ARTICLE IN PRESS

Computers & Geosciences 36 (2010) 959–969

Contents lists available at ScienceDirect

Computers & Geosciences

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

Estimating rock mass properties using Monte Carlo simulation:

Ankara andesites
Mehmet Sari a,n, Celal Karpuz b, Can Ayday c
a
Department of Mining Engineering, Aksaray University, Adana Road, 68100 Aksaray, Turkey
b
Department of Mining Engineering, Middle East Technical University, 06531 Ankara, Turkey
c
Aerospace Sciences Research Institute, Anadolu University, 26470 Eskisehir, Turkey

a r t i c l e in fo abstract

Article history: In the paper, a previously introduced method (Sari, 2009) is applied to the problem of estimating the
Received 19 November 2009 rock mass properties of Ankara andesites. For this purpose, appropriate closed form (parametric)
Received in revised form distributions are described for intact rock and discontinuity parameters of the Ankara andesites at three
24 February 2010
distinct weathering grades. Then, these distributions are included as inputs in the Rock Mass Rating
Accepted 26 February 2010
(RMR) classification system prepared in a spreadsheet model. A stochastic analysis is carried out to
evaluate the influence of correlations between relevant distributions on the simulated RMR values. The
Keywords: model is also used in Monte Carlo simulations to estimate the possible ranges of the Hoek–Brown
Ankara andesites strength parameters of the rock under investigation. The proposed approach provides a straightforward
Stochastic estimation
and effective assessment of the variability of the rock mass properties. Hence, a wide array of
Rock mass strength
mechanical characteristics can be adequately represented in any preliminary design consideration for a
Hoek–Brown failure criterion
Monte Carlo method given rock mass.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction mass properties, only a unique value is used, usually the average
of the investigated property. However, in a stochastic estimation,
A rock mass consists of two components: intact rock and it is possible to consider the full range of data concerning the
discontinuities, each of which has a significant effect on the rock specific random characteristic. This can be easily achieved with
mass strength and deformability. It is also known that weathering probability distributions, which give both the range of values that
significantly influences the engineering properties of rocks in-situ the variable could take and relative frequency of each value
(Karpuz and Pasamehmetoglu, 1997; Ehlen, 2002; Gurocak and within the range (Evans et al., 1993).
Kilic, 2005; Park et al., 2005). Therefore, reliable estimates of rock It is difficult to measure experimentally the variability of rock
mass strength and deformability are very important in rock mass mass mechanical properties. Rock engineers frequently utilize
characterization. While near surface workings in quarries are empirical methods to estimate rock mass properties with only
most strongly influenced by the structural conditions and the limited site data (Bieniawski, 1978; Hoek and Brown, 1980;
degree of weathering of the rock mass, the properties of rock Barton, 1983). These empirical methods require just a single value
formations vary greatly, both spatially and randomly, in the for each input characteristic and give a single output value. They
quarries, they can rarely be predicted with confidence. do not provide the probability distributions of the investigated
Since natural materials like soil and rock are inherently rock mass parameters; this is primarily due to the difficulty of
heterogeneous and variable, their nature of random character- handling the inherent uncertainties of the component variables in
istics needs to be appropriately described in preliminary design the empirical model.
investigations. Therefore, Muspratt (1972) emphasized the ex- Earlier research used statistical and probability methods to
istence of vast areas of potential application of probabilistic estimate rock strength and deformability from laboratory tests or
methods in geosciences because natural phenomena occur with determine the minimum number of specimens for laboratory
such variation that a stochastic rather than a deterministic system testing in rock mechanics (Yegulalp and Mahtab, 1983; Grasso
definition is more realistic. In the deterministic estimation of rock et al., 1992; Gill et al., 2005; Sari and Karpuz, 2006; Ruffolo and
Shakoor, 2009). Stochastic models are also commonly employed
to deal with uncertainties, due to the stochastic nature of the
n
Corresponding author: Tel.: + 90 382 2801345; fax: + 90 382 2801365.
geometry of rock masses and the variability of their mechanical
E-mail addresses: [email protected] (M. Sari), [email protected] properties (Dershowitz and Einstein, 1988; Kim and Gao, 1995a;
(C. Karpuz), [email protected] (C. Ayday). Kulatilake et al., 1997; Meyer and Einstein, 2002; Sari, 2009).

0098-3004/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cageo.2010.02.001
 ARTICLE IN PRESS
960 M. Sari et al. / Computers & Geosciences 36 (2010) 959–969

