The total force acting on automaton i can be represented as a sum of explicitly dened normal F nij and tangential (shear) F tij
components:

~
Fi

N i h
Ni 

i X

X
~
~
F pair;n
F pair;t
F nij ~
F tij :
hij  AP j Sij ~
~lshear
~
tij
nij ~
ij
ij
ij
j1

j1

F nij and ~
F tij are the normal and tangential pair interaction forces depending respectively on the automata overlap hij
where ~
lshear
(Fig. 1a) and their relative tangential displacement ~
(Fig. 1b) calculated with taking into account rotation of the both
ij
automata [14,15]. Note, that although the last expression of equation Eq. (2) formally corresponds to the form of element
interaction in conventional discrete element models [1721], it differs fundamentally from them in many-particle central
interaction of the automata.
Using homogenization procedure for stress tensor in a particle described in [15,21], the expression for components of the
average stress tensor in automaton i takes the form:
N

ri;ab

i
1X
q nij;a F ij;b ;
V i j1 ij

where a and b denote the axes X; Y; Z of the laboratory coordinate system, V i is the current volume of automaton i; nij; a is
the a-component of unit vector ~
nij and F ij;b is b-component of the total force acting at the point of contact between automata i and j.
The pressure Pi , or what is the same, the mean stress in the automaton volume can be determined from thus calculated
stress tensor components:

Pi 

ri;xx ri;yy ri;zz

3

Tensor components allow us to calculate the other tensor invariants in the automaton volume, in particular stress
intensity:

rxx  ryy 2 ryy  rzz 2 rzz  rxx 2 6r2xy r2yz r2xz :

rint
i p
2

From Eqs. (1)(3) it follows that the specic form of the expressions for ~
F nij and ~
F tij determines rheological behavior of a
model medium.
For further convenience, the interaction parameters of movable cellular automata are considered in relative (specic)
units. Thus, the central and tangential interaction of the automata i and j are characterized by the corresponding stresses
gij and sij :

F nij gij Sij

~
sij Sij
F tij ~

Note, that in the most of papers devoted to the description and use of MCA the equations and formulae of the method are
written for two-dimensional case. This study uses three-dimensional version of the MCA method. That is why herein shear
stress ~
nij .
sij is a vector in the plane which is normal to ~
To characterize deformation of automaton i under its normal interaction with automaton j we can use the following
dimensionless parameter (normal strain)

nij

qij  di =2
:
di =2

In general case each automaton of a pair represents different material and overlap of the pair is distributed among ith and jth
automata:

Dhij Dqij Dqji Dnij di =2 Dnji dj =2;

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

57

where symbol D denotes increment of a parameter per time step Dt of numerical integration of the motion equations Eq. (1).
The rule of strain distribution in the pair is intimately associated with the expression for computing interaction forces of the
automata. This expression for central interaction is similar to Hookes relations for diagonal stress tensor components:



2Gi
Pi ;
Dgij 2Gi Dnij 1 
Ki

where K i is the bulk modulus, and Gi is the shear modulus of the material of ith automaton, Pi is the pressure of automaton i
that may be computed using formulae Eqs. (3) and (4) at previous time step or by predictorcorrector scheme.
To determine a parameter characterizing shear deformation in pair of automata ij we start with kinematics formula for
free motion of the pair as a rigid body

~ ij  ~
~
vj  ~
vi x
r ij ;

10

~ ij is the rotational velocity of the pair as a

Rj  ~
Ri ; ~
where ~
v i is the translation velocity of the ith automaton centroid, x
rij ~
whole (rigid body). If we multiply both sides of equation Eq. (10) on the left by ~
rij and neglect rotation about the axis con~ ij  ~
necting centers of the automata of the pair (i.e. let x
r ij 0 because rotation about the axis of the pair does not produce
shear deformation) then we get the following formula

~ ij
x

~
vj  ~
v i ~nij  ~
vj  ~
vi
r ij  ~

:
rij
r2ij

11

Besides such rotation of the pair as a whole (dened by the difference in translational velocities of the automata), each
automaton rotates with its own rotational velocity (Fig. 2a). The difference between these rotational velocities produces
shear deformation. Thus, increment of shear deformations of automata i and j per time step Dt is dened by the relative tanlshear
gential displacement at the contact point D~
divided by the distance between the automata
ij

~
cij ~
cji



~ ij  x
~ i  ~
D~lshear
qij x
nij Dt
ij

:
r ij
r ij

12

The expression for tangential interaction of movable cellular automata is similar to Hookes relations for non-diagonal
stress tensor components:

D~
sij 2Gi ~cij ;

13

and is pure pairwise.

