ARTICLE IN PRESS

Journal of the Mechanics and Physics of Solids 56 (2008) 896–923

www.elsevier.com/locate/jmps

An interacting micro-crack damage model for failure of brittle

materials under compression
B. Paliwal, K.T. Ramesh
Department of Mechanical Engineering, The Johns Hopkins University, Baltimore, MD 21218, USA
Received 19 November 2006; received in revised form 1 April 2007; accepted 19 June 2007

Abstract

A model is developed for brittle failure under compressive loading with an explicit accounting of micro-crack
interactions. The model incorporates a pre-existing flaw distribution in the material. The macroscopic inelastic
deformation is assumed to be due to the nucleation and growth of tensile ‘‘wing’’ micro-cracks associated with frictional
sliding on these flaws. Interactions among the cracks are modeled by means of a crack-matrix-effective-medium approach
in which each crack experiences a stress field different from that acting on isolated cracks. This yields an effective stress
intensity factor at the crack tips which is utilized in the formulation of the crack growth dynamics. Load-induced damage
in the material is defined in terms of a scalar crack density parameter, the evolution of which is a function of the existing
flaw distribution and the crack growth dynamics. This methodology is applied for the case of uniaxial compression under
constant strain rate loading. The model provides a natural prediction of a peak stress (defined as the compressive strength
of the material) and also of a transition strain rate, beyond which the compressive strength increases dramatically with the
imposed strain rate. The influences of the crack growth dynamics, the initial flaw distribution, and the imposed strain rate
on the constitutive response and the damage evolution are studied. It is shown that different characteristics of the flaw
distribution are dominant at different imposed strain rates: at low rates the spread of the distribution is critical, while at
high strain rates the total flaw density is critical.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Dynamic fracture; Stress concentrations; Micro-cracking; Defect distributions; Ceramic

1. Introduction

Experimental observations (such as recent experiments on a transparent polycrystalline ceramic undergoing

uniaxial compression, Paliwal et al., 2006) have demonstrated that the interaction among the growing micro-
cracks has a profound influence on the failure behavior and macroscopic stress–strain response of brittle
materials, particularly under high strain rate loading. In this paper, we incorporate the crack interactions
through a crack-matrix-effective-medium approach to model the brittle failure process under predominantly
compressive loading.

Corresponding author. Tel.: +1 410 516 7735; fax: +1 410 516 7254.
E-mail address: ramesh@jhu.edu (K.T. Ramesh).

0022-5096/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jmps.2007.06.012
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 897

Brittle materials such as rocks, minerals, ceramics and ceramic composites exhibit a complex non-linear
effective response when subjected to compressive loading. Under quasi-static loading, some of the
characteristic features observed during the compressive brittle failure include (e.g. Eberhardt et al., 1999;
Shao and Rudnicki, 2000):

 Non-linearity in the stress–strain relation and degradation of the elastic constants.

 Hysteresis loops in the stress–strain curves, attributed to the frictional dissipation of energy.
 Load-induced anisotropy associated with the damage evolution.
 Dilatancy and permanent volumetric strain after complete unloading.

The onset of such behaviors is commonly attributed to the presence of intrinsic flaws such as pores,
inclusions and elasticity mismatch at the grain boundaries inside the material. These pre-existing flaws serve as
stress concentrators and locally transform the stress state, leading to the formation of open tension micro-
cracks (‘‘wing cracks’’) even under globally compressive stress fields (Tapponnier and Brace, 1976; Ashby and
Hallam, 1986; Schulson et al., 1991; Vekinis et al., 1991). Under uniaxial compression, these wing cracks grow,
interact, and eventually coalesce, triggering an axial splitting mode. Failure modes related to this type of
behavior have been extensively modeled analytically, based on the sliding crack model proposed by Brace and
Bombolakis (1963) and later analyzed in detail by Nemat-Nasser and Horii (1982). Ashby and Sammis (1990)
approximately analyzed the crack growth from pre-existing flaws and examined the conditions under which an
array of such cracks interacts; the interaction generates an additional tensile field that affects the growth of the
cracks.
In a number of applications including explosive rock blasting and ceramic armor, brittle materials are
subjected to very high loading rates (4103 s1). Under these conditions the effective mechanical response also
depends on the inertial effects related to dynamic crack propagation (Lankford, 1981). It is also possible that
micro-plasticity, in addition to micro-cracking, contributes to the dynamic failure process (Lankford, 1996).
Nemat-Nasser and Deng (1994) extended the quasi-static wing-crack model to the dynamic loading range,
where the effect of the strain rate was included through the dependence of the dynamic stress intensity factor
on the crack growth speed. The interaction effects were modeled by considering an infinite array of wing
cracks under far-field compressive loading. The interaction lead to an effective dynamic stress intensity factor
at the crack tips which is different from that which would develop if the cracks were isolated. Ravichandran
and Subhash (1995) developed a micro-mechanical model for ceramics based on uniformly distributed non-
interacting sliding micro-cracks, and identified important parameters that influence the rate sensitivity and the
compressive strength of the material. Their model assumes that when the micro-crack density reaches a critical
value, the material loses its load bearing capacity; hence the model essentially defines the failure condition
rather than deriving it. Recently, Huang et al. (2002) developed a micro-crack based dynamic damage growth
model applicable to brittle solids under uniaxial compression, at a constant stress-rate loading condition. This
model defines failure (corresponding to the compressive strength) as being reached when the rate of damage
growth D_ ! 1. This was similar in spirit to the model of Grady and Kipp (1980), who assumed a critical
crack density at failure to model the fracture and fragmentation of rocks under impact loading. However, their
analysis also did not explicitly consider the micro-crack interactions.
The micro-mechanical damage models described above capture some of the underlying physical
mechanisms during a brittle failure process, but either assume dilute pre-existing crack distributions with
no interactions or model the micro-crack interaction by assuming a periodic spatial distribution of identical
cracks.
Approximate numerical schemes have been developed to analyze the micro-mechanical interactions for a
given spatial flaw distribution, typical examples of which include the method of pseudo-tractions (Horii and
Nemat-Nasser, 1985; Kachanov, 1987; Basista and Gross, 2000). However, it is not in general feasible to
account for the stress field for each individual micro-crack and account for the interaction with all the other
micro-cracks at large crack densities. In the present work, we develop a methodology based on a complex
variable approach to obtain an approximate local effective stress field as a manifestation of micro-crack
interactions. We account for the existence of a distribution of sizes in the defect population. The pre-existing
defects are assumed to be randomly distributed in space. We analyze a specific case of uniaxial compressive
 ARTICLE IN PRESS
898 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

loading under constant strain rate and investigate the effects of flaw size distribution, imposed strain rate, and
crack-growth dynamics, on the overall constitutive behavior of the material. The outline of the paper is as
follows: Section 2 presents the theoretical framework; in Section 3 a case study is presented for the damage
evolution under uniaxial compression; Section 4 contains numerical results and discussions. There is a
summary in Section 5.

2. Theoretical framework

2.1. Characterizing defects in the material

Consider a representative volume V (area A for the 2-D case) of an elastic brittle material containing
distributed pre-existing flaws. These flaws represent regions such as inhomogeneities, pores, micro-cracks, and
elasticity mismatch at the grain boundaries that cause stress concentrations and provide sites for micro-crack
nucleation. Here we consider the 2-D, plane stress (or plane strain) situation and consider those flaws that can
be characterized as closed rectilinear slits (or closed planar slits) embedded in a homogeneous, unbounded
isotropic elastic medium loaded under overall compression. Two Cartesian coordinate systems will be used
throughout the paper: a fixed global frame of reference (x1, x2, x3) and a local one (x10 , x20 , x30 ), such that the
x20 -axis is collinear with the normal to the slit. The geometry of the medium is generally that of a plate, thin in
the x3-direction (for the plane stress case) or that of a body of infinite extent in the x3-direction (for the plane
strain case). We assume that these flaws are uniformly distributed in space, but follow a certain distribution
wðs; jÞ in size (2s) and orientation (f) with respect to the applied loading. Further, we assume that the size and
orientation distributions are not correlated. Such a distribution of the flaw geometry can be defined by a
density function,
wðs; jÞ ¼ ZgðsÞf ðjÞ, (1)
where Z is the flaw density (m2), and g (s) and f (f) are the p.d.f.s of the half flaw size and orientation,
respectively. Hence the expression wðs; jÞ d s d j represents the number of flaws per unit area A with the half
flaw size and orientation in the ranges (s, s+ds) and (f, f+df), respectively.
For simplicity, further analysis is confined to the elastic solid containing an ensemble of flaws of a specific
orientation (f ðjÞ ¼ dðj  j0 Þ), under plane stress.

2.2. Description of the load induced damage: a sliding flaw model

Under compressive load, the onset of macroscopically inelastic deformation is typically attributed to the
activation of frictional sliding over the faces of these flaws, which causes wing cracks to sprout from their tips
(P, P0 , see Fig. 1(a)).
The wing cracks nucleate, from each tip, at an angle of about y ¼ 70.51, as shown in Fig. 1(a), and follow a
path so as to maximize the mode- I stress intensity factor, KI (for details, see Nemat-Nasser and Horii, 1982;
Ashby and Hallam, 1986). After a short initial curving, these cracks align themselves in the direction of the
maximum principal compression and become rather straight. We approximate these curvilinear wing cracks as
open straight cracks (Fig. 1(b)) as has been suggested by Ashby and Hallam (1986) experiencing local tension
induced by the frictional sliding of the pre-existing closed flaws (Fig. 1(c)), and the material is considered to be
free from damage before the nucleation of these micro-cracks.

