Sunteți pe pagina 1din 8

ARTICLE IN PRESS

International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335


www.elsevier.com/locate/ijrmms

Technical note

Rock variability and establishing confining pressure levels


for triaxial tests on rocks
M. Saria, C. Karpuzb,
a
Aksaray Engineering Faculty, University of Nigde, Aksaray 68200, Turkey
b
Department of Mining Engineering, Middle East Technical University, Ankara 06531, Turkey
Accepted 30 June 2005
Available online 6 September 2005

1. Introduction suggested methods for the triaxial test, is it mentioned


clearly what procedure should be followed to establish
In teaching rock mechanics, often the theoretical part the pre-test confining pressure levels. Terms such as ‘‘the
of the lectures are demonstrated by laboratory sessions, desired lateral pressure’’, ‘‘the desired value’’, ‘‘the pre-
which are helpful in supplementing the teaching of rock determined test level’’ and ‘‘the prescribed value’’ are
mechanics principles to the students. In these laboratory used. So, the test instructor has to arbitrarily settle these
sessions, an unconfined compressive strength (UCS) test pressure values by himself/herself based on his/her past
can be followed by a conventional triaxial compressive experience.
strength (TCS) test on similar rock specimens. The Ideally, at each increment of the confining pressure
former test is the one used most in the characterization the same rock specimen should be used for the test, but
of rock strength and in many rock classification systems; since rock specimens are broken after each test,
while the latter test has been widely used for the obviously the failed specimen has to be replaced by a
determination of design parameters such as C and f new sample. Thus, some conflicting results may be
because of the equipment simplicity and convenient obtained due to inherent variability in the failure
specimen preparation and testing procedures. However, characteristics of a substituted rock specimen, e.g.
the experience of the authors in conducting such sessions sometimes the new rock specimen is broken at a lower
has shown that problems are encountered while axial load for a higher confining pressure applied than
performing the conventional triaxial compressive for the previous one. Although Kovari et al. [5,6] have
strength tests and these can be misleading to the suggested that in a ‘‘continuous failure state’’ triaxial
students. test it is possible to determine the failure envelope of a
The conventional TCS test consists of applying a rock with the aid of a single cylindrical test specimen, it
hydraulic pressure to the external surface of a cylindrical is still necessary to define a pre-test confining pressure
rock specimen and at the same time axially loading the level for this test and also for the conventional TCS test,
specimen in compression with increasing increments since it is more universal and fundamental for the
until failure of the rock. The procedure for measuring students.
the rock strengths has been standardized by the In this Technical note, a practical approach will be
American Society for Testing and Materials (ASTM) introduced for establishment of pre-determined confin-
[1,2] and the International Society for Rock Mechanics ing pressure levels for the triaxial test on rocks by
(ISRM) [3,4]. Neither in the ASTM nor in the ISRM directly using the UCS and TCS test values collected
from past laboratory sessions of a rock mechanics
Corresponding author. Tel.: +90 312 210 2655; course. Also, for the purpose of this study, the possible
fax: +90 312 210 1265. probability distribution models, which take into account
E-mail address: karpuz@metu.edu.tr (C. Karpuz). the uncertainty and variability dictated by the nature of

1365-1609/$ - see front matter r 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijrmms.2005.06.010
ARTICLE IN PRESS
M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335 329

