ORI GI NAL PAPER

Experimental and Numerical Studies on Determination

of Indirect Tensile Strength of Rocks
Nazife Erarslan

Zheng Zhao Liang

David John Williams
Received: 6 September 2011 / Accepted: 15 November 2011 / Published online: 25 November 2011
Springer-Verlag 2011
Abstract Indirect tension tests using Brisbane tuff Bra-
zilian disc specimens under standard Brazilian jaws and
various loading arcs were performed. The standard Brazilian
indirect tensile tests caused catastrophic, crushing failure of
the disc specimens, rather than the expected tensile splitting
failure initiated by a central crack. This led to an investiga-
tion of the fracturing of Brazilian disc specimens and the
existing indirect tensile Brazilian test using steel loading arcs
with different angles. The results showed that the ultimate
failure load increased with increasing loading arc angles.
With no international standard for determining indirect
tensile strength of rocks under diametral load, numerical
modelling and analytical solutions were undertaken.
Numerical simulations using RFPA
2D
software were con-
ducted with a heterogeneous material model. The results
predicted tensile stress in the discs and visually reproduced
the progressive fracture process. It was concluded that
standard Brazilian jaws cause catastrophic, crushing failure
of the disc specimens instead of producing a central splitting
crack. The experimental and numerical results showed that
20 and 30 loading arcs result in diametral splitting fractures
starting at the disc centre. Moreover, intrinsic material
properties (e.g. fracture toughness) may be used to propose
the best loading conguration to determine the indirect
tensile strength of rocks. Here, by using numerical outcomes
and empirical relationships between fracture toughness and
tensile strength, the best loading geometry to obtain the most
accurate indirect tensile strength of rocks was the 2a = 30
loading arc.
Keywords Indirect tensile strength Brazilian test
RFPA method Diametral loading of rock discs
Abbreviations
ISRM International Society for Rock Mechanics
UCS Uniaxial compressive strength test
BTS Brazilian tensile strength
1 Introduction
The indirect Brazilian test is widely used in engineering
practice to indirectly determine the tensile strength of
rocks. In 1978, the Brazilian test was ofcially proposed by
the International Society for Rock Mechanics (ISRM) as a
suggested method for determining the tensile strength of
rock materials (ISRM 2007). The Brazilian test (splitting
tension test) is performed by applying a concentrated
compressive load across the diameter of a disc specimen.
The standard Brazilian test results in a reasonably nar-
row measured data scatter. In addition, the biaxial stress
state in a test specimen subjected to the Brazilian test is
considered approximately similar to that which appears in a
rock mass when tensile failure occurs in situ. However,
there are a few unsolved problems concerning this test; for
example, how to guarantee crack initiation at the centre of
the specimen (beneath the concentrated load), how to
determine the true stress state in a test specimen and how to
locate the point from which the main crack initiates.
Although the Brazilian test has been studied extensively,
N. Erarslan (&) D. J. Williams
Golder Geomechanics Centre, School of Civil Engineering,
The University of Queensland, Brisbane, QLD 4072, Australia
e-mail: nazife.tiryaki@uqconnect.edu.au
Z. Z. Liang
School of Civil and Hydraulic Engineering,
Dalian University of Technology, Dalian 116024, China
1 3
Rock Mech Rock Eng (2012) 45:739751
DOI 10.1007/s00603-011-0205-y
both experimentally and theoretically, relatively little
attention has been directed towards researching the validity
of the test. Fairhurst (1964) rst discussed the important
issue of the validity of the Brazilian test. He stated that
failure may occur away from the centre of the test disc for
small angles of loading contact area and indicated that the
calculated tensile strength from a Brazilian test is lower
than the true value of the tensile strength. Hudson et al.
(1972) observed that failure always initiated directly under
the loading points if at steel plates were used to load the
specimen, an approach that actually invalidates the test for
the determination of tensile strength.
It is possible to estimate stress distribution in a disc
specimen theoretically if some assumptions are made.
Although various methods have been proposed for this
purpose, the common fundamental assumptions are: (1)
rock behaves elastically in both tension and compression;
(2) breakage of the loading zone does not inuence the
stress distribution at the centre of the specimen; and (3)
high compressive stress exists around the two loading
points and, therefore, the tensile crack never initiates from
this region. Based on those assumptions, classical theory
(ISRM 2007) assumed that the concentrated load was
applied over an innitesimally small width as a line load,
but clearly, this would lead to stresses of very high inten-
sity. The actual loads are not concentrated, but are dis-
tributed over a nite arc of the disc. The tensile strength of
a rock disc specimen, r
t
, is calculated using:
r
t
=
2P
pdt
(1)
where P is failure load and d and t are the diameter and
thickness, respectively, of the rock disc. According to
Jaeger and Cook (1976), if a circular cylinder of radius R is
compressed across its diameter between at surfaces that
apply concentrated loads of W per unit axial length of the
cylinder, then the stresses on this diameter are:
r
x
=
W
pR
r
y
=
W(3R
2
y
2
)
pr(R
2
y
2
)
:
(2)
By symmetry, these are principal stresses and the major
and minor principal stresses are along and across the y-axis,
respectively. If the load is applied to the circumference of
the cylinder as a pressure p distributed over an arc 2a by
using shaped platens, so that W = 2paR, then equal biaxial
compression exists near the contacts with a value of
p (Jaeger and Cook 1976).
However, the actual loads in an experiment are not
concentrated, but are distributed over nite portions of the
disc. The distributed load applied to a disc under diametral
compression is more difcult to analyse than that of the
concentrated load. Hondros (1959) analysed the Brazilian
test for the case of a thin disc loaded by a uniform pressure
applied radially over a short strip of the circumference at
each end of the disc. He obtained the full-eld stresses in a
series solution by using the series expansion technique and
applied these solutions to evaluate the Youngs modulus
and Poissons ratio in concrete, by measuring strain as
follows:
r
r
=
2p
p
a