In the literature, there are a limited number of studies For the Golbasi district, Karpuz and Pasamehmetoglu (1997)
which considers only the stochastic estimation of strength and have also distinguished three types of andesites similar to those
deformability characteristics and variations of rock masses (Kim determined by Ayday (1989): (i) brownish gray-to-grayish purple,
and Gao, 1995a; 1995b; Hoek, 1998; Sari, 2009). Two papers by fine-grained glassy andesite (A-type); (ii) pale red, fine-grained
Kim and Gao (1995a, 1995b) presented a probabilistic method of andesite (B-type); and (iii) gray, fine-grained andesite (C-type). To
estimating the mechanical characteristics of a rock mass, using determine average physical and mechanical properties, Karpuz
the third type asymptotic distribution of the smallest values (1982) conducted extensive rock mechanics experiments on the
(extreme value statistics) and Monte Carlo simulation. They used distinct weathering grades of Ankara andesites collected from
the chi-square goodness-of-fit test to prove that the distribution different locations of the studied rock.
reflects the inherent variability of the basalt properties. Hoek Studies by Karpuz (1982), Ayday (1989), Doyuran et al. (1993),
(1998) applied the same method to estimate variation in the and Karpuz and Pasamehmetoglu (1997) addressed changes in
Hoek–Brown properties of a hypothetical rock mass, and assum- the frequency, length, and appearance of joints in the rock mass as
ing that all three input parameters of the criterion can be weathering progresses. This research on vertical or steeply
represented by normal distributions. The paper by Sari (2009) was dipping joints in Ankara andesite shows that with increased
based upon traditional (not extreme) statistics, as Hoek (1998) weathering, not only there are statistically significant differences
used. However, instead of arbitrarily assuming normal distribu- in mean joint spacing and mean trace length with increased
tions for input variables used in the Hoek–Brown criterion, the weathering, but also the physical appearance of joints changes as
actual probability distributions of discontinuity properties and weathering progresses.
intact rock strength parameters were obtained directly from the Doyuran et al. (1993) investigated the most appropriate
histograms of the observed frequencies of an ignimbrite rock frequency distributions of aperture, persistence, and spacing of
mass. All of the above mentioned studies about stochastic discontinuities measured in different weathering grades of
modeling have regarded input parameters as independent vari- Golbasi andesite. The results of the statistical analyses showed
ables, but this is not true. that the degree of weathering affects the type of frequency
The aim of this study is to demonstrate the use of the Monte distributions of the discontinuity parameters even in the same
Carlo (MC) method. This is the most common sampling technique rock.
used in stochastic modeling. In the MC method, the uncertainties
of the intact rock strength and discontinuity parameters are
incorporated into a spreadsheet model using statistical distribu- 3. Collection of the necessary data for stochastic estimation
tions. Furthermore, the relationships between these parameters
are also accounted for in the stochastic analysis, using a The engineering geological properties of the exposed faces of
correlation matrix. This significantly improves the current study quarries within the study area were evaluated extensively in field
over the previous study applied to a pyroclastic rock mass by the observations/measurements and laboratory tests by two of the
first author (Sari, 2009). The new MC model is basically designed co-authors (Karpuz, 1982; Ayday, 1989). In this study, Golbasi
for the probabilistic evaluation of the variations observed in the andesites are treated broadly under three distinct categories
rock mass strength and deformability properties of Ankara according to their degree of weathering and field observations as
andesites at three different weathering grades. In each case, the A-type andesite, B-type andesite, and C-type andesite.
strength and deformation characteristics of the rock mass are To quantify the variability of the investigated discontinuity
estimated by means of the Hoek–Brown procedure. Extensive properties of Ankara andesites at different weathering grades,
field investigations and laboratory experiments previously con- closed form (parametric) distributions fitted to actual observa-
ducted on the Ankara andesites by two of the co-authors (Karpuz, tions by Ayday (1989) are directly included in the stochastic
1982; Ayday, 1989) are the starting point for the present study. modeling. In the case of any specific distributions of the
mechanical properties are not described in the original work of
Karpuz (1982) and Ayday (1989), a truncated normal distribution
2. Ankara andesites is subjectively assumed for these properties of the rock. Many
authors have shown that the frequency distributions of some
Ankara andesite is widely utilized for building and pavement results of tests on the strength of rock can be well represented by
stone production due to its appearance, color, and durability. a normal distribution (Yegulalp and Mahtab, 1983; Grasso et al.,
Andesites, described by Kasapoğlu (1980), Karpuz (1982), Ayday 1992; Hoek, 1998; Hsu and Nelson, 2002; Gill et al., 2005; Sari and
(1989), Doyuran et al. (1993), Karpuz and Pasamehmetoglu Karpuz, 2006; Sari, 2009).
(1997), and Ercanoglu and Aksoy (2004), occur extensively in The mean value of the normal probability distribution
the vicinity of Ankara (Fig. 1). This material has been used since represents the best estimate of the random variable, and the
Roman times and before in building works. There are many standard deviation or coefficient of variance (COV) of this mean
andesite quarries (Fig. 2), and many important historical value represents an assessment of the uncertainty. Generally, a
structures were built from the rocks gathered from these quarries. standard deviation equivalent to 15–20% of the mean value,
Significant changes in discontinuity spacing or characteristics which represents the typical range of COV applicable for most
within the same rock type may necessitate the division of the rock natural geotechnical materials (Rethati, 1988) was assumed as a
mass into a number of structural regions. For that reason, Ayday measure of variability in this study.
(1989) subdivided the studied quarry into three regions contain-
ing different types of rocks: A-type, pinkish andesites; B-type, 3.1. Rock mass rating (RMR) classification system
dark pinkish-gray andesites; and C-type, black andesites. The
difference of color in each region is attributed to the degree of The RMR system (Bieniawski, 1989) is one of the most widely
weathering of the rock. Black andesites are un-weathered and the used rock mass classifications. This method incorporates geolo-
pinkish are the most weathered. The different degrees of andesite gical, geometric, and design or engineering parameters in arriving
weathering have also been observed by Kasapoglu (1980). Based at a quantitative measure of the rock mass quality. This
on the thin-section studies of samples of each type, the rock may engineering classification system, developed by Bieniawski
be classified as hornblende–andesite (Ayday, 1989). (1973), utilizes the following six rock mass parameters:
 ARTICLE IN PRESS
M. Sari et al. / Computers & Geosciences 36 (2010) 959–969 961

Fig. 1. Location and geologic map of study area (modified from Karpuz and Pasamehmetoglu, 1997).

Fig. 2. A general view of an Ankara andesite rock quarry.

