You are on page 1of 12

Rock Engineering and Technology for Sustainable Underground Construction

Eurock 2012 – the 2012 ISRM International Symposium, 28-30 May 2012, Stockholm, Sweden.
© BeFo and ISRM, 2012

Stochastic estimation of the Hoek-Brown strength parameters


using spreadsheet models

M. Sari
Department of Mining Engineering, Aksaray University, 68100 Aksaray, Turkey

Abstract: Empirical models are being frequently used to estimate rock mass
mechanical properties. It is well known that most empirical methods in rock
engineering give averaged values, and that it might be significant variation between
the lowest and highest value. Besides, it is highly important to describe relations
between input parameters in an empirical model if one wants to obtain accurate results
during stochastic analysis. In this paper, we run two different Monte Carlo simulations
for the comparison of results generated from empirical Hoek-Brown failure criterion.
First simulation model assumes all input parameters used in the criterion as
independent variables and second model includes the relationships between input
parameters via a correlation matrix. The correlation matrix used in the simulations is
succeeded by consulting knowledge of some experts in the field of rock engineering. It
is found that with or without considering correlations, the mean values of simulation
outputs computing the rock mass strength parameters are not notably different.
However, the standard deviations of strength parameters are generally smaller in
simulation results taking into account of correlations between input parameters. It is
concluded that in a stochastic estimation study, defining the relationships between
input parameters would not change the simulation results drastically.

Theme: Geological Site Characterization

Keywords: Hoek-Brown criterion, Monte Carlo method, rock mass strength,


correlation matrix, deformation modulus.

Eurock 2012 Page 1


1 INTRODUCTION
A rock mass is a system basically composed of two components; intact rock pieces
and discontinuity network. Intact rock refers to unfractured blocks between structural
discontinuities. A discontinuity is described as “a plane of weakness that has zero or
low tensile strength or tensile strength lower than stress levels generally applicable in
engineering applications” (Anon. 1977). Because there are so many elements or
factors that influence the engineering properties of rock masses and the inherent
variability of the value is very large, it is important to describe and characterize rock
masses in a way to find representative ratings and values for use in engineering and
design. Consequently, the classification items and scores in most widely used rock
mass classification systems are based on the experiences and subjectivity of the
originators. Riedmuller & Schubert (1999) stated that complex properties of a rock
mass could not sufficiently be described by a single number. Therefore, rock mass
characterization and description are main issues in engineering geology and rock
engineering.
Stochastic modeling may be a useful tool for rock engineers and engineering
geologists who study in complex rock mass conditions. However, it is extremely
important to describe the correlations observed between input parameters in an
empirical strength model based on classification systems if one wants to obtain
meaningful combinations during stochastic modeling. Since it is challenging to
establish the exact nature of relationships observed between intact rock material and
discontinuity parameters in the model only subjective estimates can be made. A
plausible alternative is to consult the knowledge of experts in the field.
A simple correlation matrix quantifying the relationship between rock material
and rock mass properties used as inputs of a Monte Carlo simulation is previously
proposed elsewhere (Sari et al. 2010). A new matrix basis on the expertise of many
distinguished researchers from the rock engineering and engineering geology
community appears to be very promising and of great potential. It can be readily
incorporated as an input into stochastic model studies carried out in the future.
In this paper, the stochastic models originally developed for the estimation of rock
mass strength properties of Ankara andesites (Sari et al. 2010) and Kizilkaya
ignimbrites (Sari 2009) are discussed. The proposed spreadsheet models include input
parameters of the RMR system (Bieniawski 1989) and the generalized Hoek-Brown
failure criterion (Hoek et al. 2002) as statistical distributions. Also, a correlation
matrix, which defines the relationships between input parameters, elicited from the
opinions’ of the experts in the rock mechanics community is presented in the paper.
Whether if there is a significant difference between the results of stochastic model
considering the correlations between input parameters and the one assuming the
parameter independence is statistically investigated.