the strength of rocks under compression, have been type of uncertainty is called statistical and it can be
evaluated and the acceptability of the assumption diminished readily with increasing number of sam-
about the proposed distribution model can be further ples.
investigated.  The systematic uncertainty comes from the discre-
pancies between the laboratory and in situ condi-
tions, due to factors such as scale, anisotropy and
2. Uncertainty and variability in rock mechanics water content. Supplementary sampling may not
necessarily reduce these systematic discrepancy un-
Most engineering problems involve some degree of certainties, since the same test conditions persist and
uncertainty and variability. The causes of such uncertainty consistently yield different property values.
in an engineering content have been detailed by Bury [7]
and in particular for rock engineering by Duzgun et al. [8]. If the variables in a process demonstrate uncertainty
Bury [7] classifies sources of uncertainty as (1) data and variability, it is certainly necessary to describe the
uncertainty; (2) statistical uncertainty; (3) event uncer- nature of their uncertainty and variability. This can be
tainty; and (4) model uncertainty. achieved with probability distributions, which give both
the range of values that the variable could take and the
 Data uncertainty is caused mostly by the inherent likelihood of occurrence of each value within the range.
variability of a measured quantity. No matter how Generally, the distribution and its parameters are
carefully one measures such a quantity, variability initially unknown and need to be estimated from the
among the measured values is unavoidable because it available information about the random nature of the
is an inherent quality. rock variable. The information is usually in the form of
 Statistical uncertainty is due to the limited amount of a sample of observed values and there are many
information that is typically available for a measur- theoretical probability distributions (e.g. over 30 [7])
able quantity. The more data that are on hand, the that can represent the data with various degrees of
more information on that quantity that is available. success.
Thus, limited information implies uncertainty about The most basic categories of probability distributions
the true nature of the quantity. relate to whether they are continuous or discrete. A
 Event uncertainty is a result of little information discrete distribution may take one of a set of identifiable
available on the occurrences of events, especially rare values, each of which has a calculable probability of
events and their probabilities. occurrence. A continuous distribution, on the other
 Model uncertainty is related to the mathematical hand, is used to represent a variable that can take
models which the engineer uses in his/her work. any value within a defined range (domain) [9]. All
Typically, these models represent only one or a few of distribution types use a set of arguments to specify a
the important characteristics of the physical phenom- range of actual values and distribution of probabi-
enon in question. Models only represent a restricted lities. A normal distribution, for example, uses a mean
version of reality and, since they are an idealization of and standard deviation as its arguments. The mean
a real problem, they sometimes deviate from the defines the value around which ‘‘the bell curve’’ will be
actual situation. centered and the standard deviation defines the spread
of values around the mean. For a more comprehensive
Duzgun et al. [8] have reviewed and considered the description and use of probability distributions, the
sources of uncertainties affecting rock properties and reader is directed to Bury [7], Evans et al. [10], Johnson
highlighted the three components, namely, inherent et al. [11,12].
variability, statistical uncertainty and systematic uncer- A good example for the expression of inherent
tainties. uncertainty and variability in the context of rock
engineering can be given as the values obtained for a
 The inherent variability results from the fact that, rock material subject to unconfined compressive stress
even in a homogenous rock medium, the rock (UCS). Publication on this subject is limited, but there is
characteristics exhibit natural variability. The scatter work on the probabilistic strength determination of
of the strength values from a series of test specimens brittle materials [13–20]. The first significant contribu-
of the same rock material may be considerable. This tion to the investigation of the failure probability of
is a result of the randomly distributed weak planes, brittle materials was made by the Swedish physicist
micro cracks, or flaws in the rock, which influence the Waloddi Weibull [13], and the two or three parameter
rock strength. Weibull distribution (named after him) is mostly used to
 The sample statistics, i.e. mean, standard deviation, model the distribution of the failure strength of rocks
of a rock property are subject to uncertainty because and concrete by Li et al. [14], Huang et al. [15], Muro
of limited field sampling and laboratory testing. This et al. [16], Ghosh [17], Desayi and Rao [18], Rabie and
ARTICLE IN PRESS
330 M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335