The difference in automaton rotation leads also to deformation of relative bending and torsion (only in 3D) of the pair
2b. It is obvious that resistance to relative rotation in the pair cause the torque, which value is proportional to the difference
between the automaton rotations:

~
~ ~
K rot
ij Gi Gj hj  hi :

14

Formulae Eqs. (1)(4), (6)(9), (11)(14) describe mechanical behavior of a linearly elastic body in the framework of MCA
method. Note, that relations Eqs. (8), (9), (12) and (13) are written in increments, i.e., in the hypoelastic form. In paper [23] it
is shown that this model gives the same results as numerical solving usual equation of continuum mechanics for isotropic
linearly elastic medium by nite-difference method. That makes it possible to couple MCA method with numerical methods
of continuum mechanics. The results of paper [14] show that it is taking into account rotation that allows the movable
cellular automata correctly describe isotropic response of material.

Fig. 2. Translational and rotational velocities of the automata (a) and deformation due to relative rotation between automata in the pair (b).

58

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

We propose to apply MCA for studying brittle materials herein, but in [15,16] it is shown how to describe elastoplastic
behavior within the MCA method.
The equations of motion Eq. (1) for the system of movable cellular automata are numerically integrated with the use of
velocity Verlet algorithm [24]

8
~
F ni
~
>
Rn1
Rni ~
~
v ni Dt 2m
Dt2
>
i
>
i
>
>
>
~n
>
>
~ ni Dt Mi Dt2
>~
hn1
hni x
~
<
i
b
2J i

~n ~n1
>
>
~
v n1
~
v ni F i 2mF ii Dt
>
>
i
>
>
>
~ n M
~ n1
>
M
>
:x
~ n1
~ ni i i Dt
x
i
b
2Ji

modied by introducing a predictor for estimation of ri;ab at the current time step n + 1 in order to calculate P n1
and then
i
~
.
F n1
i
A pair of elements might be considered as a virtual bistable automaton (bound and unbound states), which permits
simulation of fracture by the MCA. Within the framework of MCA method crack initiation is imitated by bound ! unbound
transition of the pair state. Fracture criterion depends on physical mechanisms of material deformation. An important
advantage of the described above formalism is that it makes possible direct application of conventional fracture criteria
(HuberMisesHencky, DruckerPrager, MohrCoulomb, Podgorski, etc.), which are written in tensor form. Note, that
switching of a pair of automata to an unbound state would result in a changeover in the forces acting on the elements;
in particular, they would not resist moving away from one another.
Thus, a fracture criterion has to be calculated with use of mean values of stress/strain tensor components in the considered pair ij:

rij;ab

ri;ab qij rj;ab qji

rij

The condition of fracture based on HuberMisesHencky criterion will take the form:
c
c
rint
ij > minri ; rj :

where rci is assigned threshold value (strength) for the material of automaton i, stress intensity rint
ij is calculated with use of
rij;ab .
3. Computational model of porous ceramics
Herein we used MCA method for 3D computer simulation of mechanical behavior of a porous material under uniaxial
compression. Response function of automata used in this study corresponded to ZrO2 ceramics with average size of pores
commensurable with the grain size of the material and porosity equal to 2% (i.e. Youngs modulus of intact material was
equal to 80 GPa and Poissons ratio 0.3). According to the pore distribution diagram of this material [25] the automaton size
in the computations was equal to 1 lm.
All the modeled samples were bricks with a square base. In Fig. 3a one can see a solid sample as fcc packing of automata.
To simulate loading, one and the same velocity in the vertical direction was assigned for all the automata of the upper layer
(dark gray particles at the top of the sample in Fig. 3), while the automata of the lower layer (dark gray particles at the

Fig. 3. Solid (a) and poropus (b) samples for modeling using MCA method as a fcc packing of spheres.