2 PREVIOUS STUDIES
Monte Carlo (MC) methods are used in porous media flow and transport problems for
ground water contamination and remediation studies. Huang et al. (2003) applied MC
methods to study groundwater flow and solute transport in heterogeneous, dual-
porosity media and compared the results with analytical models. Lu & Zhang (2003)
demonstrated the development of an important sampling method to solve complicated
problems with MC and applied it to fluid transport problems in aquifers. Morin &
Ficarazzo (2006) used stochastic techniques and MC simulations to predict
fragmentation of rock during blasting. They have shown that the results produced by
the simulator were comparable with the data obtained from a quarry, and that the use

Eurock 2012 Page 2


of MC extended the understanding of the factors affecting blast fragmentation.
Karacan & Luxbacher (2010) described a practical approach for implementing
stochastic determination of gob gas ventholes (GGV) production performances and for
generalizing the prediction capability of deterministic models. They indicated that this
approach was a promising method of representing the variability in GGV
performances and to improve the limited and site-specific character of the
deterministic models.
In the literature, there are a limited number of studies which considers only the
stochastic estimation of strength and deformability characteristics and variations of
rock masses. Kim & Gao (1995) presented a probabilistic method of estimating the
mechanical characteristics of a rock mass, using the third type asymptotic distribution
of the smallest values (extreme value statistics) and MC simulation. They used the chi-
square goodness-of-fit test to prove that the distribution reflects the inherent
variability of the properties of a basaltic rock. Hoek (1998) applied the same method
to estimate variation in the Hoek–Brown properties of a hypothetical rock mass, and
assuming that all three input parameters of the criterion can be represented by normal
distributions. Sari (2009) and Sari et al. (2010) demonstrated the use of MC
simulations to evaluate the strength and deformability of rock masses by including the
uncertainties of the intact rock strength and discontinuity parameters. They concluded
that the MC method provided a viable means for assessing the variability of rock mass
properties.

3 MATERIALS AND METHODS

3.1 Study area


The Ihlara Special Environmental Protection Area is located in the Cappadocia region
in Central Anatolia, Turkey. The area is bordered by two large Quaternary central
stratovolcanoes: Mount Erciyes (3917 m) on the eastern margin and Mount Hasandag
(3257 m) on the southwestern margin of Cappadocia, near Aksaray (Fig. 1). The
average topography of the area is about 1300 m. The region, possessing geological,
archaeological, ethnographic, biological, agricultural, and tourism features and
activities, has a vast amount of natural and cultural values.

Figure 1. Location map of study area.

Eurock 2012 Page 3


The Kizilkaya ignimbrite subject of this study well outcrops at Kizilkaya village and
in the Ihlara Valley (Fig. 2). The study area covers the historical and touristic Ihlara
Valley, which is about 14 km in length and covers 52 km2. The rocks in the study area
are basically classified as pyroclastic, called Selime tuffs, Kizilkaya ignimbrites, and
Hasandag ashes. With the influence of both water and wind erosion, interesting rock
shapes and morphological figures have been formed in the slopes of the valley. There
are also joint systems that have developed on the rock mass as a result of cooling of
material deposited during volcanic eruptions.

Figure 2. A view of the Ihlara Valley and Kizilkaya ignimbrites.

The engineering geological properties of the exposed Kizilkaya ignimbrite in the study
area were determined on the basis of field observations/measurements and laboratory
tests. The main orientation, spacing, persistence, aperture, filling, weathering, and
roughness of the discontinuities were described using the scan-line survey method
following the ISRM (1981) description criteria. A total of 260 discontinuities were
measured along a straight outcrop surface using a measuring tape and compass. A total
of 18 rock blocks were collected from the field, then 112 cube specimens were
prepared from the blocks for laboratory testing. The uniaxial compressive strength
(UCS) tests were conducted according to Turkish Standards of methods of testing for
natural building stones (TS 699 1987). These values are given in Table 1.

Table 1. Discontinuity and intact rock properties of studied rock.

Parameter Distribution type


(Mean±Std.Dev.)
UCS (MPa) Normal (46.06±8.63)
Joint spacing (m) Lognormal (2.575±1.195)
Joint persistence (m) Normal (27.09±6.79)
Joint aperture (cm) N. Exponential (3.31±3.26)
Joint roughness 25% very rough, 50% rough,
25% smooth
Weathering 25% fresh, 50% slightly, 25%
moderately
Infill 50% unfilled, 25 % sand and
gravel, 25% soft fillings
Groundwater Dry
mi Normal (13.0±2.0)