2.3. Characterizing damaged material

The local stress fields around every flaw or every flaw-crack system (a flaw with the wing cracks at its tips) in
a real material are sensitive to the spatial distribution and the characteristics of the neighboring flaws. Micro-
crack interactions can cause cracks to be under the influence of the effective stress field different from the
macroscopic stress. Ignoring the crack-interaction effects, wing cracks from all the flaws align themselves in
the direction of the maximum macroscopic principal compressive stress. Consequently, the material becomes
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 899

Fig. 1. (a) Pre-existing closed sliding flaw PP0 and curved open wing cracks PQ and P0 Q0 under far-field compression. (b) Simplified
geometry of the wing cracks as the straight open cracks. (c) Equivalent tension cracks QQ0 with concentrated collinear forces F at the
center.

effectively orthotropic and the planar effective compliance matrix S̄ takes the following form:
0 11 11 1
s11 s22 0
B 0 C
S ¼ @ s22 22
11 s22 A. (2)
0 0 s1212

Under plane stress, the components of S̄ are

1 1 1 n̄12 n̄21
s11
11 ¼ ; s22
22 ¼ ; s12
12 ¼ ; s11
22 ¼  ; s22
11 ¼  with s11 22
22 ¼ s11 ,
Ē 1 Ē 2 2Ḡ 12 Ē 1 Ē 2
where ðĒ 1 ; Ē 2 Þ and ðn̄12 ; n̄21 Þ are respectively, the effective elastic moduli and the effective Poisson’s ratios in the
principal directions and Ḡ12 is the effective shear modulus of the damaged material. As an example, consider a
material under uniaxial compression. As a result of frictional sliding and the opening of axially aligned winged
micro-cracks, the ratio of the inelastic deformation in the lateral direction to that in the loading direction is
greater than its elastic counterpart, so that
 
i  e 
22  22   i þ e  e 
i i      22 22   22 
4
22 411 )  i 4 e  )  i . (3)
11 11  11 þ e11  e11 

Hence, Poisson’s ratio along the axial direction increases (i.e. n̄12 4n). Laboratory investigations,
quantifying the brittle fracture in rocks, have shown that the damage under quasi-static compression causes
an increase in the Poisson’s ratio (Eberhardt et al., 1999). Further, the effective stiffness in the lateral direction
becomes much smaller than in the direction parallel to the micro-cracks (i.e. Ē 2  Ē 1 oE). Nemat-Nasser and
Obata (1988) modeled the inelastic response of the brittle solids neglecting the influence of crack-interactions
 ARTICLE IN PRESS
900 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

and formulated the micro-mechanical constitutive equations. Their work shows that the damage due to axially
aligned wing cracks results in lower inelastic strain in the axial direction which causes the material to become
more compliant in the transverse direction. Further, the inelastic volumetric strain (due to the opening of these
compression-induced tension cracks) becomes positive, causing the material to dilate and making it less
compressible. These are a few of the essential characteristics of the damaged material undergoing compressive
loading with weak interactions among the flaws (also refer to Deng and Nemat-Nasser, 1992).
However, it is well known that under overall compression, the damage caused by micro-cracking tends to
localize into various domains (see Xu et al., 2004, and the experimental observations of Paliwal et al. (2006)).
High-speed photographs of a damaged ceramic undergoing dynamic compressive loading (Paliwal et al., 2006)
shows that the cracked zones contain an ensemble of cracks of various morphologies and sizes which might
not be parallel to each other or aligned in a particular direction. In fact they might orient themselves
differently (and may even cause small-scale crack coalescence to form bigger cracks) as they advance (see
illustrative schematic, Fig. 2). This damage evolution is primarily due to the pre-existing defect distribution
and the interaction among the propagating cracks.
The mechanism of nucleation and growth from the pre-existing flaws is assumed to be that described
previously; however, the micro-crack interactions cause the effective stress field to be different from the
applied macroscopic stress acting on an isolated flaw. Thus, modeling the interaction is paramount and we
address this next.

2.4. Interaction among micro-cracks: a crack-matrix-effective-medium model

First some initial idealizations are necessary. We do not intend to model the localization process and hence
assume that despite the growth of the cracks, the statistical distribution in space remains uniform. We shall
also assume that although frictional sliding of the pre-existing flaws also causes inelastic deformation, the
significant damage is due to the nucleation and growth of open wing cracks under local tension. We also note
that the level of damage-induced anisotropy that develops depends in general on the distribution of the pre-
existing flaws and the interaction among the cracks as the damage progresses. However, at this point it is not
feasible to keep track of the response of individual micro-cracks and re-define the anisotropy at every damage
increment, and so we assume orthotropy.

Fig. 2. Schematic representation of the localized cracked zone in the material under far-field compression. Inset shows numerous cracks
with local tensile splitting forces. Growth mechanism of these cracks (and hence the damage) is modeled as the wing-crack mechanism.
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 901

Unlike the case of non-interacting micro-cracks, the field of interacting micro-cracks constituting the
damage is not necessarily aligned along one of the macroscopic principal stress directions. To simplify the
problem, therefore, we allow the values of the effective secant moduli of the damage material in both the
principal directions to be the same, i.e. ðE 2 ¼ E 1 ¼ EoEÞ. We further exploit the self-consistent scheme and
employ the approach by Budiansky and O’Connell (1976), as an approximation, to obtain effective moduli of
the damaged material. Budiansky and O’Connell obtained the relation between the crack density parameter
and the effective elastic constants of a material with many randomly distributed open micro-cracks (in our
model, open wing cracks with traction-free surfaces). Hence, a relation between the damage, defined in the
form of 2-D crack density parameter, O  nl 2 ; and the effective Young’s modulus, Ē and effective shear
modulus, Ḡ12 can be written as

p2
Ē=E ¼ 1  ð1 þ n̄Þð5  4n̄ÞO,
30
p2
Ḡ 12 =G ¼ Ḡ=G ¼ 1  ð10  7n̄ÞO, ð4Þ
60
where n is the areal flaw density (m2), l is the wing-crack length, and
E and G are the Young’s modulus and the shear modulus, respectively, of the pristine material.
Recently, Wang and Ramesh (2003) conducted experiments on an armor ceramic (SiC-N), undergoing
dynamic uniaxial compression. By means of image analysis of high-speed photographs of the specimen
undergoing deformation, they demonstrated that it dilates significantly in the transverse direction, while
undergoing axial compression. This suggests that the Poisson’s ratio of the material increases as the damage
progresses under macroscopic compression. However, significant transverse dilation was not observed in
AlON specimens undergoing dynamic compression (Paliwal et al., 2006), and hence the associated effect on
the Poisson’s ratio is difficult to state for this material. There is in general not sufficient experimental data on
transverse dilatancy in brittle materials under dynamic loading to determine the Poisson’s ratio as a function
of damage, but the classical expectation is that it will increase. However, in the Budiansky and O’Connell
formulation, the effective Poisson’s ratio decreases with the increase in the crack density parameter. As a
compromise and given insufficient data, we assume it to stay constant, i.e. ðn12 ¼ n21 ¼ n ¼ nÞ. However, if an
appropriate relation between the effective Poisson’s ratio and the damage is available, it can easily be
incorporated in the present framework.
Consider a uniform spatial distribution of closed flaws in an infinite homogeneous, isotropic linear-elastic
continuum. Some of these flaws are activated as a result of the load-induced damage, and have straight open
‘‘wing’’ micro-cracks at their tips, as shown in the schematic, Fig. 3. Interactions among these micro-cracks
will generally cause the effective stress field around them to be different from that if they were isolated. In this
section, we shall model the effect of the micro-crack interaction by means of the effective stress field around a
crack and hence obtain an effective mode-I stress intensity factor K eff I at the crack tips.
Assume that a flaw with two wing cracks at its tips is embedded in an elliptical matrix of pristine material
(S), which in turn is embedded in an effective medium with the effective properties of the micro-cracked solid
ðSÞ. For the damage free material (i.e. crack density O ¼ 0), the elastic properties of the medium and the
matrix are identical and therefore, all the flaws are isolated. Continued loading causes the weaker flaws to
activate first, initiating the damage process, and the properties of the surrounding medium begin to differ from
that of the matrix. Due to the difference in the properties of the medium and the matrix, the crack inside the
inclusion responds to a stress field different from the macroscopic applied stress on an isolated crack.
The major axis 2l1 of the elliptical matrix is aligned in the direction of the straight open wing crack and the
area of the surrounding medium is chosen such that at the max crack density Omax (the maximum value that
can be used if the SCM is employed to evaluate the effective properties is around 0.51), the flaw and the two
wing cracks at the tips is completely surrounded by the matrix. For example, consider wing cracks of length
1
In practice, threshold stress is realized at relatively smaller crack density. Modeling results from Grady and Kipp (1980) and recently
Huang and Subhash (2003) suggest that the material loses its load-bearing capacity at around 12% crack density or when the damage
growth rate dD=dt ! 1 (that occur at around 12% crack density, see Huang and Subhash, 2003). Experimental observations on
transparent AlON by Paliwal and Ramesh (2006) have also suggested that material failed at around 11% crack density.
 ARTICLE IN PRESS
902 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Fig. 3. Schematic illustrating the effect of the crack interaction. Inset shows that the interaction can cause the effective stress field around a
crack to be different from that of isolated cracks.

