You are on page 1of 12

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318419408

Characterization of the distribution of physical and mechanical properties of


rocks at the Tutupan coal mine, South Kalimantan, Indonesia

Conference Paper · September 2013

CITATIONS READS

4 323

1 author:

Masagus ahmad Azizi


Universitas Trisakti
35 PUBLICATIONS   100 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

RESUME PERALATAN TAMBANG View project

Tugas Resume Paper Perhapi View project

All content following this page was uploaded by Masagus ahmad Azizi on 14 July 2017.

The user has requested enhancement of the downloaded file.


Rock Mechanics for Resources, Energy and Environment – Kwaśniewski & Łydżba (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00080-3

Characterization of the distribution of physical and mechanical properties


of rocks at the Tutupan coal mine, South Kalimantan, Indonesia

M.A. Azizi
Mining Engineering Department, Institut Teknologi Bandung, Indonesia
Mining Engineering Department, Trisakti University, Indonesia

S. Kramadibrata & R.K. Wattimena


Mining Engineering Department, Institut Teknologi Bandung, Indonesia

I.D. Sidi
Civil Engineering Department, Institut Teknologi Bandung, Indonesia

ABSTRACT: Uncertainties of factors that influence slope stability have been acknowledged by many geotech-
nical engineers, and these are drawn from the characterization of the geotechnical parameters of the slope. The
characterization is associated with a process of identifying the distribution of random variable values used in
the design of a single slope of an open mine, and also part of the probabilistic method which is an alternative
approach in estimating the stability of a slope with a Failure Probability value (FP). This paper explains the char-
acterization process using the Kolmogorov-Smirnov (K-S) method, which will determine the most appropriate
function for the distribution of the random variable. The characterization results can be used in determining the
Safety Factor including the level of confidence and Probability of Failure based on the best fitting function.

1 INTRODUCTION distribution functions, namely: normal, lognormal,


beta, gamma.
Slope design has over the years become the domain The results of the characterization process will be
of geotechnical engineers. While this has benefited used for the calculation of the safety factor and fail-
the technology of theslope design processes, it has ure probability of slopes, and the distribution data are
also separated the the mine design engineers from the obtained from Tutupan Coal Mine, South Kalimantan,
responsibility of the risk versus reward relationship. Indonesia.
Given the uncertainties that prevail within the
geotechnical discipline, there always exists a proba-
2 BASIC THEORY
bility that a slope may not perform aspredicted, and in
the worst case can result in a catastrophic failure.
2.1 Probability distribution function
The rock slope stability analysis affected by the ran-
dom input parameters such as internal friction angle, Probability distribution function describes the spread
cohesion, water pressure, seismic acceleration and of a random variable that is used to estimate the prob-
density, which cannot be properly represented by a sin- ability of the occurrence of a parameter value. It has
gle value as input parameters in slope stability analysis. typical and unique properties of distribution that make
The uncertainty in the values of rock mass properties one function different from others. But this does not
is a major factor in slope stability analysis. rule out the possibility that a distribution function is
Each input parameter has a certain distribution a derivative from another function. For example, the
function. There are several methods used in character- exponential distribution function is a special form of
izing a data distribution, namely: Chi-Square Method, the gamma distribution function withshape parameter
Kolmogorov-Smirnov method, and the method of (a) is 1. Likewise, a uniform distribution function is
Anderson Darling. This paper uses the Kolmogorov- a special form of the beta distribution function with
Smirnov method. The principle of this method is based shape parameter values α = 1 and β = 1. A triangu-
on a comparison between the experimental cumula- lar distribution function is also a special form of the
tive distribution function and cumulative distribution beta distribution function with shape parameter values
theoretical assumptions. α1 = 1 and α2 = 2.
The 4 assumptions of the cumulative distribu- This paper employs 4 assumed continuous distri-
tion function will be compared with the empirical bution functions, i.e. normal, lognormal, beta, and