Eurock 2012 Page 4


3.2 Hoek-Brown failure criterion and RMR classification system
The most recent version of the generalized Hoek-Brown failure criterion (Hoek et al.
2002) is employed to estimate the rock mass properties of the Kızılkaya ignimbrites.
At failure, the generalized Hoek-Brown criterion relates the maximum effective stress,
σ1 to the minimum effective stress σ3 through the functional relation:
a
 σ 
σ 1 = σ 3 + σ ci  mb 3 + s  (1)
 σ ci 
where mb extrapolates the intact rock constant mi to the rock mass:
 GSI − 100 
mb = mi exp  (2)
 28 − 14 D 
σci is the uniaxial compressive strength of the intact rock and s and a are constants that
depend upon the rock mass’s characteristics:
 GSI − 100 
s = exp  (3)
 9 − 3D 

a= + e
2 6
(
1 1 −GSI / 15
− e −20 / 3 ) (4)

The uniaxial compressive and tensile strengths of the jointed rock masses are
calculated from the following equations suggested by Hoek et al. (2002):
σ c ( MPa) = σ ci .s a (5)

s.σ ci
σ t ( MPa) = − (6)
mb
The static modulus of deformation is among the parameters that best represent the
mechanical behavior of a rock and a rock mass. Based upon practical observations and
the back analysis of excavation behavior in poor quality rock masses, the following
equation was proposed by Hoek et al. (2002) for σci < 100:

 D  σ ci
E m (GPa) = 1 −  .10 ((GSI −10 ) / 40) ) (7)
 2  100
D is a factor that depends upon the degree of disturbance to which the rock mass has
been subjected by blast damage and stress relaxation. Since the rock mass in this study
is naturally outcropping, it is plausible to assume that it has been undisturbed and the
value D=0 is used in the above equations.
Rock Mass Rating (RMR) system (Bieniawski 1989) is one of the most widely
used rock mass classifications. It was originally based upon case histories drawn from
civil engineering applications. This method incorporates geological, geometric and
design/engineering parameters in arriving at a quantitative value of the rock mass
quality. This engineering classification system utilizes the following six rock mass
parameters: (1) Uniaxial compressive strength (UCS) of intact rock material, (2) Rock
quality designation (RQD), (3) Spacing of discontinuities, (4) Condition of
discontinuities (persistence, roughness, infilling, weathering), (5) Groundwater
conditions, (6) Orientation of discontinuities.

Eurock 2012 Page 5


3.3 Stochastic modeling approach
It is difficult to quantitatively predict the mechanical properties of jointed rock masses
due to their complexity, even by expensive, large-scale in-situ tests. Prediction of
jointed rock mass properties by empirical models has been widely applied in rock
mechanics and engineering field of study. However, empirical models developed for
the estimation of rock mass properties involve a large number of input parameters,
each subject to substantial uncertainty. Therefore, an accurate assessment of rock mass
strength and deformability must involve stochastic analysis and must attempt to model
the inherent variability and uncertainty in these parameter estimates.
Probabilistic analysis is one method for performing a multi-way sensitivity
analysis in which all parameters subject to uncertainty are varied simultaneously by
MC sampling from the distributions postulated for those parameters. Instead of
tackling the numerical problem directly, MC allows the researcher to obtain an
approximation of the solution through setting up an experiment of statistical sampling.
As the name indicates, the method borrows from games of chance such as those
played at the famous casinos of Monte Carlo in Monaco. Ideally, to correctly apply the
MC method and obtain valid results, the sampling method employed should be
completely random. The number of realizations has to be sufficiently large to
accurately represent the distribution of the input variables.
Figure 3 is a simple illustration of the stochastic modeling methodology applied in
this study. In the stochastic estimation of rock mass strength and deformability
properties of Kizilkaya ignimbrites, the following steps are taken into consideration:

i. The data for discontinuity parameters and intact rock strength characterization was
compiled from the field studies and laboratory experiments.
ii. Closed form distribution functions, which represent both the probability and range
of values that would be expected in the field and laboratory were defined for each
parameter of the RMR classification scheme.
iii. The stochastic assessment of rock mass properties was accomplished using the
probability density distributions from the previous step. MC simulations were
executed to obtain a statistical representation of the RMR and RQD in the
spreadsheet model.
iv. UCS, GSI, and mi were defined as parametric distributions and included in
strength and deformability equations to estimate the mb, s, and a parameters of the
Hoek–Brown failure criterion for the rock mass.
v. By running the MC simulation for the rock mass properties described as statistical
distributions in previous steps, the mean, standard deviation, and confidence
intervals (range) of the uniaxial compressive strength (Sigc), uniaxial tensile
strength (Sigt), global rock mass strength (Sigcm), and deformation modulus
(Erm) of the rock mass were then achieved for the studied rock mass.