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

59

Fig. 4. Periodic pores (black) in closed packing.

bottom of the sample in Fig. 3) were xed in the vertical direction. For the stress-state to be uniformly distributed, all the
automata were allowed to move along the horizontal plane. In order to preclude dynamic effects [12], the velocity of the
upper automata increased gradually from 0 to 1 cm/s and then remained constant to the end of the loading.
Pores were generated by removing single automata from the initial fcc structure (Fig. 3b). The porosity values were varied
from 0% to 65%. Note, that the maximal porosity on the assumption of pore disconnection in closed packing of spheres for 2D
is 1/3 = 33% (Fig. 4a) while for 3D it is 1/4 = 25% (in Fig. 4b only one closed packing plane is shown for fcc). In this case the
pores are situated regularly.
The maximal value of closed porosity for random removing of single automata from fcc packing is 17.6%. The site percolation threshold for 3D fcc packing is 19.9% [26]. Consequently, all the samples used in this study with porosity greater than
20% contained clusters of interconnecting pores. Pores in the samples with porosity greater than 25% were interconnected
(made up a permeable material).
In [5,27] it is shown that porosity structure denes not only elastic, but also strength properties of the material. The clusters of interconnecting pores present a new element of structure in addition to single closed pores. Thus, with increasing of
porosity a change of elastic and strength properties of a material is expected after transition of the percolation threshold.
4. Determining the representative volume
At the rst stage it is required to determine the representative volume of the modeled material. For material without
pores the representative volume was determined based on four solid samples. For this purpose the convergence of effective
elastic modulus of the sample Eeff to the material elastic modulus E0 with increasing the sample base size a was analyzed.
The convergence was assumed to be satised if the difference between Eeff and E0 was not greater than 3%. The computations suggested that the representative volume corresponded to the brick of 10 lm base size. The fracture pattern (cracks)
in this sample is shown in Fig. 5. The cracks are generated from the stress concentrators (edges) and propagate along the
maximum tangential stress direction.

Fig. 5. Fracture pattern of a solid ceramic sample under compression: (a) lines connect linked automata; (b) lines connect unlinked automata.

60

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

Fig. 6. First cracks in samples with different values of porosity C under compression.

(Ei-E)/E (%)

(Ei-E)/E (%)

4
2
0
-2
-4
-6
16

20

24

a (m)

(a)

28

32

12

8
6
4
2
0
-2
-4
-6
-8

(Ei-E)/E (%)

4
0
-4
-8

-12
-16
16

20

24

a (m)

(b)

28

32

16

20

24

28

32

a (m)

36

40

44

(c)

Fig. 7. Convergence of effective elastic moduli of the modeled samples on their basis size for different porosity: (a) 10%; (b) 25%; (c) 50%.

If a material contains pores the cracks may initiate not in the edges but in the bulk regions of the highest local porosity.
The direction of cracks propagation in such material is dened by the porous topology. Fig. 6 shows the rst cracks in porous
samples with various values of porosity. One can see that the failure behavior of the porous material with large porosity and
the failure of a solid material differ dramatically. In particular, the path of crack propagation in porous material is rather
crinkly.
The fracture pattern of the modeled brittle porous 3D samples qualitative corresponds to 2D simulation results [27]. The
quantitative difference is dened by the fact that porosity in 2D sample is permeable in the normal direction to the
computational plane.

61

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

2000
0

(MPa)

1500

1000
1
2

500

3
4
5

0.5

1.5

2.5

(%)
Fig. 8. Loading diagrams for ceramics with porosity C n  10%.

To analyze numerically the dependence of mechanical properties of the modeled material on its porosity it is necessary to
determine the size of representative sample (volume) for each porosity value. For this purpose the base size of the samples, a,
varied in the range from 10 to 45 lm. This made it possible to determine the representative volume of simulated materials,
which would be an indication that any further increase in the sample size would cause no signicant change in the elasticity
modulus and strength limit of the simulated samples under uniaxial compression.
On the base of calculation results the convergence of elastic and strength properties was analyzed for the simulated samples as a function of increasing sample size. To do this, the elastic moduli were calculated for the samples under compression
(the slope of the rst linear part of the loading diagram); the strengths (the maxima on the diagrams) were also calculated
for each sample. The relative deviation from the mean Youngs modulus, E, of the calculated values, Ei , observed for all the
samples is demonstrated in Fig. 7. The presented data suggests that for porosity value less than 15% the base size of representative sample is 20 lm (Fig. 7). For porosity value varying from 15% up to 35% the size of representative volume is found
to be 30 lm (Fig. 7b). For porosity varying from 35% up to 50% the size of representative volume is 40 lm (Fig. 7c). Note that
for porous material the convergence criterion was not as strong as for intact material. It was assumed to be enough if the
difference between Ei and E was not greater than 5%.
5. Studying elastic and strength properties of porous ceramics
The loading diagrams of the modeled samples are shown in Fig. 8. They are in close qualitative agreement with 2D simulation results [27]. A quantitative difference between 2D and 3D results could be explained by the different inuence of
porosity on effective porous characteristics in 2D and 3D tasks.
First, it has to be noted that porous 3D samples can demonstrate quasi-viscous regime of fracture which previously was
observed for 2D models [27]. It means that ceramic samples with large porosity stochastically distributed in space can miss
the stage of rapid propagation of the main crack and a sharp drop in the strength, which is typical for brittle materials. Note
that such regime of fracture is determined entirely by the pore space structure and observed for 3D samples having porosity
greater than 40% (Fig. 8).
0