Meguid [19], Grady and Kipp [20]. Moos et al. [21] have minimum of 32.09 MPa and a maximum of 103.79 MPa
studied the effects of in situ conditions and rock strength values have been obtained during the four-year
properties on wellbore stability predictions and they periods of the tests. The mean value for the samples’
suggested that it was necessary to utilize probabilistic UCS is 68.05 MPa with a standard deviation of
methods to quantify uncertainties caused by in situ 12.44 MPa. Summary statistics for the 103 specimens
stress and rock strength. In their study, the uniaxial are presented in Table 1.
compressive strength of the rocks adjacent to a wellbore
has been modelled by a log normal curve, one member 3.1. Fitting a distribution to the UCS tests
of the asymmetrical distribution family, in an attempt to
specify uncertainties for an input parameter in the Comparison graphs are superimposed on the input
quantitative risk analysis (QRA) approach. data and the fitted distribution curve on the same graph,
Hsu and Nelson [22] have statistically fitted a normal allowing one to visually assess the quality of the fit.
distribution to the UCS of 121 rock samples taken from Also, the frequency distribution of the sample may be
the Eagle Ford shale, which is classified as extremely statistically compared with the hypothetical distribution
weak to weak rock based on the UCS. The average to see if this assumption is really warranted. The
uniaxial compressive strength of the rock was 2.07 MPa frequency histogram of the actual test results is shown
but the UCS ranged from 0.44 to 5.82 MPa. They have in Fig. 1 for the UCS of the 103 rock specimens. As can
found that the assumed normal distribution was be seen from the graph, the unconfined strength of the
acceptable according to the results of the Kolmogor- andesite rock varies considerably for core samples taken
ov–Smirnov goodness-of-fit test (K–S test) at the from the different blocks of the same quarry. The full
specified significance level a, which was 0.01. nature of the UCS for this rock cannot be characterized
in the calculations by a single value, e.g. the mean of all
the observed strengths.
3. Ankara andesite
Table 1
The rock type used in the experiments for the rock Summary statistics for UCS (MPa)
mechanics course taught by the authors is broadly
described as Ankara andesite [23–26]. Andesite is a fine- Mean 68.05
grained igneous, volcanic (extrusive) rock, of intermedi- Median 66.72
Mode 63.80
ate composition, with aphanitic to porphyritic texture. It Standard deviation 12.44
is intermediate because it contains some minerals that Sample variation 154.70
are common to rhyolite and some common to basalt. It Coefficient of variation 18.3%
can appear very much like basalt, but it is usually less Range 71.70
dark or greenish in color. The variability in the Minimum 32.09
Maximum 103.79
characteristics of andesite is considerably more pro- Sum 7009.55
nounced than in the similar igneous rock groups. Count 103
Ozcelik [23] has investigated the effects of the Standard error 1.23
mechanical and physical properties of Ankara andesite Confidence interval (95%) 72.43
on the wearing rate of diamond wire beads. Karpuz and
Pasamehmetoglu [24] have extensively studied the area
where the rock blocks are gathered for the laboratory
work. They have detailed the weathering characteristics 40
of Ankara andesite, both from the engineering geology
and rock mechanics points of view at the quarries within
the vicinity of city of Ankara. For the Golbasi district, 30
they have distinguished three types of andesite, which
Frequency

are (a) brownish gray-to-grayish purple, fine-grained 20


glassy andesite; (b) pale red, fine-grained andesite; and
(c) gray, fine-grained andesite.
Preparation of the cylindrical rock samples was 10
carried out in the laboratory and a diameter of not less
than NX core size, approximately, 2 1/8 in (54 mm) is 0
used in the experiments. The length-to-diameter (L/D) 20 30 40 50 60 70 80 90 100 110 120
ratio of 2–2.5 is applicable to most of the specimens—in UCS (MPa)
accordance with the suggestions of the ISRM [4]. A total Fig. 1. Frequency distribution of UCS of andesite with a hypothetical
of 103 rock specimens were subjected to the UCS test; a normal distribution.
ARTICLE IN PRESS
M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335 331

A more realistic and rational approach is to include addition to visual inspection in Fig. 1, a detailed w2
all the variability in the analysis to reflect the random- goodness-of-fit test was carried out and Table 3 outlines
ness exhibited by the available data. This can be results of the w2 test for the 103 UCS values of the rock
achieved directly by representing the UCS of andesite samples for the normal distribution. The result of the
with a single probability distribution function. In Fig. 1, goodness-of-fit test suggests that it is a plausible
therefore, a fitted normal distribution curve is also approximation to represent the observed frequency
shown. distribution of UCS with a normal distribution. In this
To find the best fitting distribution type, a statistical method, a test statistic may be devised by dividing a
software package, BestFits was used [27]. BestFit standard normal distribution into a number of intervals.
evaluates all statistical distributions in its built-in library The expected frequency of occurrence within each
for the given data set, whether discrete or continuous, interval can be compared with the frequency of sample
and ranks the most appropriate distributions according observations that fall within the intervals. If the number
to their statistical goodness-of-fit test criteria, chi-square in each interval deviates significantly from that expected
(w2) or Kolmogorov–Smirnov tests. The output data it seems unlikely that the sample was drawn from a
generated from the results of BestFit are tabulated in normal population [28]. The test statistic is calculated by
Table 2 for the available distributions with their the equation
given rank according to w2 criteria. Note that the lower
the w2 value, the higher the p-value, the closer the Xk
ðOj  E j Þ2
w2 ¼ , (2)
theoretical distribution appears to fit the data. The best Ej
j¼1
distribution type representing the UCS of Ankara
andesite has been found to be normal with the where Oj is the number of actual observations within the
parameters: mean, m ¼ 68:05 MPa; and standard devia- jth class, and Ej is the number of observations expected
tion, s ¼ 12:44 MPa. Details of the distribution fitting in that class. There are k classes or intervals.
procedure and goodness-of-fit test are demonstrated in In the first column of Table 3, 103 UCS values are
the following section only for the normal distribution. arbitrarily subdivided into 10 classes, each having a
The probability density function for a normal distribu- width of 10 MPa. The second column lists the observed
tion with mean m and standard deviation s is given frequencies for each class, while the third column lists
below the expected frequencies for each class as a result of the
  best fitting normal curve in Fig. 1. The fourth column
1 ðx  mÞ2
f ðxÞ ¼ pffiffiffiffiffiffiffiffiffiffi exp  s40; 1oxo þ 1. then provides the w2 term for each class calculated from
2ps2 2s2
Eq. (2) and the overall w2 statistic is calculated as 4.877,
(1) with 9 degrees of freedom (number of classes –1) at the
bottom of Table 3 (the critical w2 value for the level of
3.2. w2 Goodness-of-Fit to UCS Tests significance a ¼ 0:05 with 9 degrees of freedom is 16.92).
As given in Table 3, a p-value of 0.845 was calculated,
We can test how well the distribution of sample values which is a relatively high value. For the w2 test, a higher
conforms to a theoretical distribution by a test p-value will increase the credibility of the null hypothesis
procedure called the goodness-of-fit or w2. So, in and eventually will lead to acceptance of the statement
that the UCS values of rock samples can be modeled as
coming from a normal population. As a conclusion, at