3.4 Construction of a correlation matrix


In most rock engineering analyses, there are several parameters-often many-that may
vary randomly. If the value taken on by one variable has no influence upon the value
assumed by another variable, then these variables are said to be independent.
Independence of random variables greatly simplifies the representation and analysis of
uncertainty, and often independence is assumed even where it is not really true.
The MC method used in the previous studies (Sari 2009, Sari et al. 2010) relies
heavily on estimating GSI from a simulated RMR value using specific assumptions for
joint orientation and water conditions (where GSI=RMR89-5). Unfortunately, RMR
does not provide a unique measure (description) of the rock mass; that is many

Eurock 2012 Page 6


Figure 3. Schematic diagram of stochastic modeling methodology applied to rock masses (Sari et
al. 2010).

different combinations of the RMR inputs can yield the same RMR value (e.g. high
UCS and low RQD can give the same measure as low UCS and high RQD). Thus, the
GSI simulated value may have no real counterpart for the actual GSI based on field
observations of the rock mass. This may be cause critical errors in the MC output for
rock mass strength and deformability using the Hoek-Brown failure model.
A complete probabilistic treatment of dependent random variables requires joint
probability distributions, which for two variables may be depicted as a surface. Simple
descriptors suffice in place of full probability distributions for many applications; the
descriptors covariance and correlation coefficient indicate the degree of dependence
among the variables.
Sari (2009) considered intact rock and joint parameters of the Kizilkaya
ignimbrites probabilistic in nature; however, they were also assumed to be
independent. For this study, to overcome the problem of selecting a representative
rating of various parameters, a correlation matrix is constructed by applying opinions
of experts in the rock engineering field. For this purpose, a form of empty correlation
matrix was sent to the researchers and it was requested to complete the matrix
according to their engineering judgment. During the construction of the matrix, the
researchers were first defined the direction of the relationship observed between input
parameters as negative or positive. Then, the strength of association anticipated
between these parameters was quantified using a descriptive scale. There were five
classes into which the relationships could be assigned ranging from 0 to 1, relating to
‘‘none’’ (0), ‘‘weak’’ (0.3), ‘‘moderate’’ (0.6), ‘‘strong’’ (0.9), and ‘‘perfect’’ (1).
Only seven experts out of fifty were filled the matrix. Figure 4 captures the
correlation coefficients assigned by those experts for some of pairs of input
parameters. A final matrix was intuitively constructed by averaging of scores assigned
by the various researches for each corresponding cell. The matrix is presented in
Table 2 as a lower triangular matrix and it can be used in MC simulations directly. To
the author’s knowledge, the construction of such a matrix is probably one of the first
attempts in the field of rock engineering.

Eurock 2012 Page 7


UCS vs. RQD UCS vs. Spacing UCS vs. Persistence
1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

Fortsakis

Fortsakis

Fortsakis
Hudson

Marinos

Marinos
Romana

Romana

Marinos
Hudson

Hudson
Romana
0.2 0.2 0.2

Hoek

Hoek
Hoek

Şen

Sari

Şen

Sari

Şen
Sari
0 0 0

RQD vs. Spacing RQD vs. Weathering Spacing vs. Groundwater


1 0 0.8

Romana

Sari

Hoek

Şen

Hudson

Marinos

Fortsakis
0.6
0.8 -0.2

Romana

Hudson
0.4
0.6 -0.4 0.2
0

Şen

Marinos
Sari

Hoek

Fortsakis
0.4 -0.6 -0.2
Fortsakis
Romana

Marinos
Hudson

-0.4
0.2 -0.8
Hoek
Sari

Şen

-0.6
0 -1 -0.8

Figure 4. Scores assigned by different experts on the relationship between pairs of input
parameters.

Table 2. Matrix showing correlations between intact rock and discontinuity parameters.