n=
n=1
1 1
1
n
_ _
r
R
_ _
2
_ _
r
R
_ _
2n2
sin 2na cos 2nh
_ _
r
h
=
2p
p
a

n=
n=1
1 1
1
n
_ _
r
R
_ _
2
_ _
r
R
_ _
2n2
sin 2na cos 2nh
_ _
s
rh
=
2p
p

n=
n=1
1
r
R
_ _
2
_ _
r
R
_ _
2n2
sin 2na cos 2nh
_ _
(3)
where p is applied pressure, R is the radius of the disc,
r and h are the polar coordinates of a point in disc and a is
the half central angle of the applied distributed load
(Fig. 1). It can be seen from (3) that the magnitude of a
affects directly the stress distribution within the disc.
In addition to analytical solutions that assume the rock
material to be isotropic and homogenous, recently many
researchers have included the effect of rock heterogeneity in
their experimental and numerical studies (Van de Steen 2001;
Cai and Kaiser 2004; Van de Steen et al. 2005; Lanaro et al.
2009; Liu 2004). Van de Steen et al. (2005) experimentally and
numerically studied the irregular, granular nature of rock as
well as the effect of the presence of defects and weaknesses on
Fig. 1 A disc specimen subjected to diametric distributed
compression
740 N. Erarslan et al.
1 3
the stress distribution in a specimen. They successfully simu-
lated the deformation in disc specimens, including defects, by
using a boundary element code DIGS, and they proposed that
fracturing in the Brazilian test is initiated in shear near one of
the platens and it subsequently grows in tension.
Liu (2004) used a heterogeneous material model
implemented into the R-T
2D
code in his numerical studies.
He showed the effect of rock heterogeneity on the stress
distribution and fracturing process in tested specimens by
using a variety of numerical tests such as the uniaxial
compressive strength (UCS) test, the Brazilian tensile
strength (BTS) test, diametral compression testing of a
notched Brazilian disc as well as three-point bending and
four-point shearing tests.
Recently, the effect of pre-existing cracks in terms of
heterogeneity was examined numerically by Lanaro et al.
(2009). They found that the propagation of initiated cracks
produces a stress eld that is very different from that
assumed when considering the rock material as continuous,
homogeneous, isotropic and elastic, as in many numerical
models as well as analytical solutions. They concluded that
stress concentrations at the bridges between the cracks
could reach tensile stresses much higher than the direct
tensile strength of the intact rock that was used as input in
their numerical models. This was due to the development
of large stress gradients between the cracks.
According to the Grifth criterion (Grifth 1020), the
exact centre of the disc is the only point where the conditions
for tensile failure at a value equal to the uniaxial strength are
met. However, in some experimental studies carried out
using standard Brazilian jaws, cracks initiated just under the
loading points instead of causing a central crack (Hudson
et al. 1972; Markides et al. 2010; Yu et al. 2009). Never-
theless, in the tests described herein, catastrophic crushing
failure developed on standard Brazilian testing of brittle
Brisbane tuff disc specimens. In contrast, central cracks were
obtained, corresponding to the location of the maximum
tensile stress, for loading of Brisbane tuff specimens over an
arc length. Thus, the objective of this paper is to criticise the
standard Brazilian test by comparing the results obtained
using the test with those obtained using a loaded arc, together
with a comparison of the experimental results and the ana-
lytical and numerical modelling results.
2 Experimental Study
2.1 Indirect Tension Tests
A series of Brazilian disc tests was carried out using
specimens prepared from Brisbane tuff, a host rock of
Brisbane, Australias rst motorway tunnel, CLEM7, from
which core samples were obtained. The test specimens
were standard Brazilian discs with a diameter of 52 mm
and thickness of 26 mm (a diameter to thickness ratio of
0.5). The test load was applied by a stiff hydraulic Instron
loading frame with an ISRM-suggested loading rate of
200 N/s (ISRM 2007).
Four series of indirect tension tests were conducted: (1)
standard Brazilian jaws, (2) 15 steel loading arcs, (3) 20
steel loading arcs and (4) 30 steel loading arcs. Up to four
repetitions were carried out. The steel loading arcs were
machined from standard mild steel, as recommended by
ISRM (2007) (Fig. 2).
The tensile strength of the rock specimens tested using
standard Brazilian jaws were calculated using the formula
given by ISRM (2007). Since loading boundaries of steel
loading arcs are different from standard Brazilian jaws, the
formula given by ISRM (2007) cannot be used to calculate
indirect tensile strength of specimens under loading arcs.
The tensile strength of the samples tested under angled
loading arcs was calculated at the centre of the disc by
using the analytical solutions given in (3).
The details of the test results are given in Table 1. The
maximum recorded failure load was obtained by using
loading arcs with 2a = 30. The second highest failure
load was obtained with the standard Brazilian jaws, which
also produced the highest standard deviation among the
ultimate loads.
In our tests on Brisbane tuff, loading with Brazilian jaws
caused catastrophic crushing failure of the disc specimens
(Fig. 3a). However, a single failure plane diverged from
the load axis and secondary cracks occurred under 15
loading arcs (Fig. 3b). In contrast to the results from the
15 loading arcs, one axial splitting failure plane through
the diametral loading axis was obtained with the samples
tested by using the 20 loading arcs (Fig. 3c). Unlike the
results obtained in the standard Brazilian and the 15 and
20 loading arc tests, arrested and vertical aligned central
cracks were obtained with 30 loading arcs (Fig. 3d).
As an alternative to the uniaxial tensile test, which is
difcult to perform to acceptable standards for brittle
materials, diametral compression of Brazilian disc speci-
mens has strong practical appeal. It is clear that a diametral
compression test can never adequately replace a direct
uniaxial tensile test due to their different loading bound-
aries and specimen geometries. However, conducting
direct tests are still too difcult and expensive for routine
application to a large numbers of specimens, and diametral
compression tests appear to offer the most desirable alter-
native. On that basis, this study focuses on the failure in
standard Brazilian discs under diametral compression to
determine the most representative indirect tensile strength
of rocks. Thus, the Brazilian test under 30 loading arcs is
preferable to such a test under Brazilian jaws, in terms of
central tensile splitting. However, it is not easy to conclude
Determination of Indirect Tensile Strength of Rocks 741
1 3
that the obtained failure load under 30 loading arcs rep-
resents the real value of the indirect tensile strength of
rocks. To investigate further, numerical and analytical
analyses are used for the determination of the indirect
tensile strength of rocks in this research.
3 Numerical Modelling of Heterogeneous Rock
Fracturing Under Indirect Tensile Stresses
3.1 Numerical Method
Rock or rock-like materials have composite or heteroge-
neous microstructures with various scales. The behaviour of
rock fractures and the growth of cracks or micro-cracks in
rocks are strongly inuenced by these heterogeneous
microstructures (Grifth 1920; Liu 2004; Cai and Kaiser
2004; Lanaro et al. 2009). Consequently, the numerical
methods used for studying the fracture behaviour of rocks
should take heterogeneity into consideration. Among the
numerical methods developed in the past, some researchers
have included heterogeneity in their programs (Schlangen
and Garboczi 1997; Tang et al. 1998; Blair and Cook 1998).
A numerical approach called Rock Fracture Process