Table 2
Results of BestFit Table 3
w2 goodness of fit test of 103 rock specimens UCS values
Distribution type Rank w2 test value p-Value
UCS (MPa) range Observed frequency Expected frequency w2
Normal 1 4.88 0.845
Logistic 2 4.98 0.836 20–30 0 0.11 0.110
w2 3 6.42 0.697 30–40 2 1.13 0.670
Extreme value 4 17.79 0.038 40–50 4 6.31 0.846
Erlang 5 346 0.000 50–60 19 19.09 0.000
Lognormal 6 510 0.000 60–70 36 31.26 0.719
Student-T 7 N/A N/A 70–80 23 27.75 0.813
Exponential 8 N/A N/A 80–90 16 13.35 0.526
Pareto 9 N/A N/A 90–100 2 3.47 0.623
Triangular 10 N/A N/A 100–110 1 0.49 0.531
Uniform 11 N/A N/A 110–120 0 0.04 0.040
Weibull 12 N/A N/A
n ¼ 103; Sw2 ¼ 4:877; v ¼ 9; p-value ¼ 0.845.
ARTICLE IN PRESS
332 M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335

the level of significance a ¼ 0:05 the decision is to not squares technique; the regression equation of this line is
reject the null hypothesis of goodness-of-fit between characterized by its gradient (q) and its y-intercept (sc)
observed frequencies and theoretical frequencies. In as follows:
other words, the difference between observed frequen-
s1 ¼ 71:51 þ 4:60s3 (3)
cies and theoretical frequencies is not significant. Hence,
not only for practical considerations and simplicity of 2
with r ¼ 0:642, where, s1 and s3 are the axial stress and
working with it, but also because the UCS of rock confining pressure in MPa, respectively, q ¼ 4:60 and
specimens can be validly modeled via the normal sc ¼ 71:51 MPa. Using these parameters, f and C may
distribution in this case, this distribution is used in the be calculated using the formulae [3]
following calculations.
f ¼ arc sinððq  1Þ=ðq þ 1ÞÞ
3.3. Deriving the parameters from the triaxial tests and (4)
C ¼ sc ð1  sin fÞ=2 cos f:
The apparent cohesion (C) and internal friction angel
(f) may be calculated using triaxial compressive Here, f and C are calculated as 40.01 and 16.7 MPa,
strength test data by constructing a strength envelope respectively. In the equation, the y-intercept of the best
or Mohr’s envelope. The confining pressures and fitting line is 71.51 MPa; that is the estimated UCS of all
corresponding TCS values for the different specimens the samples and it is close to the actual calculated mean
of the same rock are plotted with the confining pressures UCS of the rock samples in Table 1 (s0 ¼ 68:05 MPa).
as abscissa and the corresponding strengths as ordinates One of the first things to do when running a regression
in Fig. 2, with the UCS test results. The best linear line model is to look at how well the model fits to the data.
passing through the points is fitted using the least The necessary model summary statistics, the analysis of
variance (ANOVA) table and regression coefficients are
presented in Table 4. The determination coefficient, r2, is
found as 0.642, which indicates that only 64.2% of the
200
changes in the TCS of rock specimens are due to
180
changes in the confining pressure; the remaining 35.8%
160
Axial Stress, σ1 (MPa)