Fig. 4. Schematic of the crack-matrix-effective-medium model. Flaw with the wing cracks at the tips is surrounded by an elliptical matrix
of pristine material (secant compliance S), which is embedded in an effective medium (secant compliance S̄).

lmax at the tips of the flaw of length 2s in an elliptical matrix, as shown in the schematic (Fig. 4). Following
Huang et al. (1994), for lengths of major and minor axes given by
l 1 ¼ a þ d; l2 ¼ d (5)
we can relate the area of the matrix with the maximum crack density as
 2
a2 a2 d d 1
Omax ¼ ¼ or þ ¼ . (6)
pl 1 l 2 pða þ dÞd a a pOmax
We now evaluate the effective stress and the effective mode-I stress intensity factor KI at the crack tips.
The medium is subjected to a remote compressive biaxial loading,
0 1
s1
B C
P ¼ @ s2 A (7)
0
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 903

and the resulting tractions acting on the flaw and the wing-crack surfaces are
( 0
0
t s  traction on the closed sliding flaw surfaces
t
t0 w  traction on the open wingcrack surfaces ð¼ 0:0Þ

as shown in Fig. 5. The mode-I stress intensity factor at the wing-crack tips is calculated using the
superposition principle described as follows. The problem of a cracked inclusion in an effective medium
subjected to remote uniform loading can be decomposed into two problems as shown in Fig. 6. In the first
configuration, see Fig. 6(a), a homogeneous elliptical inclusion of the matrix material (without the crack) is
embedded in the effective medium which is subjected to the same remote loading as in the original problem.
The tractions tr are the tractions on the planes corresponding to the cracks. In the second configuration,
shown in Fig. 6(b), only the crack surfaces are subjected to tractions teff that are derived from the first problem
such that they satisfy the following traction boundary condition

t ¼ tr þ teff . (8)

Now consider the first configuration for the case of uniform remote biaxial loading P as defined in Eq. (7).
According to the Eshelby (1957) theory the stress and the strain field in this elliptical inclusion embedded in an
otherwise homogeneous infinite medium are uniform for a uniform applied far-field stress.

Fig. 5. Schematic shows the effective medium, containing a cracked elliptical inclusion, under far-field compression P. t0 represents the
tractions on the flaw-crack surfaces.

Fig. 6. Decomposition of the problem posed in Fig. 5 into two components by using the superposition principle.
 ARTICLE IN PRESS
904 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

We use the method of complex potentials to solve for the homogeneous stress field Pe inside this elliptical
inclusion (for details, see the Appendix A) in the form of the expression,
0 e 1
s11
B e C
Pe ¼ @ s22 A, (9)
se12

where se11 ; se22 and se12 are obtained in terms of the applied loading P. The tractions due to Pe on the planes at
the imaginary location of the cracks are defined as
( r
ts  tractions on the planes corresponding to the closed sliding flaw
r
t
trw  tractions on the planes corresponding to the straight open wing crack

Each of these traction vectors has components normal and parallel to the respective planes. Hence, at the
closed sliding crack location,
( rk
ðts Þ ¼ 12ðse11  se22 Þ sin 2j þ se12 cos 2j;
r
ts (10)
ðtrs Þ? ¼ se11 cos2 j þ se22 sin2 j þ se12 sin 2j

and at the straight open wing-crack location,

(
r
ðtrw Þk ¼ se12 ;
tw (11)
ðtrw Þ? ¼ se22 ;

where the superscripts ‘‘||’’ and ‘‘?’’ indicates the directions parallel and perpendicular to the surfaces,
respectively.
The tractions t acting on the flaw-crack surfaces (see Fig. 5) can be expressed in terms of these resolved
tractions tr . If the frictional contact on the closed sliding flaw surfaces is of the Mohr–Coulomb type, we can
write an expression for the tractions t0s acting on the sliding flaw surfaces, as
(
0
ðt0 s Þ? ¼ ðtrs Þ? ;
ts (12)
ðt0 s Þk ¼ tc  mðtrs Þ? ;

where m is the coefficient of dry friction for the closed cracks and tc is the cohesion.
Now consider the second configuration as defined in Fig. 6(b). Tractions teff , which satisfies the boundary
condition provided in Eq. (8), are the effective tractions acting on the flaw-crack surfaces which drive the
inelastic deformation.
( eff
eff
ts  effective traction that drives the frictional slip
t ¼ t0  tr .
teff
w  effective traction that drives the wingcrack opening

Hence, using Eq. (10) and (12), for the sliding crack surfaces
8  k
< ðteff
s
k
Þk ¼ t0 s  ðtrs Þk ¼ tc  mðtr? Þs  ðtr Þs ¼ teff ;
eff
ts  0 ? (13)
: ðteff ?
s Þ ¼ ts  ðtrs Þ? ¼ 0

and using Eq. (11), for the wing-crack surfaces (with t0 w ¼ 0)

( k r k
eff
ðteff
w Þ ¼ ðtw Þ ;
tw ? r ? (14)
ðteff
w Þ ¼ ðtw Þ :

In order to evaluate the mode-I stress intensity factor, KI, we have used the results of Horii and Nemat-
Nasser (1986), since they provide good estimates of the SIF for the short and the long wing cracks alike.
Accordingly, the tensile stress intensity factor, ðK I Þ1 , at the crack tips arises due to the wedging effect caused
by the sliding action of the closed flaw, which can be represented by two collinear splitting forces
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 905

k e
F ¼ 2sðteff
s Þ as shown in Fig. 1, and the SIF also arises due to the stress state P , denoted by ðK I Þ2 . Hence KI
at the wing-crack tips can be expressed as
K I ¼ ðK I Þ1 þ ðK I Þ2
2sðteff k
s Þ cos j
pffiffiffiffiffi
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

 ðteff
w Þ
?
pl , ð15Þ
pðl þ l Þ
where l  ¼ 0:27s was introduce to render KI non-singular at l ¼ 0.
Using Eqs. (13)–(15) we can arrive at the following expression:
2steff cos j pffiffiffiffiffi
K I ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ se22 pl ,
pðl þ 0:27sÞ
where
 
teff ¼ tc  m se11 cos2 j þ se22 sin2 j þ se12 sin 2j
 
þ 12ðse11  se22 Þ sin 2j  se12 cos 2j ð16Þ

3. Specific case of uniaxial compression at constant strain rate

We illustrate an application of this methodology to study the damage evolution and overall constitutive
response (including the post-peak softening response) under strain-controlled uniaxial compressive loading.
Later work will include examination of the effects of confinement.
Consider an elastic brittle continuum, with a spatially uniform pre-existing flaw distribution as defined in
Eq. (1). We assume a flaw density Z with a size distribution g(s) but all of these flaws are oriented at an angle
f0 given by
 
p 1 1
j0 ¼  tan , (17)
2 m
which is the most favorable orientation of the closed sliding flaws for the nucleation of the wing cracks (Ashby
and Hallam, 1986). The distribution then becomes
wðs; jÞ ¼ ZgðsÞdðj  j0 Þ. (18)
We discretize this distribution into N families of flaws, labeled by the subscript j ¼ 1, 2, y, N. Each family
is characterized by its size 2sj and its spatial orientation f0, so that its areal density nj can be expressed as
nj ¼ Zgðsj ÞDsj , (19)
where Dsj is the increment in the half flaw size such that sjþ1 ¼ sj þ Dsj .
Now, we subject the material to a uniaxial compressive loading under a constant strain rate deformation.
For this 1-D stress problem, the constitutive equation can be written in terms of the reduced modulus Ē (Eq.
(4)) as
s ¼ Ē
) _
s_ ¼ Ē_ þ Ē
 
p2 p2 _
¼ E 1  ð1 þ n̄Þð5  4n̄ÞO _  E ð1 þ n̄Þð5  4n̄ÞO, ð20Þ
30 30
where (  ) means derivative with respect to time.
The total crack density parameter O is
X
N X
N X
N
O ¼ O ð l 1 ; l 2 ; . . . ; l N ; n1 ; n2 ; . . . ; nN Þ ¼ Oj ¼ nj l 2j ¼ Z gðsj ÞDsj l 2j , (21)
j¼1 j¼1 j¼1

where 2lj is the wing-crack length associated with each family of flaws (with flaw size 2sj).
 ARTICLE IN PRESS
906 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Hence the evolution of the crack density parameter is given by

XN  
_ ¼ qOj qOj _
O n_ j þ lj . (22)
j¼1
qnj ql j

We further assume that the number of flaws in each family does not change, i.e. no new flaws are created
and no existing flaws are allowed to vanish. We emphasize here the distinction between the number of flaws in
the material and the number of cracks. In the present work, the number of defects (flaws) in the material
remains fixed (i.e. does not evolve with deformation), but the number of cracks in the material does change.
Although it is certainly possible that additional defects (or classes of defects) become potential crack
nucleation sites under any given loading, we have chosen to neglect the evolution of the initial defect
distribution so as to focus the discussion on the effect of the distribution itself and the interaction with the
crack growth dynamics. The potential evolution of the initial defect distribution is an additional complexity
that we leave for another work (which will have to incorporate a clear sense of the microscopic mechanisms
associated with the specific defects considered). Thus, n_ j ¼ 0, and so
X
N
_ ¼ 2Z
)O gðsj ÞDsj l j l_j . (23)
j¼1