-1

-0.1
-0.2

log ( E/E0)

log (E/E0)

-2
-3
-4

-0.3
-0.4
-0.5

-5

-0.6

-6

-0.7

-7

-0.8

10

20

30

40

C (%)

(a)

50

60

70

10

15

20

25

C (%)

(b)

Fig. 9. Plots of logarithm of normalized modulus of the modeled ceramics versus its porosity: (a) all calculations; (b) data for small porosity.

62

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

Let us consider the dependence of the effective elastic modulus of 3D samples on its porosity. In Fig. 9a each point with
error bar represents the value averaged on ve representative samples with various pore distributions in space. This plot
obviously can be divided into two characteristic parts connected with porous structure: the rst corresponds to closed pores
(520%), the second corresponds to interconnected pores (2065%). To substantiate that let us t the computational results
with phenomenological equation

E E0 1  C=C max m ;

15

where C max and m are adjustable parameters. The tting curve with the parameter values C max 0:7213 and m 1:9026 is
plotted with a solid line in Fig. 9a. One can see that this curve ts very well the calculated points for porosity not greater than
50%. But if we zoom this plot and look at porosity range of 020% (Fig. 9b) then we can see that the calculated points actually
are placed right between the curve (15) and the tted exponent E E0 expmC with value of the parameter m 3:0872
(straight dashed line in Fig. 9b).
The calculated points in porosity range of 2065% t much better the following equation:

E E0 EP 1  C=C max m :

16

The curve corresponding to Eq. (16) with the parameter values EP 0:9472; C max 0:6761 and m 1:6292 is plotted in
Fig. 9a as a dashed line. Physical meaning of the adjustable parameters of Eqs. (15) and (16) is clear. C max means the maximal
value of porosity of a sample that have any strength for the given pore morphology. For example, the procedure used for
generating porosity in this work has a limit beyond which the remaining automata would not interact with each other. They
would hang in the air and not resist to any load. Indeed, it is well known that real materials having large porosity have pore
wall thickness much less than pore size. Thus, to generate larger porosity for samples based on fcc packing of automata it is
necessary to remove not a single automata, but also all its neighbors. Factor EP in Eq. (16) means that at zero porosity the
modulus is not equal to the value of intact material, but it allows to get exact value at other porosity value, for example
at percolation limit.
The inuence of percolation threshold is observed even better from dependence of strength of the simulated ceramics on
its porosity. In Fig. 10a again each point with error bar represents the value averaged on ve representative samples with
various pore distributions in space. In this plot the solid line presents the tted curve using Eq. (15) with values of the adjustable parameters C max 0:7277 and m 1:9156; the dashed line corresponds to Eq. (16) with the parameter values
EP 0:9522; C max 0:6528 and m 1:5161. The main distinction of strength variation with porosity from elastic modulus
variation can be observed for small porosity. Thus, it is shown in Fig. 10b that in porosity range of 025% the calculated
points for strength are tted much better using equation

r r0

1:0  C=C max m

:
1:0 C=C 1 n

17

The corresponding curve with the parameter values C max 0:3481; m 4:8947; C 1 0:4018 and n 0:6903 is plotted in
Fig. 10b as a dashed line.
In Fig. 11 the calculation results are plotted as tted curves, the experimental data taken from [25,28] are presented in
Fig. 11 as points with error bars for comparison. One can see that the computational elastic modulus is in very close agreement with the experimental data. Calculated strength differs from experimental evidence signicantly in the range of porosity of 1040%. It can be explained by ideal structure of the modeled material, because it contains only equiaxed pores and no
stress concentrators specic for real ceramics. Besides, according to [25] percolation transitions in ceramics ZrO2 cause

-1

-0.1
-0.2

log (/0)

log ( /0)

-2
-3
-4

-0.3
-0.4
-0.5

-5

-0.6

-6

-0.7

-7

-0.8
0

10

20

30

40

50

60

70

10

15

C (%)

C (%)

(a)

(b)

20

25

Fig. 10. Plots of logarithm of normalized strength of the modeled ceramics versus porosity: (a) all calculations; (b) data for small porosity.