unexplained part of the coefficient is attributable to the


140
other factors such as inherent variability in the rock
120
samples.
100
The ANOVA table in Table 4 is used to test the null
80
σ1 = 71.51 + 4.60 x σ3 hypothesis that there is no linear relation between the
60
r2 = 0,642 dependent variable, axial stress, and the independent
40
variable, confining pressure. The test of the null
20
hypothesis is based on the ratio of the regression mean
0
0 2 4 6 8 10 12 14 16 square to the residual mean square. In Table 4, the ratio
Confining Pressure, σ2=σ3 (MPa) of the two mean squares, labeled F, is 359.86. Since the
calculated significance level is less than 0.01, one can
Fig. 2. UCS and TCS plot of Golbasi andesite (strength envelope). reject the null hypothesis that there is no linear relation

Table 4
Model summary statistics and ANOVA table

R R2 Adjusted R2 Std. erorr Count

0.801 0.642 0.640 14.26 203

df SS MS F Significance

Regression 1 73201.48 73201.48 359.86 0.000


Residual 201 40886.96 203.42
Total 202 114088.44

Coefficients Std. error T p-Value Lower %95 Upper %95

Constant, sc 71.51 1.26 56.93 0.000 69.03 73.99


Variable, s3 4.60 0.24 18.97 0.000 4.12 5.08
ARTICLE IN PRESS
M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335 333

between the s1 and s3 variables. The linear regression tion exhibits a normal distribution, the same rock
model is a statistically valid model and it can be used specimens under different level of triaxial compression
satisfactorily for prediction purposes at different levels are theoretically expected to follow a similar normal
of confining pressure. In the same table, it has also been distribution with different parameters, whatever confin-
shown that both of the coefficients of the linear model ing pressure is applied to them.
are statistically significant, confirmed by considerably To make it understandable, the frequency distribution
higher t values. of rock specimens under 2, 4, and 6 MPa lateral stresses
The difference between the estimated UCS value from were presented in Fig. 3. Since there is an insufficient
the strength envelope and the actual one, whether number of observations at each stress level, it is hard to
meaningful or not, can be simply understood by detect any pattern in the type of distribution of rock
conducting a t-test. This t-test suggests that, since the samples under specified confining pressures. After
95% confidence interval of the actual UCS mean of evaluation of twenty plots of random samples of twenty
samples in Table 1, which is 68.0572.43 MPa, almost values drawn from a normal (100, 10) distribution, Vose
includes Eq. (3)’s estimated UCS confidence interval of [9] concluded that to establish a reasonable number
the lower 95% of 69.03 MPa, it is reasonable to say that from which to draw some inference about the type of
they may come from the same population of rock distribution is a challenging task. For this case, provided
samples. that the number of samples at each test level could be
Like UCS, the TCS of rock specimens should be approximately the number of samples used for the
essentially characterized by a probability distribution UCS test, it might be reasonable to talk about the type
function at each increment of confining pressure. For of distribution for the rock samples under triaxial
instance, if the UCS of rock samples under considera- loading.

8 σ3= 2 MPa 8 σ3 = 4 MPa


n=15 n=20
6 6
Frequency
Frequency

4 4

2 2

0 0
60 70 80 90 100 110 120 130 60 70 80 90 100 110 120 130
TCS (MPa) TCS (MPa)

8
σ3 = 6 MPa
n = 18
6
Frequency

0
60 70 80 90 100 110 120 130
TCS (MPa)

Fig. 3. TCS frequency distribution of rock samples at different confining pressures.