We thus need an evolution law for the crack length lj. Wing cracks are initiated from the existing flaws when
the SIF KI at their tip reaches the mode-I fracture initiation toughness of the material. In the present study we
assume the mode-I FIT to be rate independent and equal to the mode-I fracture toughness : KIC. The subsequent
crack growth dynamics is determined from a crack growth law relating the crack speedl to the applied loading.
Following Freund (1972, 1990), the mode-I dynamic stress intensity factor of a growing crack, with an
_
instantaneous speed lðtÞ, can be related to the static SIF of an equivalent stationary crack at the same
instantaneous length lðtÞ by,
_ ¼ kðlÞK
K dI ðl; lÞ _ I ðlÞ, (24)
where KI can be obtained from Eq. (16). The function kðlÞ_ is a universal function of the crack speed, which
represents the inertial effect on the crack growth. Deng and Nemat-Nasser (1992) obtained a simplified
approximate expression for kðlÞ _ which was reported to be in good accord with the numerically obtained
solution:
! !1
l_ _
l
_ ¼ 1
kðlÞ 1 , (25)
CR 2C R

with CR being the Rayleigh wave speed of the material. We assume that the dynamic mode-I SIF K dI cannot
exceed a critical value called the dynamic fracture toughness K dIC , a material property which may also depend
on the loading rate at the crack tip, crack tip speed, etc. and should be determined by experiments (e.g.
Rosakis and Ravichandran, 2000; Kobayashi, 2002). Work conducted by various groups including on various
ceramics by Suresh et al. (1990) and recently by Weerasooriya et al. (2006) have suggested that under dynamic
loading these materials show higher fracture toughness than under quasi-static loading. However, since many
aspects of the material behavior and the underlying mechanisms are still unclear and since determination of
the dynamic fracture toughness remains a challenge, we have assumed the toughness to stay constant (i.e.
K dIC ¼ K IC ).
Hence, the dynamic crack growth criterion can be written as

K dI ¼ K IC (26)
Combining Eqs. (24) and (25), we can express the crack growth as
K I  K IC
l_ ¼ C R . (27)
K I  K IC =2
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 907

_ as described by
This expression for the crack speed is determined with the help of the universal function kðlÞ
_
Eq. (25). This functional form of kðlÞ was obtained by solving a self-similar dynamic crack growth problem
under far-field uniform tension (see Freund, 1990). This phenomenon may be applicable for the case of
compressive loading by recognizing, as discussed earlier, that the frictional sliding of the pre-existing flaws
transforms global compressive field into a population of local tensile enclaves at the wing-crack tips. Hence, it
is assumed that a similar form of kðlÞ _ exists for this case relating the wing-crack growth rate to the applied
loading. We express the crack growth in the following form,
 g
K I  K IC
l_ ¼ cmax , (28)
K I  K IC =2

where cmax is defined as the maximum (terminal) speed of a dynamically propagating crack (henceforth written
as cmax ¼ C R =a, where aX1), and g is a fitting parameter characterizing the toughness-velocity relation. For
(a, g) ¼ (1, 1), this equation is reduced to Eq. (27) described above. Similar parameters are used by
Ravichandran and Subhash (1995) in their crack growth relation to define the universal function of crack
speed. This, form is particularly useful to fit the experimental data.
Thus, the governing equations for the specific case of uniaxial compression are:

Constitutive law:
 
p2 p2 _
s_ ¼ E 1  ð1 þ n̄Þð5  4n̄ÞO _  E ð1 þ n̄Þð5  4n̄ÞO. (29)
30 30

Distributed damage evolution:

X
N
_ ¼ 2Z
O gðsj ÞDsj l j l_j . (30)
j¼1

Crack growth dynamics:

Crack growth criterion:
K dI ¼ K IC . (31)

Crack growth law:

 
K I  K IC g
l_ ¼ cmax .
K I  K IC =2

Micro-mechanics:
2steff cos j pffiffiffiffiffi
K I ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi þ se22 pl ,
pðl þ 0:27sÞ
where;
 
teff ¼ tc  m se11 cos2 j þ se22 sin2 j þ se12 sin 2j
 
1 e e e
þ ðs  s22 Þ sin 2j  s12 cos 2j ð32Þ
2 11
and with se11 ; se22 and se12 obtained from macroscopic stress s using the analysis presented in the
Appendix A.
 ARTICLE IN PRESS
908 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

4. Numerical results and discussion

We have used Euler’s forward integration method to solve Eqs. (29)–(31) with the initial conditions,
lðtÞjt¼0 ; OðtÞjt¼0 ¼ 0 at a constant strain rate.
We examine the effects of the strain rate, crack growth dynamics, flaw distribution and friction coefficient
on the compressive strength, the stress–strain response and the induced damage in brittle materials. Unless
otherwise stated, the material constants chosen are those listed in Table 1. These nominal values of the
material parameters are similar to those of most engineering ceramics.

4.1. Effects of the imposed strain rate

We have performed numerical simulations for imposed strain rates ranging from 103 to 106 s1. Material
parameters and flaw statistics are listed in Table 1. The results are presented in Fig. 7 as a series of stress vs.

Table 1
Material properties

E n r KIC m tc a, g

300 GPa 0.24 3673 kg/m3 2.7 MPaOm 0.2 0 1, 1

Nominal parameters for the flaw distribution

g(s) Z Mean F

d(ss0) 107 m2 2s0 ¼ 50 mm F0 ¼ 50.71

Fig. 7. Plot shows the effect of the strain rate on the constitutive response of the brittle material loaded under uniaxial compression. Note
that (i) the peak stress becomes increasingly sensitive to the strain rate after about 104 s1, (ii) as the strain rate becomes smaller, post-peak
softening response and the damage growth with strain becomes faster, (iii) at higher strain rate, material can sustain more damage before
the collapse (indicated by the markers on the damage-time plots). The inset shows the variation of stress and damage with time. Note that
the damage growth rate increases with the increase in strain rate.
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 909

strain curves and associated damage vs. strain curves (dotted lines), where the stress and strain under
compression are considered positive. In each case the strain controlled loading shows the stages of linear
elasticity followed by non-linear hardening up to a maximum stress (called the compressive strength) and
finally, strain softening. For these parameters, the model prediction of the compressive strength is relatively
insensitive to strain rate until a transition strain rate of about 104 s1, after which the strength increases
dramatically with the imposed strain rate. This result is consistent with the experimentally observed strain rate
sensitivity of the compressive strength of various ceramics, e.g. Lankford (1996) and brittle rocks, e.g. Huang
and Subhash (2003). It is also interesting (Fig. 7) that the post-peak softening response is less dramatic at
higher strain rates. Similarly the damage (crack density parameter O) evolves more slowly with the strain at
high strain rates. Such behavior is to be expected in a brittle material subjected to high loading rates when the
material inertia becomes increasingly important. In this study, inertial effects in the constitutive response arise
as a result of the crack growth dynamics, which provide an upper bound to the crack speed (here the terminal
crack speed cmax is considered to be equal to CR). Fig. 8 shows the plots of the crack speed ðlÞ _ normalized by
the crack speed at the peak stress (vpeak) against (KI) normalized by ðK IC Þ, at different strain rates. The post-
peak (when l4v _ peak ) softening response of the material is primarily controlled by the nature of unstable
damage growth (relationship of O _ vs. the crack growth dynamics, Eqs. (30), (31)) which under dynamic
loading conditions strongly depends on the post-peak change of the crack speed and on the terminal crack
propagation speed, cmax. Note that vpeak increases with the increase in strain rate as is indicated on each curve.
The crack growth dynamics forces the crack speed to evolve more slowly at higher strain rates (due to
bounded terminal crack speed). This is what causes the relatively slow damage evolution with strain and slow
post-peak softening response at higher strain rates.
The level of damage associated with the peak stress is identified in Fig. 7 with a colored circle for each
applied strain rate. The value of the damage corresponding to the peak stress is relatively low (0.1–0.2), which is
consistent with the experimental observations of Paliwal et al. (2006). Further, this ‘‘peak’’ damage value
increases with the strain rate. This implies that the material can sustain a higher level of damage before
collapse begins at higher strain rates. The inset in Fig. 7 shows the variation of the stress and the damage with
time (rather than strain). This demonstrates that the damage does grow more rapidly with time at higher strain
rates as is reported by several other researchers. Clearly the damage growth rate and total damage are
significantly greater at higher strain rates. The computed damage growth rate at the time of the peak stress
increases with the applied strain rate (varying from 4  105 to 1.1  107 s1). We note that some models in the
literature assume a critical crack density at the peak stress, while others assume a critical damage growth rate.
Although in this study we have not made an attempt to relate the level of damage to the statistics of
fragmentation, the increased micro-crack density at the peak stress predicted by the model at higher strain
rates suggests that more (and finer) fragments may develop at high rates than at low rates. Various
fragmentation models (see for, e.g. Grady, 1982; Zhou et al., 2006) and also experimental investigations on

_ peak ) and the post-peak (when l4v

Fig. 8. Plots illustrates the pre-peak (i.e. when the crack speed lov _ peak ) normalized crack speed for
different strain rate. Note that the post-peak softening response (in the stress–strain curve in Fig. 7) is slower as strain rate increases
because the change in the normalized post-peak crack speed decreases with the increase in the strain rate.
 ARTICLE IN PRESS
910 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

various brittle materials (see for, e.g. Wang and Ramesh, 2003 on SiC-N) have reported that under dynamic
load, extensive micro-cracking in the material leads to fragmentation with finer fragment sizes than at lower
strain rates.

4.2. Effects of crack growth dynamics

The expression for the crack growth law in Eq. (28) contains two parameters, a and g. This
phenomenological form of the crack growth law is particularly useful to fit the limited experimental data
on crack speeds in these materials. Note that the crack growth rate is a monotonically decreasing function of
both these parameters.
In Fig. 9 the compressive strength (the maximum of the computed stress–strain curve) is plotted as a
function of the strain rate for values of a ranging from 1 to 5 with g ¼ 1 (Fig. 9(a)) and g ranging from 1 to 5
with a ¼ 1 (Fig. 9(b)). Markers A, B, C and P, Q, R refer to the compressive strength at a strain rate of 105 s1
in Fig. 9(a) and (b), respectively.
Consider the case where a ¼ g ¼ 1. The plot shows that the peak stress is insensitive to the imposed strain
rate up to around 104 s1 after which the transition to the strong rate sensitive behavior occurs. This material-
dependent critical strain rate is defined as the transition strain rate (see Lankford, 1996). It is clear from plots
in Fig. 9(a) and (b) that both the transition strain rate and the dynamic rate sensitivity (material behavior
beyond the transition strain rate) are functions of the parameters in the crack growth dynamics.