 ARTICLE IN PRESS
962 M. Sari et al. / Computers & Geosciences 36 (2010) 959–969

(1) uniaxial compressive strength (UCS) of intact rock material, (2) 3.1.4. Joint persistence
rock quality designation (RQD), (3) spacing of discontinuities, (4) Many studies of field measurements have shown that
condition of discontinuities (aperture, persistence, roughness, the negative exponential probability density distribution is
infilling, weathering), (5) groundwater conditions, (6) orientation appropriate to represent the discontinuity trace length distribu-
of discontinuities. tion (Wallis and King, 1980; Baecher, 1983; Park and West, 2001;
Kulatilake et al., 2003). Ayday (1989) found that, in Ankara
3.1.1. Unconfined compressive strength (UCS) andesites, the type of distribution changes from gamma to
The unconfined (or uniaxial) compressive strength (UCS) of a negative exponential as the intensity of weathering increases.
rock is frequently used as one of the key inputs in most The frequency distributions given in Table 1 are assumed for joint
classification schemes and design applications for the character- persistence characterization of Ankara andesites.
ization of rock material strength. Sari and Karpuz (2006) suggest a
normal distribution for 103 UCS tests on the Ankara andesite 3.1.5. Joint aperture
samples. Ayday (1989) states that the strength values of the The apertures of real discontinuities are likely to vary widely
Golbasi andesite increases, starting from the A-type through to over the extent of the discontinuity. Clearly, the variation of
the C-type. Accordingly, the normally distributed UCS values aperture will have an influence on the shear strength of the
given in Table 1 are assumed for the three different weathering discontinuity (Barton, 1983). From outcrop mapping, the joint
grades of this rock. aperture can only be roughly estimated, through the direct
observation of joint exposed at the outcrop, according to Brown
3.1.2. Rock quality designation (RQD) (1981). Ayday (1989) recommended normal distributions for the
The RQD was developed by Deere et al. (1967) to provide a joint aperture at distinct weathering degrees of Ankara andesites
quantitative estimate of rock mass quality from drill core logs. In (Table 1).
the areas, where scan-line or area mapping can be conducted, it is
not necessary to use core recovery since a better picture of the 3.1.6. Joint roughness
rock mass can be obtained from these measurements. For scan- A classification of discontinuity roughness has been suggested
line data, an average joint spacing can be obtained (number of by Brown (1981). Roughness is a potentially important compo-
features divided by traverse length). Bieniawski (1989) has linked nent of strength and therefore, a qualitative description of joint
an average joint spacing to RQD relying on previous work by Priest roughness was first accomplished for each different weathering
and Hudson (1976). The RQD in this study was estimated from the grades of Ankara andesites based on field observations (Table 1).
average joint spacing based on the following equation: Then these qualitative descriptions are converted to quantitative
values using the relevant RMR score in the stochastic model.
RQD ¼ 100:e0:1l ð0:1l þ1Þ ð1Þ
where l is the average number of discontinuities per meter, l ¼1/ 3.1.7. Infilling
(mean joint spacing). The distribution of spacing must be Filling is the term for material that separates the adjacent rock
negatively exponential, if the theoretical RQD is to be applied to walls of discontinuities. The joint can be clean or filled with
a particular rock. weathered products and deposits, ranging from sandy particles to
swelling clays. Again, a qualitative description of joint infilling
3.1.3. Spacing of discontinuities was accomplished for each distinct weathering grades of Ankara
On the data collected for the Ankara andesites by Ayday andesites based on field observations (Table 1). Then, these
(1989), chi-square goodness-of-fit tests were performed for qualitative descriptions are converted to quantitative values using
normal, gamma, lognormal, and negative exponential distribu- the relevant RMR score in the stochastic model.
tions; the last two being the distribution models commonly used
for spacing evaluation (Priest and Hudson, 1976; Hudson and 3.1.8. Weathering
Priest, 1983). This is because those theoretical distributions are When the joint surface is weathered, it often shows a change in
bounded at zero and are skewed to the right, those characteristics color and appearance. Frequently, weathered products, such as
are similar to the properties of the spacing distribution. The grain particles may remain inside the joint. According to the thin-
negative exponential probability distribution (Table 1) was section results of the andesites (Ayday, 1989), A-type is
chosen as an appropriate parametric distribution model to moderately stained, B-type is slightly stained and C-type is
represent the discontinuity spacing in Ankara andesites. virtually no staining. It was also observed that the discontinuities

Table 1
Intact rock and discontinuity properties of Ankara andesites.

Parameter A_type B_type C_type Source

UCS (MPa) Normal 53.07 10.6 n

Normal 68.0 7 10.2 Normal 128.0 7 12.8
Joint spacing (cm) N. Exponential 54.4 743.9 N. Exponential 125.9 7 97.2 N. Exponential 73.0 7 52.9
Joint persistence (m) N. Exponential 2.92 72.1 Gamma 3.88 7 2.68 Gamma 1.86 7 1.0
Joint aperture (mm) Normal 5.087 0.98 Normal 4.05 7 1.48 Normal 4.72 7 1.10
Roughness Slightly rough Slightly rough Very rough
Karpuz (1982) and Ayday (1989)
Weathering Moderately weathered Slightly weathered Unweathered
Infilling Soft fillings Hard fillings Unfilled
Groundwater Dry Dry Dry
Ei (GPa) Normal 22.87 4.6 Normal 28.1 7 4.2 Normal 45.1 7 4.5
mi Normal 4.17 1.0 Normal 7.0 71.4 Normal 10.5 71.6
D Normal 0.77 0.1 Normal 0.5 70.1 Normal 0.3 70.1

n
Mean 7Std. Dev.
 ARTICLE IN PRESS
M. Sari et al. / Computers & Geosciences 36 (2010) 959–969 963

were generally in dry condition. Table 1 describes the weathering The uniaxial compressive and tensile strengths of the jointed
grades for different types of Ankara andesites. rock masses are calculated from the following equations as
suggested by Hoek et al. (2002)
3.2. The Hoek–Brown criterion sc ðMPaÞ ¼ sci :sa ð6Þ

The most recent version of the generalized Hoek–Brown failure s:sci

st ðMPaÞ ¼  ð7Þ
criterion (Hoek et al., 2002) is employed to estimate the rock mass mb
properties of the weathered Ankara andesites. In order to use the Hoek et al. (2005) re-examined existing empirical methods for
Hoek–Brown (HB) criterion to estimate the strength and deform- estimating rock mass deformation modulus and concluded that
ability of jointed rock masses, the following three properties of none of these methods provided reliable estimates over the whole
the rock mass need to be described: the uniaxial compressive range of rock mass conditions that are frequently encountered. A
strength sci of the intact rock pieces in the rock mass, the value of related modification for estimating the deformation modulus of
the Hoek–Brown constant mi for these intact rock pieces, and the rock masses was created by Hoek and Diederichs (2006). Using a
value of the geological strength index (GSI) of the rock mass. commercial curve fitting software, they derived the following
At failure, the generalized HB criterion (Hoek et al., 2002) best-fit equation for a new set of reliable rock mass deformation
relates the maximum effective stress, s1, to the minimum modulus data from China and Taiwan.
effective stress, s3, through the functional relationship  
1D=2
 a Erm ¼ Ei 0:02 þ ð8Þ
s 1þ e ðð60 þ 15DGSIÞ=11Þ
s1 ¼ s3 þ sci mb 3 þ s ð2Þ
sci in which Ei is the elastic modulus of intact rock. The Ei (GPa)
where mb extrapolates the intact rock constant mi to the rock values found by Karpuz (1982) are given in Table 1 for Ankara
mass andesites at different weathering grades.
 