Analysis (RFPA) was used in this study to elucidate the
fracturing process in a Brazilian disc under different
boundary loading conditions. Liu (2004) has listed three
main advantages of the RFPA model over other models: (1)
by introducing the heterogeneity of rock properties into the
model, RFPA can simulate the non-linear deformation of a
quasi-brittle rock with an ideal brittle constitutive law for
the local material; (2) by introducing the reduction of the
Fig. 2 a Brazilian jaws and
steel loading arcs, b disc
between standard Brazilian jaws
and c disc between loading arcs
Table 1 Results of indirect
tensile tests on Brisbane tuff
disc specimens
Recorded maximum load (kN)
Standard Brazilian jaws 15 loading arcs 20 loading arcs 30 loading arcs
Replicate 1 25.00 12.50 17.06 21.10
Replicate 2 16.77 16.39 19.82 24.60
Replicate 3 15.43 15.65 20.23 21.13
Replicate 4 21.00 14.70 19.41 22.17
Average 19.60 14.81 19.20 22.30
Standard deviation 4.34 1.69 1.64 1.64
742 N. Erarslan et al.
1 3
material parameters after element failure, RFPA can sim-
ulate strain-softening and discontinuum mechanics prob-
lems in a continuum mechanics mode; and (3) by recording
the event rate of failed elements, RFPA can simulate
micro-seismicity associated with progressive fracture
process.
To deal with real random microstructures in a numerical
simulation, rock heterogeneity can be characterized better
with statistical approaches (Liu 2004; Tang et al. 1998). In
the RFPA
2D
software, since the numerical specimens
consist of elements with the same shape and size, there is
no priority, geometrically, in any orientation in the speci-
men. Disorder can be obtained by means of random dis-
tributions of the mechanical properties of the elements. The
statistical distribution of the elastic modulus can be
described by the Weibull distribution function, even dis-
tribution function or normal distribution function. In this
study, only the Weibull distribution function:
W(x) =
m
x
0
x
x
0
_ _
m1
exp
x
x
0
_ _
m
_ _
(4)
is used. In (4), x is the elemental mechanical parameter,
such as uniaxial compressive strength, elastic modulus,
Poisson ratio or specic weight, x
0
is the expected value of
x and m is the shape of the Weibull distribution function.
According to the Weibull distribution and the denition of
homogeneity index, a larger m implies that there are more
elements with mechanical properties approximated to the
mean value, thus a more homogeneous rock specimen.
Even though the mechanical parameters in each element
differ, which makes the specimen heterogeneous, the
mechanical properties within a single element are assumed
to be homogeneously distributed.
In RFPA
2D
, each element follows an elastic-brittle
constitutive law during the loading process. Until the stress
of the element satises the strength criterion, the elastic
modulus is a constant with the same value as before
loading. When the stress increases to a value leading to the
failure of the element, in elastic damage mechanics, the
elastic modulus of the element may degrade gradually as
damage progresses.
The elastic modulus of the damaged element is dened
as follows:
E = (1 D)E
0
(5)
where D represents the damage variable, and E and E
0
are
the elastic modulus of the damaged and undamaged ele-
ments, respectively. It must be noted that the element and
its damage are assumed to be isotropic and, therefore, E, E
0
and D are all scalar.
Here, it was assumed that each element might fail in
either tensile failure or shear failure modes. If the ele-
mental stress state satises both the tensile failure criterion
and the shear failure criterion, the tensile failure mode has
the higher priority. If the element is subjected to uniaxial
tensile stress, before the tensile stress (the minimal prin-
cipal stress) of the element reaches its tensile strength, the
element remains linear elastic. When the minimal principal
Fig. 3 Failed Brisbane tuff
specimens under a standard
Brazilian jaws, b 15 loading
arcs, c 20 loading arcs and
d 30 loading arcs
Determination of Indirect Tensile Strength of Rocks 743
1 3
stress increases beyond the tensile strength, the element
fails, the elastic modulus changes to a small value and the
elements strength falls to a small value, which we can call
residual tensile strength. When the tensile stress increases
to a larger value, which we call the ultimate tensile
strength, the element loses its capability of loading.
Accordingly, the element stays linear elastic before the
uniaxial compressive stress reaches the uniaxial compres-
sive strength. If the elemental stress level meets the shear
failure criterion, the element will damage gradually. The
damage variable D when the element is subjected to uni-
axial tension can be described as:
D =
0 (e
3
[e
r0
)
1
r
rt
e
3
E
0
(e
r0
_e
3
_e
ut
)
1 (e
3
_e
ut
)
_
_
_
_
_
_
(6)
where r
rt
is the residual strength of the element (dened as
r
rt
= k[r
t
[), e
t0
is the tensile strain at the point of failure,
e
3
is the principal strain, e
ut
is the ultimate tensile strain
(described as e
ut
= ge
t0
where g is called the ultimate
tensile strain coefcient) and k is coefcient of residual
tensile strength.
The damage evolution function mentioned above only
considers the tensile failure mode of the mesoscopic ele-
ments. Compressive softening induced by shear damage at
the mesoscopic level is also assumed to exist when the
mesoscopic element is under compressive and shear stress.
In shear failure mode, the damage variable D can be
described as follows:
D =
0 (e
1
\e
c0
)
1
r
rc
e
1
E
0
(e
1
_e
c0
)
_ _
(7)
where r
rc
is the peak strength of the element subjected to
uniaxial compression, e
c0
is the compressive strain at the
elastic limit and e
3
is the principal strain.
Variation in the damage variable is obtained when the
element is subjected to uniaxial tensile stress or uniaxial
compressive stress. If the specimens are subjected to
complex stress loading, we can extend the analysis from a
one-dimensional damage model to a three-dimensional
model by using an equivalent strain e instead of the uni-
axial tensile strain or the compressive strain in (6) and (7).
When the equivalent strain of an element reaches the
ultimate tensile strain, the damaged elastic modulus is zero,
which would make the system of equations ill conditioned.
In order to keep the physical continuum of the numerical
model, the element is not removed from the model and a
relatively small number, i.e. 1.0 9 10
-6
is specied for the
elastic modulus for this consideration. Therefore, contin-
uum mechanics can be applied to resolve discontinuous
problems.
However, it is impossible to obtain the parameters of the
mesoscopic elements. We can only obtain the mechanical
parameters of a macroscopic specimen for a specic rock
specimen in laboratory experiments. For Weibull distri-
bution determination, a parametric study can be performed
to obtain the relationships between the macroscopic
parameters (compressive strength r
c
and elastic modulus
E
0
) of the specimen and the seed parameters (mean value
of compressive strength r
c
and elastic modulus E
0
) of the
mesoscopic elements by using a linear least squares
technique:
r
c
= a
1
ln(m) b
1
[ [r
c
E
0
= a
2
ln(m) b
2
[ [E
0
(8)
where a
1
, b
1
, a
2
and b
2
are the constants in the linear least
squares technique.
3.2 Numerical Brazilian Tensile Strength (BTS) Test
This section describes our use of RFPA in numerical simu-
lations. Here, RFPA code (Tang et al. 1998) was used to
simulate Brazilian disc behaviour under various diametral
loading conditions. To model the typical brittle failure of
Brisbane tuff, the following characteristic parameters are
used in the numerical test. Based on experiments and a series
of numerical tests, the heterogeneity index for the simulated
rock is 4.0. The mean value of the uniaxial compressive