Fig. 9. Plots (a) and (b) show the individual effects of the two parameters of the growth law on the compressive strength vs. strain rate
relation. Parameter a has a stronger effect on the dynamic strain rate sensitivity (as indicated by the arrow in plot (a)), whereas parameter g
has a profound influences on the transition strain rate (shaded region in plot (b)) of the material. Plots (c) and (d) shows the plots of crack
speed l_ (normalized with the CR) against KI (normalized with the KIC) for varying a with g ¼ 1, and for varying g with a ¼ 1, respectively.
Circular markers on the plots indicate the crack speed at the peak stress for the imposed strain rate of 105 s1.
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 911

In the dynamic regime the strain rate sensitivity of a material depends on the inertial effects related to the
_ R)
dynamic crack propagation. This can be illustrated by considering (Fig. 9(c)) the variation of crack speed (l=C
against (KI/KIC) for g ¼ 1 and a ¼ 1, 2, 5 (essentially this is a plot of the assumed crack growth dynamics). The
circular markers A, B and C on these curves indicate the crack speed at the peak stress for 105 s1 strain rate
loading (corresponding to the markers A, B and C in Fig. 9(a)). Fig. 9(c) shows that the crack grows at 65% of
its terminal speed for a ¼ 5 as compared to 27% of terminal speed for a ¼ 1. For any one of these assumed
dynamics (Fig. 9(c)), higher strain rates will correspond to moving the marker (e.g. A) further to the right (i.e.,
higher crack speeds) at the time of the peak stress. Clearly, the change in (KI/KIC) required to obtain an increase
in crack speed is much higher for a ¼ 5 than for a ¼ 1, so that higher compressive strengths are developed for
a ¼ 5, resulting in higher rate sensitivity. Thus the crack growth dynamics directly determines the rate-sensitivity
of the compressive strength, and higher a gives a higher rate-sensitivity.
Similarly, the influence of g on the rate dependence of the compressive strength can be illustrated through
Fig. 9(d), which shows the crack growth dynamics curves obtained with fixed a and varying g. Certainly
changing g also results in an increased rate sensitivity at high rates, since the curves all rise with increasing
KI/KIC. However, the characteristic influence of g is best seen through the shape of the curves with increasing
gU Examining the case of g ¼ 1, we observe that near KI/KIC ¼ 1 one can move to the right (increasing crack
speed) with small change in KI/KIC as compared to the case of g ¼ 5 where the change in KI/KIC is much
higher. This indicates that the effect on the compressive strength is smaller as one increases the strain rate for
smaller values of g.
Thus, there is little rate dependence until much higher strain rates are achieved in the g ¼ 1 case as
compared to the case where g ¼ 5. Hence, the value of the parameter g is effective at capturing transition rate
behavior (see the shaded region in Fig. 9(b)).
These results can be summarized as follows:

 A material with higher g in the crack growth dynamics becomes strongly rate sensitive at lower strain rates.
 Materials with the same terminal crack speed (i.e. same a) but different values of g shows similar strain rate
sensitivity beyond their respective transition strain rates.
 As a increases, the rate sensitivity of the compressive strength of the material also increases, but with only a
weak influence on the transition strain rate.

These observations related to the dynamic effects are important characteristics of the brittle failure mechanism
under high strain rate loading. Taking it further, we now look at the effect of the two parameters on post-peak
stress–strain response of the material under dynamic loading (i.e. beyond the transition strain rate). For
illustration, Fig. 10 shows a series of stress–strain plots at 105 s1 strain rate loading for three sets of a and g
denoted by (a, g) ¼ (1, 1), (5, 1) and (1, 5). This strain rate is greater than the transition strain rate for either value
of the parameters (Fig. 9). Note that material with parameters (5, 1) is highly rate sensitive at this strain rate as
compared to either (1, 5) or (1, 1). Now, comparing their post-peak strain softening response plots, it is interesting
to note that g does not really influence the softening characteristics (see plots for (1, 1) and (1, 5)), whereas higher
a causes delayed softening with strain (plots for (1, 1) and (5, 1)). As in the discussion of Fig. 9, this is best
understood in terms of the change in the crack speed with (KI/KIC) shown in Fig. 11 (note, however, that the
crack speed ðlÞ_ is normalized by the crack speed at the peak stress, vpeak). The post-peak softening is controlled by
the damage growth rate, which is determined by the crack growth rate. For (a, g) ¼ (5, 1), when the peak stress is
reached the crack speed has already reached 65% of the maximum crack speed (see Fig. 11(a)), and for further
softening a substantial increase in KI is needed to develop the additional damage, so that the stress drops more
slowly. This is less of an issue when a ¼ 1. In contrast (Fig. 11(b)), g does not appreciably influence the post-peak
change of (KI/KIC) with the change in the normalized crack speed. Thus, the strain softening response in the
stress–strain curve (Fig. 10) is only weakly affected.

4.3. Effects of the flaw distributions

In this section we illustrate the effect of the flaw size distribution on the compressive strength and on the
induced damage as a function of the imposed strain rate.
 ARTICLE IN PRESS
912 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Fig. 10. Plot shows the effect of the crack growth law parameters (a and g) on the stress–strain response for _ ¼ 105 s1 . Note that in
addition to the compressive strength, the post-peak softening response is also affected by these parameters.

_ peak ) and the post-peak (when l4v

Fig. 11. Plots illustrates the pre-peak (i.e. when the crack speed lov _ peak ) normalized crack speed for
different (a) a, and for different (b) g, for the imposed strain rate of 105 s1.

Fig. 12(a1) shows a Gaussian and two Weibull probability density functions all of which have the same
mean (2s0 ¼ 50 mm) and standard deviation (s ¼ 5 mm) representing different flaw size distributions. The two
Weibull distributions have different skew parameters, G ¼ 1.2 and 2.4. Note that unlike the Gaussian
distribution, Weibull distributions have a lower limit on the flaw size. For convenience in the numerical
analysis, we assume the lower and the upper limit to be smin ¼ s0  5s and smax ¼ s0 þ 5s, respectively, where
the probability of finding flaws in the range (sosmin and s4smax) is negligible.
The effect of the distribution (Fig. 12(a2)) is shown in the figure below in terms of stress–strain plots for two
strain rates, 1 and 105 s1. Note that the stress–strain response does not seem to differ significantly, with the
change of the distribution function. However, the material with the Weibull distribution of higher asymmetry
(G ¼ 1.2) seems to have lower compressive strength at lower strain rate. The reason appears to be that at lower
strain rates, most of the damage is due to the bigger flaws in the distribution (discussed in detail later in this
section). The low-end cut-off for the Weibull law with higher asymmetry (around 19 mm) is greater than that
for the other two laws and hence a greater number of flaws are engaged in the damage process (causing the
critical damage) at an early stage of loading. As a result, we expect lower compressive strength for such
distributions compared with the less skewed ones. On the other hand, at higher strain rates smaller flaws also
cause significant damage, and thus the stress–strain curves do not seem to be as different.
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 913

Fig. 12. Shows the p.d.f. (plots (a1) and (b1)) of the half flaw size (with the mean flaw size 2s0 ¼ 50 mm) and the corresponding
stress–strain curves (plots (a2) and (b2), respectively). Two strain rates are considered here, 1 and 105 s1. Plot (a) effect of different
distribution functions for same variance and (b) shows the effect of the standard deviation, s, on the stress–strain response of the material.

Now consider two Weibull distributions as shown in Fig. 12(b1) with identical mean (2s0 ¼ 50 mm) and
skew parameter (G ¼ 1.2) but different standard deviations. The wider distribution (solid line) has s ¼ 5 mm
and the narrower one (dashed line) has s ¼ 0.5 mm. The stress–strain plots for these two half flaw size
distribution functions and for two strain rates, 1 and 105 s1 are shown alongside the p.d.f.s in Fig. 12(b2). The
material with the narrower distribution function shows higher strength for both strain rates. However, the two
responses are closer to each other at the higher strain rate. This can be understood using the same concept as
with the skew parameter. Although the narrower distribution function has a higher cut-off for smin, it has an
extremely low probability of having bigger flaws as compared to the wider distribution. As a result, damage
initiates at relatively early stages of loading in the wider distribution due to the presence of bigger flaws. This
causes the interactions among the cracks to occur earlier as well and hence more flaws, even at the lower end of
the distribution, are activated. This also causes a lower compressive strength in a material with a wider
distribution of flaw size. However, the effect is lower at higher strain rates because large numbers of flaws are
activated within smaller amount of time and hence the stress–strain curves are close to each other for either
distribution. An important point to note here is that this case is different from the previous one where the
distributions differ in terms of the asymmetry w.r.t. the mean value. This asymmetry factor does not have
much of an effect on the tail end of the distribution as compared with the distributions plotted in Fig. 12(b1)
but it does have significant effect on the lower cut-off (i.e. smin).
It is interesting to see the distribution of this evolving damage over the pre-existing flaws. Fig. 13(a) and (b)
shows the surface plots of the damage evolution and its distribution over the existing flaws for 1 and 105 s1,
respectively. The z-axis represents the wing-crack length and x- and y-axis represents the strain and half flaw
 ARTICLE IN PRESS
914 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Fig. 13. Surface plot shows the evolution of crack length, w.r.t the applied strain, for a Gaussian distribution of initial flaw size with mean
2s0 ¼ 50 mm and standard deviation s ¼ 5 mm. Two strain rates are considered: (a) 1 s1 and (b) 105 s1. Calculations are performed till the
peak stress is reached in both the cases. Note that the damage caused by the bigger flaws is more and its spread over the population of
flaws is less uniform, at lower strain rate.