ARTICLE IN PRESS
334 M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335

4. Approach to setting the minimum confining pressure may be seen that the necessary minimum confining
pressure percentage applicable for any rock type should
Since the distribution of UCS (Fig. 1) is fairly be almost the half of the CoV of the rock samples.
symmetrical in shape and it has been previously Hence, it can be inferred that 10% of the mean UCS
approved by the w2 test that the UCS of samples follows value of any rock type can be simply suggested as the
probability rules for a normal distribution, it is initial confining pressure level for TCS testing.
applicable in the establishment of confining pressures.
This rule formally states that 73 times the standard
deviation from the mean should contain almost all of
the sample values [28,29]. Therefore, a practical and 5. Conclusions and recommendations
simple way can be the addition of 3 times the standard
deviation to the mean, so that this new value will be the A probabilistic approach has been explicitly detailed
highest UCS value observable for rock samples under in this study towards the solution of a testing problem
consideration. If this value is immediately located in that can be encountered frequently in rock mechanics.
Eq. (3) in place of s1 and resolving the equation in terms Because of vague terminology usage in the Suggested
of s3, it will give us the potential confining pressure that Methods, researchers and students were facing difficul-
should be applied to the specimens in an attempt to ties concerning the pre-setting of confining pressure
remove some of the inherent variability factor. levels for a conventional TCS test on rock. This work
The maximum observable UCS ¼ 3  Std of UCS+ reported here aims to solve this problem and some of the
Mean of UCS ¼ 3  12.44+68.05 ¼ 105.4 MPa. If we conclusions drawn from this work are as follows:
subsequently place this value into Eq. (3)
 UCS frequency distribution of rock samples can
s1 ¼ 71:51 þ 4:60  s3 ; exhibit a high variation and, instead of using only one
105:4 ¼ 71:51 þ 4:60  s3 ; point estimate, it is more convenient to represent this
s3 ¼ 7:4 MPa: value with a probability distribution representing the
natural uncertainty and variability.
This number can be clarified intuitively as the  Frequency distribution model for the UCS values of
necessary minimum confining pressure appropriate for Ankara andesite used in the case study here is
the rock specimen of Ankara andesite so as to account considered to be normal with the parameters,
for some of the natural variability inherent in the triaxial mean, m ¼ 68:05 MPa and standard deviation,
compressive strengths of rock samples. A confidence s ¼ 12:44 MPa.
interval will have been settled between the lower and  Additional TCS tests are required on similar rock
the next applicable lateral stress, so that the first specimens to remove some of the statistical uncer-
observable maximum failure strength would not be tainties and to delineate the type of possible
greater than the expected failure strength of the next distribution at different levels of confining pressure.
one. Expressed in another way, one is avoiding rock  A rough thumb rule may be recommended—the
failure whilst the confining pressure is increased to a necessary minimum confining pressure applicable for
prescribed level; its expected failure load would not be any rock type should be around the 10% of the mean
lower than the maximum observed UCS of the same rock. UCS. Another useful approach is to take the half of
Accordingly, by putting 105.4 MPa in Eq. (3) as the the CoV as the percentage of the mean UCS for the
expected sc, and calculating the maximum observable initial confining pressure level.
strength as 3  Std of UCS+105.4 MPa, and then  When this rule is followed, since the randomness in
putting this value in the place of s1, finally resolving the failure of rock samples has been taken into
the equation in terms of s3 will produce the next account, there will be much less risk of encountering
applicable value for the confining pressure. The same situations in which rocks will fail at a lower axial
procedure can be followed for the establishment of other loading when a higher confining pressure is applied to
levels of confining pressure. them.
If it is desired to express 7.4 MPa for simplicity as a
percentage in a similar notation as used in the coefficient
of variation (CoV), dividing this value by the actual Acknowledgments
mean of the UCS will give us a unitless ratio related to
the percentage of observed mean suitable as a minimum The authors gratefully acknowledge the following for
confining pressure on the rock samples. The CoV of this their help during the course of this study: Technicians
rock is found to be 18.3% (Table 1) and the pro- Nizamettin Aydemir and Hakan Uysal for their kind
posed minimum confining pressure will be 7.4 MPa/ help in the preparation of the rock specimens and
68.05 MPa ¼ 10.8% of the mean UCS. From here it conducting the tests; Research Assistants, Sinem Sener
ARTICLE IN PRESS
M. Sari, C. Karpuz / International Journal of Rock Mechanics & Mining Sciences 43 (2006) 328–335 335