strength and elastic modulus for the rock are 202.5 MPa and
26 GPa, respectively, and the Poisson ratio value and friction
angle are 0.25 and 35, respectively. With regard to the
loading steel arcs (plates), the heterogeneity index is a suf-
ciently high value to consider the material to be almost
homogenous. The mean strength and elastic modulus are
1,500 MPa and 100 GPa, respectively. The stress is applied
by using a constant displacement movement of 0.0025 mm
per step on the top of the steel arc (plate), while the bottom
arc is xed in the vertical direction. The numerical model can
be assumed as a plain stress problem.
In this study, a numerical model for the BTS test is
based on the geometry used in our experiments, i.e. the
diameter of the disc is 52 mm. The three simulated loading
arcs and Brazilian jaws are shown in Fig. 4. The degree of
greyness as shown in Fig. 4 represents the distribution of
the elastic modulus. The lighter the grey colour, the higher
is the value of the elastic modulus.
3.3 Numerical Simulation Results
Figure 5 shows the progressive fracture process in heter-
ogeneous Brazilian disc samples under standard Brazilian
jaws and three different loading arcs. As the loading
increases using standard Brazilian jaws, cracks start at
points close to the upper contact along the vertical diameter
of the disc (Fig. 5a). Subsequently, cracks unstably
744 N. Erarslan et al.
1 3
propagate radially outward in the upper half of the disc,
resulting in a diametral fracture plane with small crack
coalescence.
When the angle of the steel arcs is increased, it was
difcult for the fractures to open just under the loading
point, due to the constraint stress in the horizontal direction
under the loading arcs. It is also interesting to mention that
any crushed zone just below the loading arcs, due to
the high compressive loading, was not observed in the
numerical simulations. In all loading arc simulations, the
rst failure and/or crack initiation occurred at the centre of
the disc. In contrast, primary crack initiation occurred very
close to the loading point in standard Brazilian jaw simu-
lations. Central primary crack initiation took place at the
centre of the disc with the 15, 20 and 30 loading arc
simulations (Fig. 5b, c, d). At the end of all fracture pro-
cess simulations, crack coalescence and failure took place
along the vertical diameter of the discs.
In addition to the importance of determining the central
crack initiation point, the propagation paths of the primary
cracks are also important. In the simulations there are some
secondary failure planes beside the main splitting failure
plane in rock, just next to the contact points between the
15 loading arc and the disc specimen (Fig. 5b-3). This
result supports the experimental ndings shown in Fig. 3.
Recorded peak loads obtained during simulations are
given in Table 2. The obtained ultimate loads during
experiments are also provided in Table 2 to allow com-
parison between simulation and experimental results.
Similar to the trend in the experimental results, the
numerical ultimate load obtained from simulations done by
using RFPA
2D
was found to increase with increasing
loading arc angles.
An analytical solution for a disc loaded by a uniform
pressure that is applied radially over a short strip of the
circumference at each end of the disc is given by Hondros
(1959). Under plane stress (disc) conditions, the theoretical
horizontal tensile stress along the vertical diameter, as
shown in Fig. 1, is given by:
r
h
=
2p
p
sin 2a
1 2
r
R
_ _
2
cos 2a
r
R
_ _
4
1
r
R
_ _
2
_ _
_
arctan
r
R
_ _
2
sin 2a
1
r
R
_ _
2
cos 2a
_ _
a
_
(9)
where p is the applied stress, R is the radius of the speci-
men and t its thickness and r is the polar coordinate.
A comparison between the analytical solution obtained
by using (9) for 20 loading arc angle (b = 20) and the
numerical simulation results of all loading geometries are
given in Fig. 6. The theoretical horizontal stress distribu-
tion is reasonably consistent with the numerical results,
although there are clear discrepancies towards the bound-
aries. The reason for these discrepancies may come from
differences in the assumed boundary conditions for the two
types of analyses.
As shown in Fig. 6, tensile stresses reach a maximum
at the centre of the disc (r = 26 mm) and persist over
more than half of the diameter of the specimen. In
general, the analytically determined tensile stress distri-
bution moves to the centre of the disc at a greater
increasing rate than the numerically calculated distribu-
tion. On the other hand, the increment in the tensile
stress rate of the Brazilian jaw simulation is the lowest
when approaching the centre of the disc in all simula-
tions and under all analytical values. The results indicate
that concentration of the tensile stress under Brazilian
jaw loading geometry through the centre of the disc is
lower than that using loading arcs. Details of the
obtained stress values from simulations and analytical
results are given in Table 3.
As shown in Fig. 6, the r
x
stress distribution inside the
numerical model changes dramatically because of the
Fig. 4 Numerical models for the BTS test under a standard Brazilian jaws, b 15 loading arcs, c 20 loading arcs and d 30 loading arcs
Determination of Indirect Tensile Strength of Rocks 745
1 3
heterogeneity effect (Liu 2004). For example, at the posi-
tions of crack initiation, the tensile stress drops to zero,
while at the bridges between the cracks, the tensile stress
might rise to values larger than the tensile strength of the
rock material.
Figures 7 and 8 show simulation-derived ngerprint
patterns of the maximum principal stress and minimum
principal stress induced in disc specimens subjected to the
same displacement loading, but under different loading
geometries. The stress distribution near the centre of disc
under diametral loading is quite uniform for both Brazilian
jaws and diametral loading arc cases. The stress distribution
in the y-axis direction is more uniform than that in the x-axis
direction. When the angle of the loading arc increases, the
maximum principal stress distribution (compressive stress)
becomes more complex along the loading direction com-
pared with the standard Brazilian jaw results (Fig. 7). On
the other hand, the minimum principal stress (maximum
tensile stress) is focused more at the centre of the disc as the
angle of the loading arcs increases (Fig. 8). Thus, the
obtained ngerprint patterns are of great signicance for
elucidating the crack initiation point in non-transparent rock
material under different diametral loading boundaries. The
ngerprint patterns show that the crack initiation point
Fig. 5 Progressive fracture
process induced between
a standard Brazilian jaws, b 15
loading arcs, c 20 loading arcs
and d 30 loading arcs
746 N. Erarslan et al.
1 3
moves to the centre of the disc under increasing loading arc
angles, similar to the results for the progressive fracture
process shown in Fig. 5.
To obtain a proper loading condition to determine the
most accurate indirect tensile strength of rocks, some
relationships between intrinsic material properties, such as
fracture toughness, crack initiation and propagation in a
sample, were considered. There are many suggested
methods to determine fracture toughness of rocks. Here, it
was decided to use the method of Guo et al. (1993) to
determine fracture toughness value from Brazilian tests,
since it uses Brazilian discs without a notched crack. To
determine the mode I fracture toughness (K
IC
) from Bra-
zilian tests, crack propagation in a Brazilian disc was
studied analytically and experimentally by Guo et al.
(1993). Their fracture toughness value is determined by:
K
IC
= P B U(a=R) (10)
where, K
IC
is fracture toughness (MP