sizes, respectively. The surface shows the evolution of the wing-crack length (representation of the damage)
from the existing flaw families of different size with strain, till the peak stress is reached in the material.
Simulations are conducted for the Gaussian distribution of the flaw size with 2s0 ¼ 50 mm and s ¼ 5 mm. Note
that the strain rate of 1 s1 is before the transition strain rate region (see Fig. 9). Therefore, as discussed in the
previous section, the damage (shown in the figure) evolves slowly with strain at higher strain rate.
It is interesting to see that the wing-crack length distribution is more uniform over the existing flaws at
105 s1 strain rate as compared to 1 s1 strain rate. The reason appears to be that at high strain rate loading
large stresses are achieved in a very short time as compared to the low strain rate loading. Thus, large numbers
of flaws are activated in a short time and this contributes to the damage and causes it to be distributed more
uniformly. The surface plots also show that the maximum wing-crack lengths from the bigger flaws is around
800 mm at lower strain rate, as compared to the maximum length of about 300 mm at higher strain rate (the
‘‘peak’’ crack density is 10% and 12% for strain rate of 1 and 105 s1, respectively). Thus, at a given damage
level the lower strain rate loading corresponds to the nucleation and growth of bigger flaws as compared with
the higher strain rate loading. At lower strain rates, the damage caused by the bigger flaws is more significant
and the smaller flaws remain relatively dormant.
These quantitative aspects of the damage have also been observed in experiments. Our work (Paliwal et al.,
2006) on transparent polycrystalline ceramic AlON undergoing uniaxial compression indicated that during a
dynamic test many more micro-cracks are nucleated forming a damage zone of high crack density, whereas
under quasi-static loading damage development is due to the formation of a very few macro-cracks. Sarva and
Nemat-Nasser, 2001 conducted uniaxial compression experiments on hot-pressed sintered SiC and observed
that the fragment size under dynamic loading is considerably lower than under quasi-static loading. Similar
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 915

observations on hot-pressed SiC-N ceramic were reported by Wang and Ramesh (2003). They suggested that
many micro-cracks are nucleated during a dynamic test as compared to the quasi-static test and the length of
the cracks is much smaller as is manifested by smaller fragment sizes.
The effect of the flaw density parameter Z for a Gaussian distribution of flaw size (2s0 ¼ 50 mm and
s ¼ 5 mm) is shown in Fig. 14 in terms of the variations of the strain rate. Values of Z range from 105 to
pffiffiffi
108 m2 (that correspond to the normalized mean separation of sites, ð12 s0 ZÞ, ranging from 2 to 64),
representing dilute and concentrated flaw distributions, respectively. An increase in the flaw density leads to
lower compressive strength for all strain rates, but it is interesting to note that the material with lower Z
becomes strongly rate sensitive at higher strain rates so that the reduction of strength with increasing Z
increases with the increase in strain rates. In other words, it seems that the decrease in flaw density is
analogous to the dynamic effects associated with the higher strain rates. This result can be explained as
follows: as Z decreases the mean spacing between the flaws increases (i.e. the cracks are farther apart), which
implies that the wing-crack length associated with the flaws has to be larger in order to develop the effective
crack interactions causing the stress collapse. Hence the crack growth dynamics becomes more important at
lower flaw densities and the transition to the dynamic rate sensitivity occurs at lower strain rates. On the other
hand, higher flaw densities cause the interactions among the cracks to dominate at lower wing-crack lengths
(i.e. at relatively early stages of crack growth) and hence the transition occurs at higher strain rates. Similar
observations were also reported by Nemat-Nasser and Deng (1994) where the transitional strain rate decreases
with increasing micro-flaw spacing (i.e. decreasing flaw density).
It is also interesting to look at the distribution of the damage over the pre-existing flaws for different Z.
Fig. 15(a) and (b) shows the surface plots of the evolution of the wing-crack length and its distribution over
the existing flaws for Z ¼ 108 and 105 m2, respectively, under strain rate of 105 s1. The z-axis represents the
wing-crack length and x- and y-axis represents the strain and half flaw sizes, respectively. It is evident from the
plot that the distribution of the wing-crack length over the existing flaws is more uniform for smaller Z.
Arrows indicate the wing-crack length attained at the peak stress, from the flaws of various sizes. It shows that
the relative contribution of the smaller flaws in causing the damage is larger for smaller Z which is similar to
the dynamical effects associated with the increasing of the loading rates (Fig. 13).

Fig. 14. Shows the effect of the flaw density Z, on the strain rate dependence of the compressive strength of the material. Distribution
function is Gaussian with mean 2s0 ¼ 50 mm and standard deviation s ¼ 5 mm. Note that the material with lower flaw density tends to have
higher strain rate sensitivity.
 ARTICLE IN PRESS
916 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Fig. 15. Surface plot shows the evolution of crack length, vs. the applied strain, for an imposed strain rate of 105 s1 and with a Gaussian
distribution of initial flaw size with mean 2s0 ¼ 50 mm and standard deviation s ¼ 5 mm. Two flaw density parameters are considered:
(a) Z ¼ 108 and (b) Z ¼ 105. Calculations are performed till the peak stress is reached in both the cases. Note that the wing-crack length
distribution is more uniform over the population of flaws, at lower Z. The arrows indicate the wing-crack lengths from various flaws at the
peak stress.

4.4. Effect of the mean flaw size (2s0)

pffiffiffiffi
Fig. 16 shows the variation of the compressive strength normalized by K IC = s0 against the initial
3 1
flaw size for a strain rate of 10 s . The dashed horizontal line shows the normalized initiation stress
(Ashby and Hallam, 1986). A delta distribution of pre-existing flaw size was considered for the simulations.
The result suggests that the failure stress decreases with the increase in flaw size. However, unlike initiation
stress, it shows a stronger non-linear response than the inverse square-root of the flaw size. Bhattacharya
et al. (1998) developed an energy based model to investigate the compressive failure that accounts
for the heterogeneity in brittle materials. They predicted an inverse square-root relationship of uniaxial
compressive strength with the grain size of the material (which was the only length scale in the model). The
addition of the interaction and growth dynamics in the present model appears to lead to a stronger
dependence on flaw size.

4.5. Effect of the friction coefficient (m)

Fig. 17(a) shows the variation of the peak stress and the initiation stress (stress required to nucleate the wing
cracks from the flaws) normalized by the respective stress at m ¼ 0. A delta distribution of pre-existing flaw
size was considered with mean 2s0 ¼ 50 mm. Note that the compressive strength increases with the increase in
friction coefficient and approximately doubles as m varies from 0.0 to 0.5. Dashed lines are the straight lines
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 917

Fig. 16. Shows the effect of the flaw size on the compressive strength of the material for a delta distribution of the flaw size with mean
2s0 ¼ 50 mm, at 103 s1 imposed strain rate. Note the stronger non-linear variation than the square-root of the flaw size of the compressive
pffiffiffiffi
strength with the flaw size. The straight dashed line is the initiation stress normalized with K IC = s0 .

Fig. 17. (a) Shows the plot of the normalized compressive strength and the initiation stress against the friction coefficient (m), with the
respective stress at m ¼ 0, for a delta distribution of pre-existing flaw size with mean 2s0 ¼ 50 mm at a imposed strain rate of 103 s1 and
(b) shows the wing-crack length plotted against the strain for different m..

which show that the peak stress exhibits stronger non-linear variation with the friction coefficient than the
initiation stress. This physical phenomenon can be explained as follows. During the deformation process, flaw
surfaces slide against each other causing the stress concentration at their tips (see Fig. 1 above). Friction exists
on these surfaces and resists the sliding motion. Hence, higher stress is required to initiate the wing cracks
from the flaws. Fig. 17(b) shows the wing-crack length plots for m ¼ 0.0, 0.2 and 0.4 and for a constant strain
rate of 103 s1. Note that increase in the friction coefficient requires higher strain (and hence higher stress) for
the initiation of the wing cracks. It is also seen that the length decreases with the increase in the friction
coefficient, consistent with the predictions of Deng and Nemat-Nasser (1992). Therefore, the load bearing
capacity of the material increases with the increase in frictional resistance as the damage tends to accumulate
over a longer duration. It is also evident from the figure that the crack speed decreases with the increase in the
friction coefficient, which explains the stronger non-linear variation of the peak stress with the friction
coefficient than that of initiation stress.
 ARTICLE IN PRESS
918 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

5. Summary

A micro-mechanical damage model for brittle failure under compression is proposed based on frictional
sliding of pre-existing flaws. These pre-existing flaws are assumed to be uniformly distributed in space but follow
a certain distribution in size. As a result of frictional sliding due to external loading, wing cracks are nucleated at
the tips of these flaws inducing the damage in the material. The model incorporates the interactions among the
growing micro-cracks by means of a crack-matrix-effective-medium approach, which gives an effective stress
field around the cracks which is different from that if they were isolated. We define load-induced damage as a
scalar crack density parameter, evolution of which is a function of the pre-existing flaw distributions and the
crack-growth law for a particular material. The methodology is applied for the case of uniaxial compression
under constant strain rate loading (lateral compression and overall confinement effects are currently under study
and will be addressed in a subsequent work). The results can be summarized as follows:

 The model predicts a macroscopic stress–strain response and a peak stress (defined as the compressive
strength) with an associated transition strain rate beyond which the compressive strength of the material
becomes highly strain rate sensitive. Results also shows that as the strain rate increases, the post-peak stress
collapse and the damage evolution become slower with strain, with an increase in the associated damage at
the peak stress.
 Dynamic compressive strength of the material is sensitive to the crack growth law ðl_ ¼ cmax
ððK I  K IC Þ=ðK I  K IC =2ÞÞg Þ. Parameter g has a profound effect on the transition strain rate where as
parameter a strongly influences the dynamic strain rate sensitivity. Model also predicts that g does not have an
influence on the softening characteristics of the material as compared to the parameter a. Increase in a
(associated with lower terminal crack speed of the material) causes delayed softening response with strain.
 Material with narrower distributions (i.e. lower s) of initial flaw size about the same mean value has a
higher compressive strength. This effect is more prominent at lower strain rates. However, the stress–strain
response does not seem to vary with the change in the distribution functions from Gaussian to Weibull
having the same value for mean (2s0) and standard deviation (s).
 The model predicts that a material with lower flaw density (Z) has a higher compressive strength and this
effect of Z increases with the increase in strain rate. The model also predicts that as Z increases, damage at
the peak stress (in the form of wing-crack length) is distributed more uniformly over the pre-existing flaws.
 At lower strain rates, the damage caused by the bigger flaws is more and its over all distribution over the
pre-existing flaws is less uniform.
 The model also predicts that the compressive strength of the material increases with the decrease in the
mean flaw size and increase in the coefficient of friction.

In conclusion, the approach provides a sound basis for predicting several physical phenomena related to brittle
fracture. However, several assumptions are made in constructing the model, primarily due to lack of sufficient
experimental data in the dynamic loading regime. For example: (i) the fracture toughness of the material is
assumed to be a constant and independent of the applied strain rate, and (ii) the form of the crack growth law.

Acknowledgment

This work was performed under the auspices of the Center for Advanced Metallic and Ceramic Systems
(CAMCS) at the Johns Hopkins University, supported by the Army Research Laboratory under the
ARMAC-RTP Cooperative Agreement No. DAAD19-01-2-0003.

Appendix A

A.1. Auxiliary problem: intact elliptical inclusion in a homogeneous matrix

Consider an isotropic elastic inclusion embedded in an infinite orthotropic elastic matrix, subjected to
remote principal stresses N1 and N2, as shown in Fig. 18. The semi-major and the semi-minor axes of the
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 919

Fig. 18.

inclusion are a and b, respectively. The interface (G) between the matrix (Om) and the inclusion (Oi) is assumed
to be perfectly bonded and the axis of anisotropy of the surrounded matrix is aligned with the vertical and the
horizontal axis. We shall use the complex variable method to construct the solution for the homogeneous
stress state inside the inclusion (Fig. 18).
Following Green and Zerna (1968), Section 6.7, Chen (1967), let (x, y) be rectangular cartesian coordinates
and let
)
z ¼ x þ iy; z ¼ x  iy
, (A.1)
zi ¼ z þ gi z; zi ¼ z þ gi z

with
jg1 jo1; jg2 jo1, (A.2)
where g1 and g2 are complex material constants which are the roots of the following characteristic equation
(Green and Zerna, 1968, Section 6.9),

s22 4 11 12 2 11
22 a  2ðs22 þ 2s12 Þa  s11 ¼ 0
1þg a1 1 þ gk
a¼ ; g¼ ; ak ¼ ; ðk ¼ 1; 2Þ ðA:3Þ
1g aþ1 1  gk
and sijkl are the components of the planer compliance tensor S of the material,
2 11 11 3
s11 s22 0
6 s22 0 7
S¼4 22 5. (A.4)
s12
12

If we express Airy’s stress function in terms of two complex potentials as (Green and Zerna, 1968, Section
9.1),

f ¼ Oðz1 Þ þ Ōðz1 Þ þ oðz2 Þ þ oðz2 Þ, (A.5)

) Displacements and tractions (and hence the stresses) can be written as follows,
0
D ¼ ux þ iuy ¼ d1 O0 ðz1 Þ þ r1 Ō ðz1 Þ þ d2 o0 ðz2 Þ þ r2 o0 ðz2 Þ, (A.6)

0
P ¼ X þ iY ¼ 2ifg1 O0 ðz1 Þ þ Ō ðz1 Þ þ g2 o0 ðz2 Þ þ o0 ðz2 Þg, (A.7)
 ARTICLE IN PRESS
920 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

00 )
Y ¼ sxx þ syy ¼ 4g1 O00 ðz1 Þ þ 4g1 Ō ðz1 Þ þ 4g2 o00 ðz2 Þ þ 4g2 o00 ðz2 Þ
00 , (A.8)
F ¼ sxx  syy þ 2isxy ¼ 4g21 O00 ðz1 Þ  4Ō ðz1 Þ  4g22 o00 ðz2 Þ  4o00 ðz2 Þ

where prime denotes the differentiation with respect to the argument.

In these expressions,
)
d1 ¼ ð1 þ g1 Þb2  ð1  g1 Þb1 ; d2 ¼ ð1 þ g2 Þb1  ð1  g2 Þb2
, (A.9)
r1 ¼ ð1 þ g1 Þb2 þ ð1  g1 Þb1 ; r2 ¼ ð1 þ g2 Þb1 þ ð1  g2 Þb2

where

bk ¼ s11 22 2
22  s22 ak ðk ¼ 1; 2Þ. (A.10)
This problem involves an elliptical boundary (G), so it is convenient to use the following conformal
transformation which maps the elastic region in the physical z-plane onto a region of circular geometry in the
z-plane (Green and Zerna, 1968, Section 9.19):

z ¼ zðzÞ ¼ cz þ d=z; z ¼ exþiZ , (A.11)

where
aþb ab
c¼ ; d¼ . (A.12)
2 2
This function maps the region outside the elliptical contour with semi-axis (a, b) onto a region outside the
circle |z| ¼ 1.
We define two other complex variables z1 and z2 by the transformation:
zk ¼ ðc þ gk dÞzk þ ðd þ gk cÞ=zk ðk ¼ 1; 2Þ. (A.13)
The zk-planes are chosen so that the circles |zk| ¼ 1 (k ¼ 1, 2) correspond to the circle |z| ¼ 1 in the z-plane
and to the ellipse in the z-plane. For convenience we may imagine that the three circles |z1| ¼ 1, |z2| ¼ 1,
|z| ¼ 1 coincide at the elliptical boundary and are given by

z1 ¼ z2 ¼ z ¼ eiZ ; z=1=z. (A.14)

Now we put,
O0 ðz1 Þ ¼ f ðz1 Þ; o0 ðz2 Þ ¼ gðz2 Þ, (A.15)
and substitute into Eqs. (A.6) and (A.7) to get the expressions for the displacements and tractions in terms of
z k.
Next we have to determine the complex potentials, both in the matrix and the inclusion, which should
satisfy the following boundary conditions.

1. Across the elliptical interface there is continuity of tractions and displacements, i.e. the values of P and D
for the inclusion and the matrix must be the same at the interface.
2. Far away from the inclusion the stress approaches the prescribed values defined by N1 and N2.

We select the following form of the complex potential for the matrix:

f ðz1 Þ ¼ H 1 z1 þ Ḡ 1 =z1 ; gðz2 Þ ¼ H 2 z2 þ Ḡ2 =z2 , (A.16)

where H1 and H2 are given by
H 1 ¼ ðB þ iCÞðc þ g1 dÞ; H 2 ¼ ðB0 þ iC 0 Þðc þ g2 dÞ, (A.17)
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 921

with B, C, B0 , C0 are given in the terms of the boundary stresses, by solving the following system of equations
(refer to Green and Zerna, 1968, Section 9.1):
)
N 1 þ N 2 ¼ 4g1 ðB þ iCÞ þ 4g1 ðB  iCÞ þ 4g2 ðB0 þ iC 0 Þ þ 4g2 ðB0  iC 0 Þ
, (A.18)
N 2  N 1 ¼ 4g21 ðB þ iCÞ þ 4ðB  iCÞ þ 4g22 ðB0 þ iC 0 Þ þ 4ðB0  iC 0 Þ

and G1 and G2 are yet to be determined.