for providing the UCS test results and Arman Kocal for [13] Weibull W. A statistical distribution function of wide applic-
providing some components of the TCS test values. ability. J Appl Mech 1951;18:293–7.
[14] Li QS, Fang JQ, Liu DK, Tang J. Failure probability pre-
diction of concrete components. Cem Concr Res 2003;33:
1631–6.
References [15] Huang C, Subhash G, Vitton SJ. A dynamic damage growth
model for uniaxial compressive response of rock aggregates. Mech
[1] American society for testing and materials. Standard test method Mater 2002;34:267–77.
for unconfined compressive strength of intact rock core speci- [16] Muro T, Takegaki Y, Yoshikawa K. Impact cutting property of
mens. ASTM Designation D 2938-79, ASTM standards 4.08: rock material using a point attack bit. J Terramech 1997;34:
Philadelphia; 1984. 83–108.
[2] American society for testing and materials. Standard test method [17] Ghosh A. A FORTRAN program for fitting Weibull distribution
for triaxial compressive strength of undrained rock core speci- and generating samples. Comput Geosci 1999;25:729–38.
mens without pore pressure measurements. ASTM designation D [18] Desayi P, Rao BK. Probabilistic analysis of tensile strength of
2664-67 (reapproved 1974), ASTM standards 4.08: Philadelphia; ferrocement. Int J Cem Compos Lightweight Concrete 1988;10:
1984. 15–25.
[3] Suggested methods for determining the strength of rock materials [19] Rabie AM, Meguid SA. An evaluation of a probabilistic
in triaxial compression. In: Brown ET, editor. Rock characteriza- engineering approach to the failure prediction of brittle materials.
tion, testing and monitoring—ISRM suggested methods. Oxford: Eng Fracture Mech 1986;24:243–53.
Pergamon Press; 1981. [20] Grady DE, Kipp ME. Continuum modelling of explosive fracture
[4] Suggested methods for determining the uniaxial compressive in oil shale. Int J Rock Mech Min Sci Geomech Abstr 1980;
strength and deformability of rock materials. In: Brown ET, 17:147–57.
editor. Rock characterization, testing and monitoring—ISRM [21] Moos D, Peska P, Finkbeiner T, Zoback M. Comprehensive
suggested methods. Oxford: Pergamon Press; 1981. wellbore stability analysis utilizing Quantitative Risk Assessment.
[5] Kovari K, Tisa A, Einstein HH, Franklin JA. Suggested methods J Pet SciEng 2003;38:97–109.
for determining the strength of rock materials in triaxial [22] Hsu S- C, Nelson PP. Characterization of Eagle Ford shale. Eng
compression—revised version. Int J Rock Mech Min Sci 1983;20: Geol 2002;67:169–83.
283–90. [23] Ozcelik Y. Multivariate statistical analysis of wearing on diamond
[6] Kovari K, Tisa A, Attinger RO. The Concept of ‘‘Continuous beads in the cutting of andesitic rocks. Key Eng Mater
Failure State’’ triaxial tests. Rock Mech Rock Eng 1983;16: 2003;250:118–30.
117–31. [24] Karpuz C, Pasamehmetoglu AG. Field characterization of
[7] Bury K. Statistical distributions in engineering. Cambridge: Cambridge weathered Ankara andesites. Eng Geol 1997;46:1–17.
University Press; 1999. [25] Ercanoglu M, Aksoy H. Potential instability map for rock slopes
[8] Duzgun HSB, Yucemen MS, Karpuz C. A probabilistic model for at Ankara Castle and vicinity. Yerbilimleri 2004;29:97–114 (in
the assessment of uncertainties in the shear strength of rock Turkish).
discontinuities. Int J Rock Mech Min Sci 2002;39:743–54. [26] Sener S, Tutluoglu L. Fracture toughness analysis of Ankara
[9] Vose D. Risk analysis: a quantitative guide. 2nd ed. Chichester: andesite. In: Proceedings of 18th international mining congress
Wiley; 2000. and exhibition of Turkey, IMCET; 2003. p. 89–92.
[10] Evans E, Hastings N, Peacock B. Statistical distributions. 2nd ed. [27] BestFits and @RISK user manual, New York, Palisade Corporation.
New York: Wiley; 1993. 2001.
[11] Johnson NL, Kotz K, Kemp AD. Univariate discrete distribu- [28] Davis JC. Statistics and data analysis in geology. New York:
tions. New York: Wiley; 1993. Wiley; 1973.
[12] Johnson NL, Kotz K, Balakrishnan N. Continuous univariate [29] Harnett DL, Horrell JF. Data, statistics, and decision models with
distributions, vol. 1–2. New York: Wiley; 1994–1995. Excel. New York: Wiley; 1997.

S-ar putea să vă placă și