GSI100
mb ¼ mi exp ð3Þ 3.2.1. Geological strength index (GSI)
2814D
In the HB criterion, the GSI is the most important input
sci is the uniaxial compressive strength of the intact rock and s parameter in terms of the relationship between the strength and
and a are constants that depend upon the rock mass’s character- deformation properties determined in the laboratory and those
istics assigned to the field scale rock mass (Hoek, 1998). In earlier
 
GSI100 versions of this criterion, Bieniawski’s RMR was used for this
s ¼ exp ð4Þ scaling process and the use of different rating scores for each
93D
parameter made the RMR more suitable for numerical computa-
1 1 GSI=15 20=3 tions. Therefore, it is possible to directly obtain a frequency
a¼ þ ðe e Þ ð5Þ
2 6 distribution from the rock mass rating scores that is crucial for the
The GSI introduced by Hoek (1994) and Hoek et al. (1995) purpose of simulation in a probabilistic analysis. On the other hand,
provides a number which, when combined with the intact rock GSI can only be determined by field observations of the blocks and
properties, can be used to estimate the reduction in rock mass discontinuity surface conditions. The descriptive and largely
strength for different geological conditions. GSI takes into account qualitative nature of the GSI tables does not allow a quantitative
the geometrical shape of the intact rock fragments as well as the definition for the variability of the rock mass properties. As a result,
condition of the joint surfaces. Angular rock pieces with clean, the type of parametric distribution that represents the rock mass
rough discontinuity surfaces will result in a much stronger rock characteristics can only be made in a subjective manner, such as
mass than one which contain rounded particles surrounded by assuming a normal distribution as Hoek (1998) did.
weathered and altered material.
The parameter D is a factor that quantifies the disturbance of rock 3.2.2. Material constant mi
masses. It varies from 0 (undisturbed) to 1 (disturbed) depending on The empirical criterion formulated by Hoek and Brown (1997)
the amount of stress relief, weathering, and blast damage resulting allows the use of an approximate value of material constant mi for
from nearby excavations. In order to allow some variations in the a particular rock. The constant mi is changed with the type of rock,
values of this parameter in the stochastic analysis and to treat it as a its mineral composition, interlocking of grains, grain size, etc.
random variable of the HB model, a narrow normal distribution is Karpuz (1982) recommended different material constants mi for
assigned for distinct grades of this rock (Table 1). the distinct weathering grades of Ankara andesite samples based

Table 2
Correlation matrix incorporated into MC simulation model.

UCS RQD S P A R I W GW mi Ei D

UCS 1.0 UCS—uniaxial compr. strength

RQD 0.9 1.0 RQD-rock quality designation
S 0.9 0.9 1.0 S-spacing
P  0.6  0.6  0.6 1.0 P-persistenc
A  0.6  0.3  0.6 0.6 1.0 A-aperture
R 0.6 0.6 0.3  0.3  0.6 1.0 R-roughness
I  0.6  0.3  0.3 0.6 0.9  0.3 1.0 I-infilling
W  0.9  0.6  0.9 0.9 0.9  0.6 0.9 1.0 W-weathering
GW  0.6  0.6  0.6 0.6 0.9  0.3  0.3 0.6 1.0 GW-groundwater
mi 0.6 0.6 0.6  0.3  0.3  0.3  0.3  0.6  0.3 1.0 mi-material constant
Ei 0.9 0.6 0.6  0.6  0.9 0.6  0.6  0.9  0.3 0.6 1.0 Ei-elasticity modulus
D  0.6  0.6  0.6 0.6 0.9  0.6 0.6 0.9 0.6  0.6  0.9 1.0 D-disturbance factor
 ARTICLE IN PRESS
964 M. Sari et al. / Computers & Geosciences 36 (2010) 959–969

on triaxial strength data. Accordingly, a normal distribution was achieved as in the use of parametric distributions. Also, since a
assumed with different mean values and different COVs (15%, 20%, normal distribution goes from plus to minus infinity, it should be
and 25% of the mean) for distinct types of this rock (Table 1). used with truncation to describe variables belonging to natural
materials.
The core of the MC method is a random-number generator. A
4. Stochastic estimation of rock mass strength parameters random number is a uniformly distributed random variable over
the interval [0,1]. Random numbers can be generated using the
4.1. Monte Carlo (MC) simulation Excel RAND() function, which produces a new random number
every time the spreadsheet is recalculated. The next step is to
The most common sampling technique used in stochastic transform uniformly distributed random numbers into a non-
analysis is the Monte Carlo (MC) method. The term stochastic is uniform distribution. For a more comprehensive description of
used when there is a random component to a model, e.g. values statistical distributions and how the MC simulation works, the
for the input variables are sampled from a statistical distribution. reader is directed to Ripley (1987), Evans et al. (1993), Johnson
The technique uses random or pseudo-random numbers to et al. (1993, 1994, 1995), Gentle (1998), Law and Kelton (2000),
sample from a probability distribution. Any given sample may and Vose (2000).
fall anywhere within the range of the input distribution but, Dependencies do exist between variables in a system being
samples are more likely to be selected from the regions of the modeled and they will often need to be correlated to ensure that
distribution that have higher probabilities of occurrence. With only meaningful scenarios are generated for each iteration of the
sufficient iterations, MC sampling ‘‘recreates’’ input distributions model. One problem in the MC simulation is the difficulty of
and with the aid of a computer, hundreds or thousands of ‘‘what- incorporating the covariance between input variables. The
if’’ scenarios can be conducted, modeling most combinations of covariance between random parameters plays an important role
input parameters and quantifying the statistical distributions of in probabilistic analysis, although there are limited conclusions
outcomes (Vose, 2000). This estimation is reasonably accurate from the research involving the accurate evaluation of covariance
only if the number of simulations is very large. The advantage of between random parameters in a rock mass (Kim and Gao, 1995a;
this method is that the complete probability distribution for the Hoek, 1998; Sari, 2009).
mechanical properties can be successfully obtained if the For instance, a weathered discontinuity could be expected to
probability density function of input parameters are accurately have wider openings and be completely filled with weathering
assessed and the interdependence between these parameters are products. A stronger rock would be expected to have a shorter
quantified correctly. trace length and narrower opening. A closer spacing would result
When using the MC simulation, it is not necessary to first in a lower RQD and shorter persistence. Pairs of input variables
describe the input variables with a closed form distribution, these such as UCS and mi, RQD and mean joint spacing, and GSI and
can be directly sampled from the frequency distributions mean joint spacing, certainly have interdependence in nature too.
(histograms). In fact, in this way inaccuracies can be avoided, Simulation software allows the incorporation of the rank order
including the normal approximation. Clearly, this is not as easily correlations observed between input parameters as a correlation

Fig.3. Schematic diagram of stochastic modeling methodology applied to an Ankara andesite.