UCS RQD S P A R I W GW m E D
UCS 1
RQD 0.51 1
Spacing 0.30 0.91 1
Persistence 0.30 -0.47 -0.30 1
Aperture 0 -0.40 -0.20 0.33 1
Roughness 0.43 0.21 0.17 0.09 -0.39 1
Infilling -0.30 -0.33 0 0.31 0.63 -0.47 1
Weathering -0.61 -0.71 -0.43 0.47 0.13 -0.56 0.77 1
GW -0.21 -0.36 0 0.43 0.47 -0.11 -0.46 0.64 1
m 0.71 0.26 0 0 0 0 -0.13 -0.40 0 1
E 0.89 0.49 -0.49 -0.34 -0.56 0.51 -0.39 -0.70 -0.17 0.46 1
D -0.26 -0.50 -0.41 0.27 0.51 0 0.36 0.51 0.39 -0.30 -0.69 1
UCS - uniaxial compressive strength, RQD - rock quality designation, S – spacing, P – persistence,
A – aperture, R – roughness, I – infilling, W – weathering, GW - groundwater, m - material constant,
E - elasticity modulus, D - disturbance factor.

4 RESULTS AND DISCUSSION


The software program used to perform the data analysis is @RISK from Palisade
(Palisade 2001). This software allows for a basic data fitting by the use of Maximum
Likelihood Estimators to estimate the distribution parameters (i.e. to determine the
parameters that maximize the likelihood of the sample data). It explicitly includes the
uncertainty present in inputs to generate outputs that show all possible alternatives.
The program provides a simple and intuitive implementation of the MC simulation
together with the generalized Hoek–Brown failure criterion, allowing users to easily
obtain reliable estimates of rock mass properties, and to obtain a visualization of the
changing effects of input variables on rock mass parameters.
A spreadsheet model for carrying out @RISK simulations, with a listing of all cell
formulas for input and output parameters, is given in Figure 5. For the model, 5,000
iterations are performed with Latin Hypercube sampling to closely resemble the
resulting probability distribution. This means that every run of the simulation yields
5,000 different possible combinations of input variables, which are sampled randomly
from the defined distributions.
Eurock 2012 Page 8
A B C D E F G H I
1 Rating Roughness Infilling Weathering
2 UCS (MPa) 46.06 4 x p x p x p
3 RQD (%) 95.0 20 6 0.05 6 0.5 0 6 0.25
4 Spacing (m) 2.575 20 5 0.20 2 0.2 5 5 0.50
5 Persistence (m) 27.09 0 3 0.40 0 0.2 5 3 0.25
6 Aperture (cm) 3.31 0 1 0.20
7 Roughness 3 0 0.15
8 Infilling 2 B2 =Normal(46.06; 8.63;Truncate(0; 100))
9 Weathering 5 B3 =100*Exp(-0.1*1/B4)*(0.1*1/B4+1)
10 Groundwater 15 B4 =Lognorm(2.575; 1.295;Truncate(0 ; ))
11 RMR 69.0 B5 =Normal(27.09; 9.79;Truncate(0; 60))
12 B6 =Expon(3.31;Truncate(0; ))
13 C7 =Discrete(D3:D7;E3:E7)
14 GSI 64.0 C8 =Discrete(F3:F5;G3:G5)
15 mi 13.0 C9 =Discrete(H3:H5;I3:I5)
16 mb 3.688 C10 =15
17 s 0.022 C11 =Sum(C2:C10)
18 a 0.502 B14 =C11-5
19 B15 =Normal(13.0; 2.0;Truncate(8; 18))
20 B16 =B15*Exp((B14-1 00)/28)
21 Sigc 6.591 MPa B17 =Exp((B14-100)/9)
22 Sigt -0.266 MPa B18 =1/2+1/6*(Exp(-B14/15)-Exp(-20/3))
23 Sigcm 12.643 MPa B21 =B2*B17^B18
24 Erm 16.074 GPa B22 =-B17*B2/B16
25 B23 =B2*(B16+4*B17-B18*(B16-8*B17))*
26 (B16/4+B17)^(B18-1)/(2*(1+B18)*(2+B18))
27 B24 =(B2/100)^1/2*10^((B14-10)/40)
Figure 5. Spreadsheet model used in MC simulation analysis.