63

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

0.8

0.8

0.6

0.6

E/E0

/0

0.4

0.4

0.2

0.2

10

20

30

40

50

60

70

10

20

30

40

C (%)

C (%)

(a)

(b)

50

60

70

Fig. 11. Calculated properties of modeled ceramics versus porosity (tted curves as solid lines) and corresponding experimental data (points with error
bars): (a) normalized Youngs modulus; (b) normalized strength.

microstructure change. In particular, internal stresses in ceramics with continuous porous structure restrain grain growth.
But one can see that experimental data clear exhibits break in the plot of strength versus porosity. Unfortunately, a lack
of data for small porosity does not allow to see such a break in the plot of elastic modulus versus porosity.
Thus, simulation of uniaxial compression of brittle porous 3D samples by the movable cellular automaton method allow
to conclude that percolation transition from the closed pores to interconnected pores in the porous material leads to change
in the dependence of its elastic and strength properties on porosity.
It is necessary to note that this result could be obtained only in three-dimensional simulation because two-dimensional
samples with interconnected porosity are not topologically connected and do not resist to mechanical loading. Besides, the
porosity value at which formation of interconnected clusters occurs in a two-dimensional case (22.4%) is considerably less
than the corresponding percolation threshold (69.6%).
6. Conclusions
Deformation and fracture of porous ceramics have been successfully simulated using the movable cellular automaton
method in wide range of porosity. The method based on discrete element approach allows taking into account pores explicitly and modeling crack initiation and development. To reveal the inuence of porosity percolation on the porosity dependence of mechanical properties of the material we consider only equiaxed pores of one and the same size stochastically
distributed in space. We use this simplication to avoid the inuence of other structure parameters on the studied problem.
The main results are summarized as follows:
 Use of simple fracture criterion based on threshold for stress intensity (von Mises stress) shows that cracks in porous
ceramics initiate in the regions of highest local porosity, which could be far away from the usual stress concentrators
for solid samples.
 The simulations for wide range of porosity show that the plots of strength and elastic modulus of the brittle porous samples versus porosity have a change at the porosity value corresponding to percolation threshold. The obtained results are
in close agreement with available experimental data.
 The proposed 3D MCA model of porous ceramics can be obviously extended for the case of hierarchical porous structure
using the approach proposed for 2D computations [29].

Acknowledgement
This work was supported by the interdisciplinary integration projects Nos. 39 and 66 of the Siberian Branch of the Russian
Academy of Sciences.
References
[1] Wu TT. The effect of inclusion shape on the elastic moduli of a two-phase material. Int J Sol Struct 1966;2:18.
[2] Walpole LJ. On the overall elastic moduli of composite materials. J Mech Phys Solids 1969;17:23551.
[3] Bruno G, Efremov AM, Levandovskyi AN, Clausen B. Connecting the macro- and microstrain responses in technical porous ceramics: modeling and
experimental validations. J Mater Sci 2011;46:16173.
[4] Roberts A, Garboczi E. Elastic properties of model porous ceramics. J Am Ceram Soc 2000;83(12):30418.