 ARTICLE IN PRESS
M. Sari et al. / Computers & Geosciences 36 (2010) 959–969 965

matrix in the spreadsheet models. However, since it is challenging i. The data for discontinuity parameters and intact rock strength
to establish the exact nature of relationships observed between characterization was compiled from the studies of Karpuz
rock material and discontinuity parameters only subjective (1982) and Ayday (1989).
estimates can be made. ii. Closed form distribution functions, which represent both the
In a previous study (Sari, 2009), intact rock and joint probability and range of values that would be expected in
parameters were considered to be probabilistic in nature; the field and laboratory were defined for each parameter of
however, they were also assumed to be independent. For this the RMR classification scheme.
study, to overcome the problem of selecting a representative iii. In the third step, the stochastic assessment of rock mass
rating of various parameters, a correlation matrix (Table 2) is properties was accomplished using the probability density
constructed intuitively for use in MC simulations. The correlation distributions from the previous step. MC simulations were
matrix is a lower triangular matrix with ones along the diagonal. executed to obtain a statistical representation of the RMR and
The construction of this matrix is solely based on the direction of RQD in the spreadsheet model.
the relationships and relative weights of the rock mass iv. UCS, GSI, Ei, mi, and D were defined as parametric distributions
parameters considered in the RMR classification scheme and and included in strength and deformability equations to
engineering judgment of the authors. To the authors’ knowledge, estimate the mb, s, and a parameters of the Hoek–Brown
the construction of such a matrix is probably one of the first failure criterion for the rock mass.
attempts in the field of rock engineering. v. By running the MC simulation for the rock mass properties
During the construction of the matrix, the direction of the described as statistical distributions in previous steps, the
relationship observed between input parameters is first defined as mean, standard deviation, and confidence intervals (range) of
negative or positive. Then, the strength of association anticipated the uniaxial compressive strength (Sigc), uniaxial tensile
between these parameters is quantified using a descriptive scale. strength (Sigt), and deformation modulus (Erm) of the rock
There are five classes into which the relationships can be assigned mass were then achieved for each distinct type of the Ankara
ranging from 0 to 1, corresponding to ‘‘none’’ (0.0), ‘‘weak’’ (0.3), andesites.
‘‘moderate’’ (0.6), ‘‘strong’’ (0.9), and ‘‘perfect’’ (1.0).
Using Excel add-in @RISK, that processes these calculations in
standard spreadsheet packages; it is possible to explicitly include
the uncertainty present in inputs to generate outputs that show
4.2. Results and discussion all probable alternatives (Palisade, 2000). The program provides a
simple and intuitive implementation of an MC simulation
Fig. 3 is a simple illustration of the stochastic modeling together with prepared spreadsheet models. For this purpose,
methodology applied in this study. In the stochastic estimation of different spreadsheet models are prepared separately for three
rock mass strength and deformability properties of weathered distinct types of Ankara andesite. These models include the
Ankara andesites, the following steps were taken into generalized Hoek–Brown failure criterion formulae and allows
consideration: users to easily obtain reliable estimates of rock mass strength

Fig. 4. Spreadsheet model used in MC simulation for A-type andesite.

 ARTICLE IN PRESS
966 M. Sari et al. / Computers & Geosciences 36 (2010) 959–969

properties. One of the models prepared for the A-type andesite is related cells and they give a total of 57 RMR value. When a new
presented in Fig. 4 with related cell formulas. suitable value in the relevant cells is generated randomly from the
In the spreadsheet model, cells B2:B6 include formulae for the specified parametric distributions the spreadsheet recalculates
probability distributions defined previously for rock mass dis- many different combinations of these ratings according to the
continuity and strength parameters. Cells C2:C10 contain the correlation matrix given in Table 2. Cells, C2:C10, which contain
appropriate RMR scores given in Bieniawski’s (1989) classification the input parameters for the RMR classification system will
scheme. Many different combinations of these ratings are automatically assign the scores according to the appropriate
simulated to produce a distribution of RMR in cell C11. In Fig. 4, range of the RMR values given in the 1989 version of Bieniawski’s
only average values of the discontinuity and intact rock strength classification scheme. These ratings can then be combined to
parameters are rated according to the RMR classification in the acquire a final RMR frequency distribution with an average value

40 45
35 40
A
30 B C 35
A
Frequency (%)

Frequency (%)
25 30
25 B
20
20
15
15 C
10 10
5 5
0 0
20 30 40 50 60 70 80 90 100 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
GSI (%) mb
55 80
50
C 70
45 A
40 60
Frequency (%)

Frequency (%)

35 50
30 B
40
25
20 A 30
15 20 B
10 C
5 10
0 0
0.5 0.502 0.504 0.506 0.508 0.51 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
a s
55 45
50 40
A
45 A 35
40
Frequency (%)

Frequency (%)

30
35
30 25 B
25 B 20
20 15
15 C
C 10
10
5 5
0 0
0 10 20 30 40 50 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
Sigc (MPa) Sigt (MPa)
40
35 A
30
Frequency (%)

25 B
20
C
15
10
5
0
0 5 10 15 20 25 30 35 40 45
Erm (GPa)

Fig. 5. Frequency distributions of rock mass parameters simulated by spreadsheet model.