m
_
), P is the mini-
mum load in a loaddisplacement curve (MN), U(a/R) is a
dimensionless stress intensity factor, B = 2=(p
3
2
R
1
2
t a), R is the radius of the disc (m), a is the half loading
arc angle (radian) and t is disc thickness (m).
Some of the fracture toughness values determined by
using the Brazilian test are close to the results obtained by
using the Chevron Bend (CB) method, which is the method
suggested by ISRM (Guo et al. 1993; Alkilicgil 2010).
According to the assumptions of Guo et al. (1993), the
crack initiation point cannot be guaranteed to locate at the
disc centre, and the stress distribution on the loading arc is
assumed to be uniform. On that basis, Wang and Xing
(1999) modied the Brazilian disc method of Guo et al.
(1993) by attening the loading ends of the disc. According
to the solution given by Wang and Xing (1999), the frac-
ture toughness value is determined by:
K
IC
=
P
min

R
_
t
U
max
(11)
where P
min
is the local minimum load determined from
recorded forcedisplacement curve, R and t are the radius
and thickness of the specimen, respectively, and U
max
is the
maximum dimensionless stress intensity factor.
The maximum mode I stress intensity factor (K
Imax
)
values calculated from the numerical analysis given by
Guo et al. (1993) and Alkilicgil (2010) are used in the
calculation of the maximum dimensionless stress intensity
factors (U
max
) for loading angles from 2a = 5 to
2a = 50 (Fig. 9). As expected, the maximum dimen-
sionless stress intensity factor decreases with increasing
loading angles.
Figure 10 shows the recorded forceloading displace-
ment curves in diametral compression of the standard
Brazilian jaws (Fig. 5a) and the three different loading arc
angles (Fig. 5b, c, d). Initially, the rock specimen behaves
as an intact elastic material with increasing load resulting
in an increasing diametral displacement until the diametral
crack initiates and propagates unstably. Subsequently, the
load reaches its local minimum corresponding to the
Table 2 Experimental and numerical results of failure loads for
Brisbane tuff
Loading mode Experimental ultimate
load (kN) (average
of 4 repeats)
Normalized
numerical ultimate
load (kN)
Standard Brazilian 19.6 17.1
15 loading arcs 14.8 18.6
20 loading arcs 19.2 19.2
30 loading arcs 22.3 21.1
Fig. 6 Horizontal stress distribution along the vertical diameter (BY)
of the disc (-r
x
: tension; -r
y
: compression)
Table 3 Experimental, numerical and analytical results of failure
loads for Brisbane tuff
Loading
mode
Experimental
ultimate load
(kN)
Theoretical
tensile stress at
the centre of the
disc (MPa)
Numerical tensile
stress at the
centre of the disc
(MPa)
Standard
Brazilian
jaws
19.60 9.20 7.91
15 loading
arcs
14.81 6.61 8.04
20 loading
arcs
19.20 8.21 8.29
30 loading
arcs
22.30 8.85 8.70
Determination of Indirect Tensile Strength of Rocks 747
1 3
maximum dimensionless stress intensity factor. After that
time, an increase in the applied load application is required
to make the crack propagate further. At that time, P
max
can
be used to calculate the splitting stress (indirect tensile
strength) of the Brazilian disc; however, determination of
fracture toughness is possible by using the local minimum
load (P
min
) during unstable crack propagation, derived
from (11).
Fracture toughness values calculated by using (11) and
the numerical simulations of the standard Brazilian jaws,
15 loading arcs, 20 loading arcs and 30 loading arcs
are: 0.85 MPa