64

A.Yu. Smolin et al. / Engineering Fracture Mechanics 130 (2014) 5364

[5] Konovalenko IgS, Smolin AYu, Korostelev SYu, Psakhe SG. Dependence of the macroscopic elastic properties of porous media on the parameters of a
stochastic spatial pore distribution. Tech Phys 2009;54(5):75861.
[6] Skripnyak VA, Skripnyak EG, Kozulin AA, Skripnyak VV, Korobenkov MV. Computer simulation of the relation between mechanical behavior and
structural evolution of oxide ceramics under dynamic loading. Russ Phys J 2009;52(12):13008.
[7] Schpfer MPJ, Abe S, Childs C, Walsh JJ. The impact of porosity and crack density on the elasticity, strength and friction of cohesive granular materials:
insights from DEM modelling. Int J Rock Mech Mining Sci 2009;46:25061.
[8] Miled K, Sabb K, Roy RL. Effective elastic properties of porous materials: homogenization schemes vs experimental data. Mech Res Commun
2011;38(2):1315.
[9] Chelidze T, Reuschle T, Darot M, Gueguen Y. On the elastic properties of depleted relled solids near percolation. J Phys C: Solid State Phys
1988;21:L100710.
[10] Johnson GR, Stryk RA. Conversion of 3D distorted elements into meshless particles during dynamic deformation. Int J Impact Engng 2003;28:94766.
[11] Psakhie SG, Moiseyenko DD, Smolin AYu, Shilko EV, Dmitriev AI, Korostelev SYu, et al. The features of fracture of heterogeneous materials and frame
structures. Potentialities of MCA design. Comput Mater Sci 1999;16(14):33343.
[12] Psakhie SG, Smolin AYu, Korostelev SYu, Dmitriev AI, Shilko EV, Alekseev SV. About the features of transient to steady state deformation of solids. J
Mater Sci Technol 1997;13(1):6972.
[13] Psakhie SG, Zavshek S, Jezershek J, Shilko EV, Dmitriev AI, Smolin AYu, et al. Computer-aided examination and forecast of strength properties of
heterogeneous coal-beds. Comput Mater Sci 2000;19(14):6976.
[14] Smolin AYu, Roman NV, Dobrynin SA, Psakhie SG. On rotation in the movable cellular automaton method. Phys Mesomech 2009;12(34):1249.
[15] Psakhie SG, Horie Y, Shilko EV, Smolin AYu, Dmitriev AI, Astafurov SV. Development of discrete element approach to modeling heterogeneous elasticplastic materials and media. Int J Terraspace Sci Engng 2011;3(1):93125.
[16] Psakhie SG, Smolin AYu, Shilko EV, Anikeeva GM, Pogozhev YuS, Petrzhik MI, et al. Modeling nanoindentation of TiCCaPON coating on Ti substrate
using movable cellular automaton method. Comput Mater Sci 2013;76:8998.
[17] Cundall PA, Strack ODL. A discrete numerical model for granular assemblies. Geotechnique 1979;29(1):4765.
[18] Jing L, Stephansson O. Fundamentals of discrete element method for rock engineering: theory and applications. Oxford: Elsevier; 2007.
[19] Sibille L, Nicot F, Donze FV, Darve F. Material instability in granular assemblies from fundamentally different models. Int J Numer Anal Meth Geomech
2007;31(3):45781.
[20] Jauffrs D, Liu X, Martin CL. Tensile strength and toughness of partially sintered ceramics using discrete element simulations. Engng Fract Mech
2013;103:13240.
[21] Potyondy DO, Cundall PA. A bonded-particle model for rock. Int J Rock Mech Min Sci 2004;41(8):132964.
[22] Daw MS, Foiles SM, Baskes MI. The embedded-atom method: a review of theory and applications. Mater Sci Rep 1993;9(78):251310.
[23] Psakhie SG, Smolin AYu, Stefanov YuP, Makarov PV, Shilko EV, Chertov MA. Modeling the behavior of complex media by jointly using discrete and
continuum approaches. Tech Phys Lett 2004;30(9):7124.
[24] Swope WC, Andersen HC, Berens PH, Wilson KR. A computer simulation method for the calculation of equilibrium constants for the formation of
physical clusters of molecules: application to small water clusters. J Chem Phys 1982;76:63749.
[25] Kulkov SN, Buyakova SP, Maslovsky VI. Structure, phase composition and mechanical properties of zirconia based ceramics. Bull Tomsk State Univ
2003;13:3457 (in Russian).
[26] Lorenz CD, Ziff RM. Universality of the excess number of clusters and the crossing probability function in three-dimensional percolation. J Phys A: Math
Gen 1998;31:814757.
[27] Smolin AYu, Konovalenko IgS, Kulkov SN, Psakhie SG. Quasi-viscous fracture of brittle media with stochastic pore distribution. Tech Phys Lett
2006;32(9):73840.
[28] Kulkov SN, Buyakova SP, Smolin AYu, Roman NV, Kinelovskii SA. Percolation transitions in porous structure and their effect on physicochemical
properties of ceramics. Tech Phys Lett 2011;37(4):3603.
[29] Konovalenko IgS, Smolin AYu, Psakhie SG. Multiscale approach to description of deformation and fracture of brittle media with hierarchical porous
structure on the basis of movable cellular automaton method. Fratt Int Strutt 2013;24:7580.