 ARTICLE IN PRESS
M. Sari et al. / Computers & Geosciences 36 (2010) 959–969 967

of 60.9 and a standard deviation of 6.34, which typically in-situ stress should not be thought to modify the classification
resembles a normal distribution. index, because they are considered in the design stage of rock
Since the 1989 version of the Bieniawski’s RMR classification structures. For example, when RMR is used for rock mass strength
system is applied for the calculation of ratings in the cells, cell B13 estimation, the rock mass should be assumed to be completely
calculates the GSI as equal to RMR89-5 as suggested by Hoek and dry and have a very favorable discontinuity orientation to avoid
Brown (1997), where RMR89 has the groundwater rating set to 15 double counting (Hoek et al., 1995, 2002).
(dry) and the adjustment for joint orientation is set to 0 (very For each model simulation, 1000 iterations are performed
favorable). It should be noted that when a rock mass classification using the Latin Hypercube sampling. This method uses a stratified
system is used for estimating rock mass strength (and deforma- sampling to better resemble the resulting probability distribution
tion properties), only the inherent parameters of the intact rock with fewer iterations compared to MC sampling (Palisade, 2000).
and discontinuities need to be considered for evaluation of the Every run of the simulation yields 1000 different possible
classification index. Other parameters such as groundwater and combinations of input variables that are sampled randomly from
the previously defined parametric distributions taking into
account the correlations provided in Table 2. Once the values of
Table 3 the GSI in cell B13 have been estimated from the previous steps,
@Risk inputs summary statistics. the truncated normal distributions located in cells B2 for UCS, B14
for mi and B16 for D can be used to estimate the Hoek–Brown rock
Standard
Input name Minimum Maximum Mean 5% 95% mass parameters mb, s, and a values in cells B18, B19, and B20,
deviation
respectively. Then, possible combinations of the Hoek–Brown
UCS_A 16.53 104.58 53.02 10.70 35.50 70.42 failure parameters simulated in the previous step are entered into
UCS_B 31.34 109.99 68.01 10.26 51.17 84.74 the equations contained in cells B22, B23, and B24 for uniaxial
UCS_C 83.70 198.93 128.03 12.96 106.91 149.04 compressive strength (Sigc), uniaxial tensile strength (Sigt), and
Spacing_A 0.03 6.95 0.52 0.52 0.06 1.49
Spacing_B 0.09 16.35 1.27 1.25 0.15 3.59
deformation modulus (Erm) to obtain the final rock mass strength
Spacing_C 0.06 8.87 0.73 0.70 0.10 2.05 and deformability distributions of the Ankara andesites.
Persistence_A 0.50 17.80 3.21 2.51 0.68 8.21 All the frequency distributions of the output variables
Persistence_B 0.44 28.10 3.85 2.66 0.92 8.84 simulated by stochastic modeling are given in Fig. 5. It is
Persistence_C 0.81 9.77 1.98 1.23 0.21 4.15
important to note that stochastically estimated rock mass
Aperture_A 2.02 9.86 5.08 0.99 3.47 6.69
Aperture_B 0.10 11.77 4.07 1.48 1.65 6.49 mechanical properties of the Ankara andesites never have a
Aperture_C 1.01 9.91 4.72 1.11 2.90 6.53 single value, in fact, all exhibit significant variations between
Ei_A 8.59 44.42 22.81 4.63 15.22 30.33 some specified intervals. Information obtained from the MC
Ei_B 14.73 47.36 28.11 4.23 21.17 35.01 simulation runs is still relevant, and gives further insight into
Ei_C 30.08 64.79 45.11 4.53 37.66 52.50
D_A 0.41 1.00 0.70 0.10 0.54 0.86
variability of the rock mass strength and deformability properties
D_B 0.22 0.80 0.50 0.10 0.34 0.66 estimated from the HB criterion. The rock mass strength and
D_C 0.01 0.60 0.30 0.10 0.14 0.46 deformability characteristics of the Ankara andesites are generally
mi_A 0.84 8.87 4.10 1.01 2.45 5.74 seen to resemble asymptotic distributions skewing to larger
mi_B 2.02 13.59 7.00 1.41 4.70 9.29
values, a similar trend that has been observed in intact rock
mi_C 5.28 19.24 10.50 1.62 7.86 13.13
materials.

Table 4
@Risk outputs summary statistics.

Output name Minimum Maximum Mean Standard deviation 5% 95%

RQD_A 20.68 99.99 89.62 16.38 48.64 99.78

RQD_B 69.79 100.00 97.33 5.11 85.40 99.96
RQD_C 51.93 99.99 94.32 9.25 71.79 99.88
RMR_A 32 77 60.9 8.56 42 72
RMR_B 51 86 70.9 6.08 59 79
RMR_C 62 91 79.9 5.73 69 89
GSI_A 27 72 55.9 8.56 37 67
GSI_B 46 81 65.9 6.08 54 74
GSI_C 57 86 74.9 5.73 64 84
mb_A 0.03 1.62 0.42 0.25 0.09 0.91
mb_B 0.24 4.17 1.44 0.57 0.59 2.48
mb_C 1.06 7.51 3.76 1.14 1.98 5.76
s_A 0.000 0.027 0.003 0.004 0.000 0.010
s_B 0.000 0.094 0.015 0.013 0.002 0.039
s_C 0.003 0.210 0.058 0.041 0.010 0.140
a_A 0.501 0.527 0.505 0.004 0.502 0.514
a_B 0.501 0.508 0.502 0.001 0.501 0.504
a_C 0.500 0.504 0.501 0.001 0.500 0.502
Sigc_A 0.04 15.20 2.85 2.13 0.31 6.63
Sigc_B 0.53 33.66 7.99 4.43 2.12 16.69
Sigc_C 4.82 91.14 29.75 13.21 11.21 51.53
Sigt_A  3.04  0.01  0.38 0.33  1.01  0.03
Sigt_B  4.51  0.03  0.67 0.49  1.59  0.14
Sigt_C  7.23  0.17  1.88 1.09  3.91  0.50
Erm_A 0.28 18.72 4.47 2.99 0.76 10.06
Erm_B 1.12 33.24 11.02 5.00 3.79 19.90
Erm_C 6.32 55.02 28.65 8.21 14.95 41.60
 ARTICLE IN PRESS
968 M. Sari et al. / Computers & Geosciences 36 (2010) 959–969

The summary statistics computed from the @RISK simulation 100 Intact
for the input and output variables are listed in Tables 3 and 4,
respectively. For the A-type andesite, the RMR ranges 32–77 with
a mean value of 60.9. For the B-type andesite, it ranges 51–86
with a mean value of 70.9. For the C-type andesite, it ranges 80
62–91 with a mean value of 79.9. The model simulates the mean
values for Sigc of A-, B-, and C-type andesites as 2.85, 7.99, and