In the current view of the model spreadsheet, only average values of the input
parameters are shown in the related cells and they result in a total of 69 RMR ratings.
In fact, these ratings are replaced every time with a new suitable value generated
randomly from any of the specified parent distribution. The cells, C2-C10, containing
the input parameters are recomputed automatically according to the appropriate range
of the RMR ratings given in the Bieniawski’s 1989 classification scheme.
Since the 1989 version of Bieniawski’s RMR classification is used in this
research, in the strength equations, the GSI is settled RMR89-5 as suggested by Hoek
& Brown (1997), where RMR89 has the groundwater rating set to 15 and the
adjustment for joint orientation set to zero. These ratings can then be assembled to
acquire a GSI distribution with a mean value of 64.35 and standard deviation of 4.5.
The resulting population tends to be normally distributed (Fig. 6), a consequence of
the Central Limit Theorem. It is known that variables that occur as a sum of a number
of random effects, none of which dominate the total, are normally distributed.
The summary statistics computed from the @RISK simulations for output
variables are presented in Table 3. Here, case_1 corresponds to the simulation without
considering the correlation and case_2 corresponds to the simulation considering the
correlation. For case_1, the model calculates the mean values for UCS, tensile
strength, global strength, and deformation modulus of the Kizilkaya ignimbrites as
6.586, 0.263, 12.65 MPa, and 16.06 GPa, respectively. For case_2, the model
estimates the same rock mass properties as 6.548, 0.257, 12.70 MPa, and 15.97 GPa,
respectively. It can be easily seen that the difference between mean values for two
cases is practically negligible. Since two approaches produce almost similar results in
the stochastic analysis it can be concluded that including the correlations between

Eurock 2012 Page 9


input parameters would not produce considerable changes on the results of MC
simulations.

X <= 57 X <= 72
0.1 5% 95%

Mean = 64.35

0.08
% Frequency

0.06

0.04

0.02

0
50 55 60 65 70 75 80
GSI_1 (%)

Figure 6. Frequency distribution of estimated GSI values simulated from RMR.

Table 3. Summary of output variables obtained from MC simulations for two cases.
Output Min. Max. Mean Std Dev COV 5% 95%
GSI_1 (%) 50 79 64.35 4.51 7.0 57 72
GSI_2 (%) 50 78 64.35 4.12 6.4 57 71
Sigc_1 (MPa) 1.311 22.29 6.586 2.38 36.1 3.416 11.28
Sigc_2 (MPa) 1.526 15.16 6.548 2.22 33.9 3.532 10.76
Sigt_1 (MPa) 0.034 1.527 0.263 0.12 46.4 0.116 0.502
Sigt_2 (MPa) 0.064 0.714 0.257 0.10 39.0 0.124 0.446
Sigcm_1 (MPa) 4.096 28.34 12.65 3.13 24.8 8.128 18.36
Sigcm_2 (MPa) 3.484 23.98 12.70 3.33 26.2 7.780 18.68
Erm_1 (GPa) 4.939 44.95 16.06 4.86 30.2 9.298 25.35
Erm_2 (GPa) 5.809 35.99 15.97 4.46 27.9 9.578 24.41

Output values obtained from the MC simulations also indicate that the properties of
rock mass might vary more extensively if the input variables are assumed as
independent in the analysis. Therefore, the standard deviations of rock mass strength
parameters are usually greater in the MC simulations not taking into account of
correlations. As can be followed from Table 3, coefficient of variation (COV), which
is defined as the observed standard deviation expressed as a percentage of the
observed mean, is generally slightly lower in the results assuming the parameter
dependence than from the ones considering the parameter independence. This may be
as a result of decreasing the chance of obtaining combinations sampled from the
extremes of input distributions during simulation.
One of the most common graphical representations of a probability distribution is
a histogram in which the fraction of all observations falling within a specified interval
is plotted as a bar above that interval. The histograms of some of the output variables
evaluated by stochastic modeling are given in Figure 7. It is important to note that the
graphs obviously show that none of the output variables have a single value; in fact,
all exhibit considerable variations between some specified intervals. Information
obtained from the simulation is still relevant, and gives further insight into evaluation
of the strength parameters estimated from the Hoek-Brown criterion.

Eurock 2012 Page 10


X <= 3.42 X <= 11.28
X <= .12 X <= .45
5% 95%
0.2 5% 95%
Mean=6.58
MPa 4 Mean = 0.257
0.16 MPa

% Frequency

% Frequency
0.12

2
0.08

0.04 1

0 0
0 4 8 12 16 20 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Sigc_1 (MPa) Sigt_2 (MPa)

X <= 9.3 X <= 25.35


5% 95% X <= 8.13 X <= 18.36
0.1 5% 95%
Mean = 16.06 0.14
GPa Mean = 12.65
0.08 0.12
MPa
% Frequency

0.1

% Frequency
0.06
0.08

0.04 0.06

0.04
0.02
0.02

0
0
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30
Erm_1 (GPa) Sigcm_1 (MPa)

Figure 7. Histograms of some of the output variables generated from the MC simulations.