Substituting these into Eqs. (A.6) and (A.7), we find that at the elliptical boundary, |z| ¼ 1:
D ¼ ðd1 H 1 þ r1 G 1 þ d2 H 2 þ r2 G 2 Þz þ ðd1 G 1 þ d2 G 2 þ r1 H 1 þ r2 H 2 Þz, (A.19)

P ¼ 2ifðg1 H 1 þ G1 þ g2 H 2 þ G 2 Þz þ ðH 1 þ g1 G 1 þ H 2 þ g2 G 2 Þzg. (A.20)

Now consider the elliptical region Oi. Parameters referring to the region are distinguished by a superscript,
e.g. gI1 ; dI1 ; rI1 , etc. Assuming that the stresses are uniform inside the inclusion, complex potentials inside the
inclusion are:
O0 ðz1 Þ ¼ A1 z1 ; o0 ðz2 Þ ¼ A2 z2 , (A.21)
where A1 and A2 are unknown in the expression. A new complex variable, Z, is introduced:
Z ¼ x=a þ iy=b, (A.22)
On the inclusion surface, |Z| ¼ 1. We now define four parameters mj, nj (j ¼ 1, 2):
)
2mj ¼ ð1 þ gIj Þa þ ð1  gIj Þb
, (A.23)
2nj ¼ ð1 þ gIj Þa  ð1  gIj Þb

so that
zj ¼ mj Z þ nj Z. (A.24)
Substituting into Eqs. (A.6) and (A.7), we can get an expression for the displacement D and the force P in
the inclusion:
D ¼ ðdI1 A1 m1 þ rI1 A1 n1 þ dI2 A2 m2 þ rI2 A2 n2 ÞZ
þ ðdI1 A1 n1 þ rI1 A1 m1 þ dI2 A2 n2 þ rI2 A2 m2 ÞZ, ðA:25Þ

P ¼ 2ifðgI1 A1 m1 þ A1 n1 þ gI2 A2 m2 þ A2 n2 ÞZ
þ ðgI1 A1 n1 þ A1 m1 þ gI2 A2 n2 þ A2 m2 ÞgZ. ðA:26Þ
Next we set the tractions and the displacements along the interface, represented by |Z| ¼ 1 and |z| ¼ 1
between the inclusion and the matrix, equal. This results in the following simultaneous equations,
9
gI1 A1 m1 þ A1 n1 þ gI2 A2 m2 þ A2 n2  G 1  G 2 ¼ g1 H 1 þ g2 H 2 >
>
>
>
A1 m1 þ gI1 A1 m1 þ A2 m2 þ gI2 A2 m2  g1 G 1  g2 G2 ¼ H 1 þ H 2 =
I I I I . (A.27)
d1 A1 m1 þ r1 A1 n1 þ d2 A2 m2 þ r2 A2 n2  r1 G 1  r2 G 2 ¼ d1 H 1 þ d2 H 2 > >
>
>
rI1 A1 m1 þ d1 A1 n1 þ rI2 A2 m2 þ d2 A2 n2  d1 G 1  d2 G 2 ¼ r1 H 1 þ r2 H 2 ;
I I

The four unknown complex constants G1, G2, A1 and A2 can be determined by solving this system of
equations. Plugging the obtained values of A1 and A2 into Eq. (A.21), we can determine the homogeneous
stress state inside the inclusion by solving Eq. (A.8). Substituting the obtained values of G1 and G2 (and
already known H1 and H2) in the potential functions defined by Eqs. (A.16) and (A.15), stress state in the
matrix can be found by solving Eq. (A.8).
This complex variable approach provides a complete solution for the stresses and the displacements to the
elasticity problem.
 ARTICLE IN PRESS
922 B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923

Hence, for the case of uniaxial compression the far-field stress P (the problem posed in Fig. 6(a)) takes the
0 1
s
B C
form P ¼ @ 0 A, so that N1 ¼ s and N2 ¼ 0. Then using the approach described above, one can find the
0
0 e 1
s11
e B se C
homogeneous stress state P ¼ @ 22 A inside the inclusion, which is used to compute the effective Mode-I SIF
se12
at the crack tips.

References

Ashby, M.F., Hallam, S.D., 1986. The failure of brittle solids containing small cracks under compressive stress states. Acta Metall. 34,
497–510.
Ashby, M.F., Sammis, C.G., 1990. The damage mechanics of brittle solids in compression. PAGEOPH 133, 489–521.
Basista, M., Gross, D., 2000. A note on crack interactions under compression. Int. J. Fract. 102, L67–L72.
Bhattacharya, K., Ortiz, M., Ravichandran, G., 1998. Energy-based model of compressive splitting in heterogeneous brittle solids.
J. Mech. Phys. Solids 46, 2171–2181.
Brace, W.F., Bombolakis, E.G., 1963. A note on brittle crack growth in compression. J. Geophys. Res. 68, 3709–3713.
Budiansky, B., O’Connell, R.J., 1976. Elastic moduli of a cracked solid. Int. J. Solids Strcut. 12, 81–97.
Chen, W.T., 1967. On an elliptic elastic inclusion in an anisotropic medium. Quart. J. Mech. Appl. Math 20, 307–313.
Deng, H., Nemat-Nasser, S., 1992. Dynamic damage evolution in brittle solids. Mech. Mater. 14, 83–104.
Eberhardt, E., Stead, D., Stimpson, B., 1999. Quantifying progressive pre-peak brittle fracture damage in rock during uniaxial
compression. Int. J. Rock Mech. Minor. Sci. 36, 361–380.
Eshelby, J., 1957. The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc. R. Soc. London A A 241,
376–396.
Freund, L.B., 1972. Crack propagation in an elastic solid subjected to general loading II. Nonuniform rate of extension. J. Mech. Phys.
Solids 20, 141–152.
Freund, L.B., 1990. Dynamic Fracture Mechanics. Cambridge University Press, New York.
Grady, D.E., 1982. Local inertial effects in dynamic fragmentation. J. Appl. Phys. 53, 322–325.
Grady, D.E., Kipp, M.E., 1980. Continuum modelling of explosive fracture in oil shale. Int. J. Rock Mech. Miner. Sci. 17, 147–157.
Green, A.E., Zerna, W., 1968. Theoretical Elasticity. Clarendon Press, Oxford.
Horii, H., Nemat-Nasser, S., 1985. Elastic fields of interacting inhomogeneities. Int. J. Solids Strcut. 21, 731–745.
Horii, H., Nemat-Nasser, S., 1986. Brittle failure in compression: splitting, faulting and brittle-ductile transition. Philos. Trans. R. Soc.
London A 319, 337–374.
Huang, C., Subhash, G., 2003. Influence of lateral confinement on dynamic damage evolution during uniaxial compressive response of
brittle solids. J. Mech. Phys. Solids 51, 1089–1106.
Huang, C., Subhash, G., Vitton, S.J., 2002. A dynamic damage growth model for uniaxial compressive response of rock aggregates. Mech.
Mater. 34, 267–277.
Huang, Y., Hu, K.X., Chandra, A., 1994. A generalized self-consistent mechnaics method for microcracked solids. J. Mech. Phys. Solids
42, 1273–1291.
Kachanov, M., 1987. Elastic solids with many cracks: a simple method of analysis. Int. J. Solids Strcutures 23, 23–43.
Kobayashi, A.S., 2002. Dynamic fracture of ceramics and CMC. In: McCauley, J.W., Crowson, A., Gooch, Jr., W.A., Rajendran, A.M.,
Bless, S.J., Logan, K.V., Normandia, M., Wax, S. (Eds.), Ceramic Armor Materials by Design. The American Ceramic Society,
Westerville, OH, pp. 185–196.
Lankford, J., 1981. Mechanisms responsible for strain rate dependent compressive strength in ceramic materials. J. Am. Ceram. Soc. 64,
C33–C35.
Lankford, J., 1996. High strain rate compression and plastic flow of ceramics. J. Mater. Sci. Lett. 15, 745–750.
Nemat-Nasser, S., Deng, H., 1994. Strain-rate effect on brittle failure in compression. Acta Metall. Mater. 42, 1013–1024.
Nemat-Nasser, S., Horii, H., 1982. Compression induced nonplanar crack extension with application to splitting, exfoliation and
rockburst. J. Geophys. Res. 87, 6805–6821.
Nemat-Nasser, S., Obata, M., 1988. A microcrack model of dilatancy in brittle materials. Trans. ASME J. Appl. Mech. 55, 24–35.
Paliwal, B., Ramesh, K.T., McCauley, J.W., 2006. Direct observation of the dynamic compressive failure of a transparent polycrystalline
ceramic (AlON). J. Am. Ceram. Soc. 89, 2128–2133.
Ravichandran, G., Subhash, G., 1995. A micromechanical model for high strain rate behavior of ceramics. Int. J. Solids Struct. 32,
2627–2646.
Rosakis, A.J., Ravichandran, G., 2000. Dynamic failure mechanics. Int. J. Solids Strcut. 37, 331–348.
 ARTICLE IN PRESS
B. Paliwal, K.T. Ramesh / J. Mech. Phys. Solids 56 (2008) 896–923 923

Sarva, S., Nemat-Nasser, S., 2001. Dynamic compressive strength of silicon carbide under unaxial compression. Mater. Sci. Eng. A 317,
140–144.
Schulson, E.M., Kuehn, G.A., Jones, D.A., Fifolt, D.A., 1991. The growth of wing cracks and the brittle compressive failure of ice. Acta
Metall. Mater. 39, 2651–2655.
Shao, J.F., Rudnicki, J.W., 2000. A microcrack-based continuous damage model for brittle geomaterials. Mech. Mater. 32, 607–619.
Suresh, S., Nakamura, T., Yeshurun, Y., Yang, K.-H., Duffy, J., 1990. Tensile fracture toughness of ceramic materials: effects of dynamic
loading and elevated temperatures. J. Am. Ceram. Soc. 73, 2457–2466.
Tapponnier, P., Brace, W.F., 1976. Development of stress-induced microcracks in Westerly granite. Int. J. Rock Mech. Miner. Sci. 13,
103–112.
Vekinis, G., Ashby, M.F., Beaumont, P.W.R., 1991. The compressive failure of alumina containing controlled distributions of flaws. Acta
Metall. Mater. 39, 2583–2588.
Wang, H., Ramesh, K.T., 2003. Dynamic strength and fragmentation of hot-pressed silicon carbide under uniaxial compression. Acta
Mater. 52, 355–367.
Weerasooriya, T., Moy, P., Casem, D., 2006. A four-point bend technique to determine dynamic fracture toughness of ceramics. J. Am.
Ceram. Soc. 89, 990–995.
Xu, X.H., Ma, S.P., Xia, M.F., Ke, F.J., Bai, Y.L., 2004. Damage evaluation and damage localization of rock. Theor. Appl. Fract. Mech.
42, 131–138.
Zhou, F., Molinari, J.-F., Ramesh, K.T., 2006. Effects of material properties on the fragmentation of brittle materials. Int. J. Fract. 139,
169–196.