Shear stress,τ (MPa)

29.75 MPa, respectively. It estimates the average values for Sigt of 60
A-, B-, and C-type andesites as  0.38, 0.67, and  1.88 MPa, C
respectively. The Erm averages of A-, B-, and C-type andesites are
found as 4.47, 11.02, and 28.65 GPa, respectively.
40
For different weathering stages of the Ankara andesites, the
failure envelopes in the principal stress space (Fig. 6) and shear-
normal stress space (Fig. 7) were obtained as the final step in the
study. The strength envelopes are drawn by entering the mean 20 B
values for the parameters of each rock mass previously estimated
from the stochastic analysis. Necessary calculations for this are A
carried out by solving the Hoek–Brown empirical equations 0
proposed for intact rock and rock masses. It is obvious in both -20 0 20 40 60 80
graphs that as the weathering intensity of Ankara andesites Normal stress,σn(MPa)
increases, the relative change in the strength envelopes becomes
smaller. Fig. 7. Generalized Hoek–Brown failure envelopes for Ankara andesites in sn  t
space.

5. Conclusions  MC simulated results of strength and deformability parameters

were found to show distributions skewing to larger values.
Stochastic modeling is a technique where an existing model of  interdependence between rock mass parameters has to be
a system is used to quantify the random variation that is expected incorporated into any stochastic model to obtain meaningful
in the system under investigation. In this study, this technique has combinations.
been used to estimate the variation in the rock mass strength and  values of the rock mass strength and deformability that are
deformability characteristics of weathered Ankara andesite. The found will most likely more closely characterize the complex
results showed that: geological environments encountered in the real world.

 variations in the mechanical properties of the rock mass could Overall, the rock mass properties affected by the random nature
be easily estimated, if appropriate techniques were applied of discontinuity characteristics and intact rock properties, which
that took into account the frequency distribution of the input are widely scattered and variable, cannot be sufficiently repre-
parameters. sented by a single value for each input characteristic and a single
output value. Therefore, it is highly recommended that the
probabilistic analysis should be applied, particularly in cases
where there is significant scatter in the data of discontinuity and
intact rock parameters.
350 Note to readers: the interested readers who want to obtain the
Intact original data associated with this paper should contact the
authors directly: [email protected] (C. Karpuz) and cayday@a-
300
nadolu.edu.tr (C. Ayday). However, some valuable raw data can
also be obtained from open literature, such as Doyuran et al.
250 (1993) and Karpuz and Pasamehmetoglu (1997).
Major principal stress,σ1 (MPa)

200 Acknowledgements
C
The authors are grateful to Dr. Ali Shafiei from University of
150 Waterloo and an anonymous referee for critically reviewing the
manuscript, and improving the quality of the paper.

100 B References

A Ayday, C., 1989. Statistical analysis of discontinuity parameters of Gölbas- ı (Ankara)

50
andesites, Süpren (Eskis-ehir) marble, and Porsuk Dam (Eskis-ehir) peridotite.
Ph.D. Dissertation, Middle East Technical University, Ankara, 186pp.
Baecher, G.B., 1983. Statistical analysis of rock mass fracturing. Journal of
0 Mathematical Geology 15, 329–347.
-15 -5 5 15 25 35 45 Barton, N., 1983. Application of Q-system and index tests to estimate shear
strength and deformability of rock masses. In: Proceedings of the International
Minor principal stress, σ3 (MPa) Symposium on Engineering Geology and Underground Construction, Lisbon,
Portugal, pp. 51–70.
Fig. 6. Generalized Hoek–Brown failure envelopes for Ankara andesites in s1  s3 Bieniawski, Z.T., 1973. Engineering classification of jointed rock masses. Transac-
space. tions of the South African Institution of Civil Engineers 15 (12), 335–344.
 ARTICLE IN PRESS
M. Sari et al. / Computers & Geosciences 36 (2010) 959–969 969