m
_
, 0.7 MPa

m
_
, 0.81 MPa

m
_
and 1.24
MPa

m
_
, respectively. In contrast, the mode I fracture
toughness of Brisbane tuff was found experimentally to
be 1.18 MPa

m
_
(Erarslan 2011), using the Crack
Chevron Notched Brazilian Disc (CCNBD) method
suggested by ISRM (1995). The CCNBD specimens have
the same diameter and thickness as standard Brazilian
disc specimens; thus, a relationship between fracture
toughness and indirect tensile strength of rocks is both
helpful and meaningful. The numerical stress intensity
value closest to the experimental fracture toughness of
1.18 MPa

m
_
, was obtained with the 2a = 30 loading
arc simulations (1.24 MPa

m
_
). To support that result, a
well-known empirical relationship between the mode I
fracture toughness (K
IC
) and the tensile strength of rocks
(r
t
) is used (Whittaker et al. 1992). According to this
Fig. 7 Fringe patterns of maximum principal stress induced between a standard Brazilian jaws, b 15 loading arcs, c 20 loading arcs and d 30
loading arcs
Fig. 8 Fringe patterns of minimum principal stress induced between a standard Brazilian jaws, b 15 loading arcs, c 20 loading arcs and d 30
loading arcs
Fig. 9 Maximum dimensionless stress intensity factors for various
loading angles
748 N. Erarslan et al.
1 3
relationship, K
IC
can be calculated using (Whittaker
et al. 1992):
K
IC
(MPa

m
_
) = 0:27 0:107r
t
(MPa): (12)
The indirect tensile strength was found to be 8.55 MPa
by using the experimental mode I fracture toughness value
as 1.18 MPa