In general, the strength and deformability characteristics of the Kizilkaya ignimbrite


can be seen to approximately resemble asymptotic distributions skewing to larger
values similar to intact rock materials. Kim & Gao (1995) obtained that the laboratory
and field test data showed the same type of distributive character although the test
methods and the volumes of the test specimens were substantially different.

5 CONCLUSIONS
Stochastic modeling is a technique where an existing mathematical model of a system
is used to quantify the uncertainty and variability. In this study, this technique has
been used to estimate the variability in the strength and deformability characteristics
of a pyroclastic rock mass. Mechanical properties of studied rock mass have been
satisfactorily modeled by including the frequency distributions of intact rock strength
and discontinuity parameters in the analysis. It is believed that the results of such a
probabilistic analysis most likely will more closely representative of the complex rock
mass conditions dealt with in the real world.
Overall, the rock mass properties affected by the random discontinuity
characteristics and intact rock properties, which are widely scattered and variable,
cannot be sufficiently represented by a single value input parameters and a single
output value. Therefore, it is recommended that the uncertainty analysis should be
applied, particularly in cases where there is significant scatter in the data of
discontinuity and intact rock parameters. However, the study evidently shows that the
consideration of the relationship between input parameters in a stochastic analysis will
not significantly affect the results generated from MC simulations. Adding correlation
coefficients into MC simulations will only slightly diminish the standard deviation of
the results.

Eurock 2012 Page 11


REFERENCES
Anonymous, 1977. The Description of Rock Masses for Engineering Purposes. Report
by the Geological Society Engineering Group Working Party. Quart. J. Eng.
Geol., 29:67-81.
Bieniawski, Z.T. 1989. Engineering Rock Mass Classification. New York: John Wiley
& Sons.
Hoek, E. 1998. Reliability of the Hoek-Brown estimates of rock mass properties and
their impact on design. Int J Rock Mech Min Sci, 35:63–68.
Hoek, E. & Brown, E.T. 1997. Practical estimates of rock mass strength. Int. J. Rock
Mech. Min Sci., 34:1165-1186.
Hoek, E., Carranza-Torres C. & Corkum, B. 2002. Hoek–Brown failure criterion –
2002 Edition. In Proceedings of the +orth American Rock Mechanics Society,
Toronto, July 2002.
Huang, H., Hassan, A.E. & Hu, B.X. 2003. Monte Carlo study of conservative
transport in heterogeneous dual-porosity media. J. Hydrology, 275:229-241.
ISRM, 1981. ISRM Suggested Methods: Rock Characterization, Testing and
Monitoring. ET Brown (ed.). London: Pergamon Press.
Karacan, C.Ö. & Luxbacher, K. 2010. Stochastic modeling of gob gas venthole
production performances in active and completed longwall panels of coal mines.
Int. J. Coal Geology, 84:125-140.
Kim, K. & Gao, H. 1995. Probabilistic approaches to estimating variation in the
mechanical properties of rock masses. Int. J. Rock Mech. Min. Sci., 32:111-120.
Lu, Z. & Zhang, D. 2003. On the importance sampling Monte Carlo approach to
uncertainty analysis for flow and transport. Advances in Water Resources,
26:1177-1188.
Morin, A.M. & Ficarazzo, F. 2006. Monte Carlo simulation as a tool to predict
blasting fragmentation based on the Kuz-Ram model. Computers & Geosciences,
32:352-359.
Palisade, 2001. @RISK Program and User Manual. New York: Palisade Corporation.
Riedmuller, G. & Schubert, W. 1999. Rock mass modeling in tunneling versus rock
mass classification using rating methods. In Proceedings of the 37th US Rock
Mechanics Symposium, Vail, Colorado, June 1999.
Sari, M. 2009. The stochastic assessment of strength and deformability characteristics
for a pyroclastic rock mass. Int. J. Rock Mech. Min. Sci., 46:613-626.
Sari, M., Karpuz, C. & Ayday, C. 2010. Estimating rock mass properties using Monte
Carlo simulation: Ankara andesites. Computers & Geosciences, 36:959-969.
TS 699, 1987. Tabii Yapı Taşları-Muayene ve Deney Metotları, Ankara: Türk
Standartları Enstitüsü, [in Turkish].

Eurock 2012 Page 12

You might also like