Bieniawski, Z.T., 1978. Determining rock mass deformability: experience from case Johnson, N.L., Kotz, K., Balakrishnan, N., 1995. Continuous Univariate Distributions,
histories. International Journal of Rock Mechanics and Mining Sciences 15, vol. 2. John Wiley & Sons, New York 719pp.
237–247. Johnson, N.L., Kotz, K., Kemp, A.D., 1993. Univariate Discrete Distributions. John
Bieniawski, Z.T., 1989. Engineering Rock Mass Classifications. John Wiley & Sons, Wiley & Sons, New York 565pp.
New York 251pp. Karpuz, C., 1982. Rock mechanics characteristics of Ankara andesites in relation to
Brown, E.T. (Ed.), 1981. Testing and Monitoring: ISRM Suggested Methods. their degree of weathering. Ph.D. Dissertation, Middle East Technical
Pergamon, London 211pp. University, Ankara, 157pp.
Deere, D.U., Hendron, A.J., Patton, F.D., Cording, E.J., 1967. Design of surface and Karpuz, C., Pasamehmetoglu, A.G., 1997. Field characterization of weathered
near surface construction in rock. In: Fairhurst, C. (Ed.), Failure and Breakage of Ankara andesites. Engineering Geology 46, 1–17.
Rock, Society of Mining Engineers of AIME, pp. 237–302. Kasapoglu, K.E., 1980. Engineering Geological Characteristics of Foundation of
Dershowitz, W.S., Einstein, H.H., 1988. Characterizing rock joint geometry with City of Ankara. Hacettepe University, Geological Engineering Department,
joint system models. Rock Mechanics and Rock Engineering 21, 21–51. Ankara.
Doyuran, V., Ayday, C., Karahanoglu, N., 1993. Statistical analyses of discontinuity Kim, K., Gao, H., 1995a. Probabilistic approaches to estimating variation in the
parameters of Golbasi (Ankara) andesites, Supren (Eskisehir) marble, and mechanical properties of rock masses. International Journal of Rock Mechanics
Porsuk dam (Eskisehir) peridotite in Turkey. Bulletin of the International and Mining Sciences 32, 111–120.
Association of Engineering Geology 48, 15–31. Kim, K., Gao., H, 1995b. Probabilistic site characterization strategy for natural variability
Ehlen, J., 2002. Some effects of weathering on joints in granitic rocks. Catena 49, assessment of rock mass properties. In: Proceedings of the 10th Conference on
91–109. Engineering Mechanics, ASCE, University of Colorado, Part 1, pp. 21–24.
Ercanoglu, M., Aksoy, H., 2004. Potential instability map for rock slopes at Ankara Kulatilake, P.H.S.W., Fiedler, R., Panda, B.B., 1997. Box fractal dimension as a
Castle and vicinity. Yerbilimleri 29, 97–114 [in Turkish]. measure of statistical homogeneity of jointed rock masses. Engineering
Evans, E., Hastings, N., Peacock, B., 1993. Statistical Distributions second ed. John Geology 48, 217–229.
Wiley & Sons, New York 170pp. Kulatilake, P.H.S.W., Um, J., Wang, M., Escandon, R.F., Varvaiz, J., 2003. Stochastic
Gentle, J.E., 1998. Random Number Generation and Monte Carlo Methods fracture geometry modeling in 3-D including validations for a part of
(Statistics and Computing). Springer, New York 398pp. Arrowhead East Tunnel, California, USA. Engineering Geology 70, 131–155.
Gill, D.E., Corthesy, R., Leite, M.H., 2005. Determining the minimal number of specimens Law, A.M., Kelton, W.D., 2000. Simulation Modeling and Analysis third ed.
for laboratory testing of rock properties. Engineering Geology 78, 29–51. McGraw-Hill, New York 784pp.
Grasso, P., Xu, S., Mahtab, M.A., 1992. Problems and promises of index testing of Meyer, T., Einstein, H.H., 2002. Geologic stochastic modeling and connectivity
rocks. In: Tillerson, Wawersik (Ed.), In: Proceedings of the 33rd U.S. Symposium assessment of fracture systems in the Boston area. Rock Mechanics and Rock
on Rock Mechanics (Santa Fe, NM). Balkema, Rotterdam, pp. 879–888. Engineering 35, 23–44.
Gurocak, Z., Kilic, R., 2005. Effect of weathering on the geomechanical properties of Muspratt, M.A., 1972. Numerical statistics in engineering geology. Engineering
the Miocene basalts in Malatya. Eastern Turkey. Bulletin of Engineering Geology 6, 67–78.
Geology and the Environment 64, 373–381. Palisade, 2000. Guide To Using @RISK; Risk Analysis and Simulation Add-In for
Hoek, E., 1994. Strength of rock and rock masses. ISRM (International Society for Microsofts Excel. Palisade Corporation, Newfield, NY.
Rock Mechanics) News Journal 2 (2), 4–16. Park, H.J., West, T.R., 2001. Development of a probabilistic approach for rock wedge
Hoek, E., 1998. Reliability of the Hoek–Brown estimates of rock mass properties failure. Engineering Geology 59, 233–251.
and their impact on design. International Journal of Rock Mechanics and Park, H.J., West, T.R., Woo, I., 2005. Probabilistic analysis of rock slope stability and
Mining Sciences 35, 63–68. random properties of discontinuity parameters, Interstate Highway 40,
Hoek, E., Brown, E.T., 1980. Underground Excavations in Rock. Institution of Western North Carolina, USA. Engineering Geology 79, 230–250.
Mining and Metallurgy, London 527pp. Priest, S.D., Hudson, J.A., 1976. Discontinuity spacings in rock. International Journal
Hoek, E., Brown, E.T., 1997. Practical estimates of rock mass strength. International of Rock Mechanics and Mining Sciences 13, 135–148.
Journal of Rock Mechanics and Mining Sciences 34, 1165–1186. Rethati, L., 1988. Probabilistic Solutions in Geotechnics. Elsevier, Amsterdam
Hoek, E., Carranza-Torres, C.T., Corkum, B., 2002. Hoek–Brown failure criter- 451pp.
ion—2002 edition. In: Proceedings of the 5th North American Rock Mechanics Ripley, B.D., 1987. Stochastic Simulation. John Wiley & Sons, New York 237pp.
Symposium and 17th Tunneling Association of Canada Conference: NARMS- Ruffolo, R.M., Shakoor, A., 2009. Variability of unconfined compressive strength in
TAC, Toronto, ON, pp. 267–271. relation to number of test samples. Engineering Geology 108, 16–23.
Hoek, E., Diederichs, M., 2006. Empirical estimates of rock mass modulus. Sari, M., 2009. The stochastic assessment of strength and deformability
International Journal of Rock Mechanics and Mining Sciences 43, 203–215. characteristics for a pyroclastic rock mass. International Journal of Rock
Hoek, E., Kaiser, P.K., Bawden, W.F., 1995. Support of Underground Excavations in Mechanics and Mining Sciences 46, 613–626.
Hard Rock. Balkema, Rotterdam 215pp. Sari, M., Karpuz, C., 2006. Rock variability and establishing confining pressure
Hoek, E., Marinos, P., Marinos, V., 2005. Characterization and engineering properties levels for triaxial tests on rocks. International Journal of Rock Mechanics and
of tectonically undisturbed but lithologically varied sedimentary rock masses. Mining Sciences 43, 328–335.
International Journal of Rock Mechanics and Mining Sciences 42, 277–285. Vose, D., 2000. Risk Analysis: A Quantitative Guide second ed. John Wiley & Sons,
Hsu, S.C., Nelson, P.P., 2002. Characterization of Eagle Ford shale. Engineering Chichester, UK 418pp.
Geology 67, 169–183. Wallis, P.F., King, M.S., 1980. Discontinuity spacings in a crystalline rock.
Hudson, J.A., Priest, S.D., 1983. Discontinuity frequency in rock masses. Interna- International Journal of Rock Mechanics and Mining Sciences 17, 63–67.
tional Journal of Rock Mechanics and Mining Sciences 20, 73–89. Yegulalp, T.M., Mahtab, M.A., 1983. A proposed model for statistical representation
Johnson, N.L., Kotz, K., Balakrishnan, N., 1994. Continuous Univariate Distributions, of mechanical properties of rock. In: Proceedings of the 24th U.S. Symposium
vol. 1. John Wiley & Sons, New York 756pp. on Rock Mechanics, Texas A&M University, pp. 61–69.