m
_
in (12). The closest indirect tensile stress
to this value is obtained by 2a = 20 and 2a = 30 loading
arc simulations.
4 Discussion
The main objective of this study was to analyse the tensile
stress distribution and failure of disc specimens under
loading arcs of different angles and compare them with the
results of the standard Brazilian test. This was tested
experimentally, as well as numerically and analytically.
In the experiments described herein, the disc specimens
shattered into several pieces under the standard Brazilian
jaws. It is likely that these multiple fractures were caused by
the release of energy stored in the testing machine after the
specimen split into two hemi-cylinders, making the standard
Brazilian test very sensitive to the stiffness of the testing
machine being used. In contrast, applying distributed
diametral loads by using loading arcs causes single failure
planes along the loaded diameter in disc specimens tested.
According to Jaeger and Cook (1976), shear fracture at the
contacts can be decreased if the load is applied to the cir-
cumference of the cylinder. They stated that experiments
with distributed loads over a narrow arc, usually about 15,
yielded tensile strength values little different from those
obtained with line loads and gave rise to similar diametral
extension fractures (Jaeger and Cook 1976).
The indirect tensile strength of Brisbane tuff was cal-
culated as 9.2 MPa using the ISRM (ISRM 2007) sug-
gested formula (Table 3). According to the analytical
solutions of Hondros (1959), when 2a \11 and r/R tends
to zero, the tangential stress along the vertical diameter
may be calculated using the same formula. This indicates
that the formula proposed by ISRM is not just for line
loading, but is also suitable for diametral loading over an
arc of up to 2a = 11. This outcome is also supported by
the work of Jaeger and Cook (1976). Based on experiments
with distributed loads over a narrow arc, usually up to
about 2a = 15, they reported that tensile strength varied
little from that obtained using line loads and gave similar
diametral extension fractures.
Analytically and numerically calculated tensile stresses at
the centre of the disc specimen are given in Table 3. Although
the highest failure load was found experimentally with a 30
loading arc, the highest tensile stress was found with Brazilian
tests after analytical calculations. This result shows that fur-
ther theoretical studies are needed to formulate the relation-
ship between line and diametral loading on disc specimens.
On the other hand, no large difference between obtained
tensile stresses for Brazilian and loading arc tests were found
among the numerical simulations (Table 3). The reason may
be related to material heterogeneity, since the horizontal stress
distribution changes dramatically along the vertical diameter.
To account for heterogeneity, the maximum tensile stress at
the centre of the disc was calculated by averaging values from
5 to 10 tests. Although a difculty related to heterogeneity did
arise, there was agreement between numerical and experi-
mental results, presumably because we included material
heterogeneity in the numerical model.
In numerical simulations, rock heterogeneity was found as
a stress concentrator, similar to the presence of pre-existing
cracks under loading. In front of the crack propagation
direction, if the strength of the elements is not too high, the
crack propagates in a straight manner, which may be con-
sidered as the trans-granular failures observed in experiments.
On the other hand, if the strength of the elements is very high,
the crackwill propagate aroundelements, anda tortuous crack
propagation path is observed, which can be considered as the
inter-granular failures observed in experiments.
Hondross (1959) solutions indicate that the simple Bra-
zilian formula for principal tensile stress at the centre of the
Fig. 10 Loaddisplacement curves by a standard Brazilian jaws and
b loading arcs
Determination of Indirect Tensile Strength of Rocks 749
1 3
sample is in error byless than2%for contact arcs of upto 15.
Jaeger and Hoskins (1966) indicate that the ring stress con-
centration factors for line loads differ from those for loads
distributed over 15 arcs by an error of less than 2% (Jaeger
and Hoskins 1966; Mellor and Hawkes 1971). On that basis,
it is concluded that 15 is an acceptable upper limit for
analytical solution agreements between line and diametral
distributed loading. However, although the highest failure
load was obtained experimentally, the tensile stress for 30
loading arcs was smaller than that given for line loading by
analytical solutions. This lack of agreement indicates that a
detailed analytical solution is needed to determine the
applicability of Hondross (1959) equations for calculations
of tensile stresses along the loading diameter of disc.
According to the Grifth criterion (Grifth 1920) used for
failure in solid discs, the exact centre of the disc is the only
point at which the conditions for tensile failure at a value
equal to the uniaxial strength are met. This result is also
indicated by analytical solutions given by Hondros (1959).
Fairhurst (1964) generalized the Grifth criterion to account
for variation of the compression/tension strength ratio
(n) from the theoretical value, n = 8, so that the conditions
for failure according to the generalized Grifth criterion
become available. His results indicate that to assure fracture
initiation near the centre of a homogeneous specimen, it is
necessary to spread the applied load over an appreciable
contact arc (20 or more). He suggests that with a narrow
contact strip (a = 5), there will be a pronounced tendency
for off-centre fracture initiation in rocks that have a low
compression/tension strength ratio. He also showed that if
the compression/tension strength ratio is relatively small and
the contact strip is narrow, there will be a systematic
underestimation of tensile strength, if the fracture occurs
near the centre of the specimen. Our experimental and
numerical results support Fairhursts results (1964).
Guo et al. (1993) indicate that the standard Brazilian test
has some disadvantages: crack initiation and propagation
cannot be guaranteed to be located at the centre, arc
loading is hard to apply and stress distribution on the
loading arc cannot be assumed to be uniform. By consid-
ering these disadvantages, a modied Guo et al. (1993)
method was used in fracture toughness calculations by
using a attened Brazilian disc method (Wang and Xing
1999). However, extensive analytical, numerical and
experimental validation of the method is needed.
5 Conclusions
Indirect tensile tests, which induce non-uniform stress
elds controlled partly by the properties of the rock
material and loading geometry, can never fully substitute
for a direct uniaxial tensile test. However, a Brazilian disc
under diametral compression is capable of giving a very
good measure of uniaxial tensile strength for Grifth-type
and/or heterogeneous materials when it is carefully per-
formed, with special attention paid to control of contact
stresses. In reality, there is no perfect source creating direct
tensile stresses for rock materials; thus, the use of indi-
rectly produced tensile stresses and indirect tension tests
are much more applicable to rock mechanics applications.
From the results of the experimental studies described
herein using Brisbane tuff disc specimens, it was concluded
that indirect tensile strength of rocks is strongly dependent
on the type of diametral loading. A general increase in
failure load was obtained with an increasing angle of
loading arc for all disc specimens tested. The maximum
recorded failure load, corresponding to the nest and most
central-located crack, was obtained with 2a = 30 loading
arcs. The highest failure load standard deviation was
obtained with standard Brazilian jaws, which resulted in
catastrophic specimen failure.
Numerical simulations show that rock can be specied
by using characteristic parameters. Numerical simulations
indicated that the maximum tensile stress appears at the
centre of the disc under 30 loading arcs, while the lowest
maximum tensile stress appears at the centre of the disc
under standard Brazilian jaws.
The RFPA numerical model successfully simulated rock
fracturing in Brazilian disc specimen under line and
diametral loading. Since fracture formation and the
mechanism of fracture propagation were monitored as a
progressive process with R-T
2D
code, it was possible to
analyse crack initiation and propagation as a continuous
process in a Brazilian disc under various diametral loading.
Based on the agreement between experimental and
numerical results, the RFPA method is deemed reasonable
and applicable for research into the fracture process of
heterogeneous materials, such as rocks.
Experimentally obtained fracture toughness values of
Brisbane tuff were compared with the fracture toughness
values obtained from numerical simulations. It was con-
cluded that the best loading geometry to nd indirect tensile
strength of rock material was obtained from the 2a = 20
and 2a = 30 loading arc simulations. To support that result,
well-known empirical relationships between fracture
toughness and tensile strength of rocks were applied.
Acknowledgments Acknowledgement is made to Leighton Con-
tractors, who provided core samples of Brisbane tuff from the
CLEM7 Project, and to Professor Ted Brown AC, Les McQueen,
Mark Funkhauser and Rob Morphet of Golder Associates Pty Ltd for
their assistance and advice. The work described forms part of the rst
authors PhD research carried out at the Golder Geomechanics Centre
at The University of Queensland. The rst author was supported by an
Australian Postgraduate Award/UQRS and the Golder Geomechanics
Centre.
750 N. Erarslan et al.
1 3
References
Alkilicgil C (2010) Development of specimen geometries for mode I
fracture toughness testing with disc type rock specimens. PhD
thesis, Middle East Technical University, Ankara
Blair SC, Cook NGW (1998) Analysis of compressive fracture in rock
using statistical techniques Part I: A non-linear rule-based
model. Int J Rock Mech Min Sci 35(7):837848
Cai M, Kaiser PK (2004) Numerical simulations of the Brazilian test
and tensile strength of anisotropic rocks and rocks with pre-
existing cracks. Int J Rock Mech Min Sci 4:478483
Erarslan N (2011) Static and cyclic laboratory testing of Brisbane
rocks, The University of Queensland, Brisbane
Fairhurst C (1964) On the validity of the Brazilian test for brittle
materials. Int J Rock Mech Min Sci 1:535546
Grifth AA (1920) The phenomena of rupture and ow in solids. Phil
Trans R Soc Lond A221:163
Guo H, Aziz NI, Schimidt LC (1993) Rock fracture toughness
determination by the Brazilian test. Eng Geol 33:177188
Hondros G (1959) The evaluation of Poissons ratio and the modulus
of materials of a low tensile resistance by Brazilian (indirect
tensile) test with particular reference to concrete. Aust J Appl Sci
10:243268
Hudson JA, Brown ET, Rummel F (1972) The controlled failure of
rock discs and rings loaded in diametral compression. Int J Rock
Mech Min Sci 9(2):241244
ISRM (1995) Suggested method for determining mode I fracture
toughness using cracked chevron notched Brazilian disk
(CCNBD) specimens. Int J Rock Mech Min Sci Geomech Abstr
32(1):5764
ISRM (2007) Suggested methods for determining tensile strength of
rock materials. In: Ulusay R, Hudson JA (eds) The complete
ISRM suggested methods for rock characterization, testing and
monitoring: 19742006, pp 177184
Jaeger JC, Cook NGW (1976) Fundamentals of rock mechanics.
Chapman and Hall, London
Jaeger JC, Hoskins ER (1966) Stresses and failure in rings of rock
loaded in diametral tension or compression. Brit J Appl Phys
17:685692
Lanaro F, Sato T, Stephansson S (2009) Microcrack modelling of
Brazilian tensile tests with the boundary element method. Int J
Rock Mech Min Sci 46:450461
Liu H (2004) Numerical modelling of the rock fragmentation process
by mechanical tools. PhD thesis, Lulea University of Technol-
ogy, Lulea, Sweden
Markides CF, Pazis DN, Kourkoulis SK (2010) Closed full-eld
solutions for stresses and displacements in the Brazilian disk
under distributed radial load. Int J Rock Mech Min Sci
47:237247
Mellor M, Hawkes I (1971) Measurement of tensile strength by
diametral compression of discs and annuli. J Eng Geol
5:173225
Schlangen E, Garboczi EJ (1997) Fracture simulations of concrete
using lattice models: computational aspects. Eng Fract Mech
57:319332
Tang CA, Fu YF, Kou SQ, Lindquist PA (1998) Numerical simulation
of loading inhomogeneous rocks. Int J Rock Mech Min Sci
Geomech Abstr 35:10011007
Van de Steen B (2001) Effect of heterogeneities and defects on the
fracture pattern in brittle rock. PhD thesis, Katholieke Univer-
siteit Leuven, Leuven, Belgium
Van de Steen B, Vervoort A, Napier JAL (2005) Observed and
simulated fracture pattern in diametrically loaded discs of rock
material. Int J Fract 131:3552
Wang QZ, Xing L (1999) Determination of fracture toughness KIC by
using the attened Brazilian disk specimen for rocks. Eng Fract
Mech 64:193201
Whittaker BN, Singh RN, Sun G (1992) Rock fracture mechanics:
principles, design and applications. Elsevier, Amsterdam
Yu Y, Zang JX, Zang J (2009) A modied Brazilian disc tension. Int J
Rock Mech Min Sci 46:421425
Determination of Indirect Tensile Strength of Rocks 751
1 3