213
Table 1. Assumed probability density function (Tse, 2009). Table 2. Critical value (Tse, 2009).

Distribution Probability density function Mean & Varians Significance level α 0.10 0.05 0.01

x−µ 2 1.22 1.36 1.63


 
1 −1
Normal fx (x) = √ e 2 σ
µ Critical value √ √ √
σ 2π σ2 n n n
−∞ < x < ∞
1
ln x−µ 2 µ+ σ2
 
1 − 21
Lognormal fx (x) = √ e σ
e 2
xσ 2π 2 2
e2µ+σ (eσ −1 )
x>0
xα1 −1 (1 − x)α2 −1 α1 rejected. The equation of this cumulative distribution
Beta f (x) = ;
B (α1 , α2 ) α1 + α 2 function is as follows:
0≤x≤1 α1 α2
f (x) =0, x others (α1 + α2 )2 (α1 + α2 + 1)2

xα−1 e−x/β
Gamma f (x) = ;x>0 αβ
βα Ŵ (α)
f (x) =0, x others αβ2

x1 , x2 , . . . , xn are the values of the sample data that is


already set, and n is the sample size.
In the K-S test, the maximum difference between
Sn (x) the and F(x) for all ranges of x is a measure of the
difference between assumed theoretical distributions
and experimental data distribution. The maximum
difference of both distributions can be written as
follows:

Theoretically Dn is a random variable whose distri-


Figure 1. Probability Distribution Function.
bution depends on n. To a certain significance level
α, K-S test compares the maximum difference of the
observations in above equation with the critical value
gamma. These are considered to represent the prop- Dαn . Table 2 describes the critical value with a particular
erties of the distribution of the random variables (see significance level α. Thus with a significance level α,
Table 1). the null hypothesis is rejected if Dn is higher than the
Figure 1 describes a Probability Density Function critical value. To facilitate this method assisted with
(PDF) and Cumulative Distribution Function (CDF). Matlab software Ver.7.12.0.
Probability density functionsexplains the area in which When all functions have a Dn value lower than the
the relative probability of a random number can be critical value, it can be retested using 2 indicators,
assumed to be a unique value compared to the other i.e. Akaike Information Criteria (AIC) and Bayesian
value. Failure probability (FP) is determined from Information Criteria (BIC). The Akaike information
the distribution of the safety factor. The area under criterion, denoted by AIC, proposes to penalize large
the curve (FK < 1) is defined as a failure probability models by subtracting the number of parameters in the
value. model from its log-likelihood. Thus, AIC is defined as:

2.2 Fitting distribution function


If a theoretical distribution has been assumed, then where log L(θ̂; x) is the log-likelihood and p is thenum-
the validity of the assumed distribution could be con- ber of estimated parameters in the model. Based on this
firmed or refuted by a statistical test with goodness approach the modelwith the largest AIC is selected.
of fit. There are several methods to test the goodness The above problem of the AIC can be corrected by
of fit of some assumed distribution functions, i.e. Chi imposing a differentpenalty on the log-likelihood. The
Square method, Kolmogrov-Smirnov (K-S) method, Schwarz information criterion, also calledthe Bayesian
and Anderson-Darling method. This paper only dis- information criterion, denoted by BIC, is defined as:
cusses the K-S method. The basic procedure of this
method includes a comparison between experimental
and theoretical cumulative frequency distribution of
the types of assumed distributions. If the difference of
both frequencies is higher than those of a certain sam- The details of these methods can be seen in Tse
ple size, the assumed theoretical distribution will be (2009).

214
2.3 Monte Carlo simulation
This method effectively simulates the response of the
SF performance function to randomly select discrete
values of the component variables. The process is
repeated many times to obtain an approximate discrete
PDF of the resulting SF values. The component ran-
dom variables for each calculation are selected from a
sample of random values that are based on the selected
PDF of the random variables. Although the PDFs can
take on any shape, the normal, lognormal, beta and
gamma distributions are among the most favored for
the analysis.
It should be noted that each Monte Carlo simula-
tion will use a different sequence of random values,
and the resulting probabilities, means, variances, and
histograms will be slightly different. As the number of
trials increases, the differences will become smaller. Figure 2. Empirical CDF.
Because the Monte Carlo simulation requires a
large number of calculations, it often has been viewed
as a less desirable option than the Taylor series or point
estimation method. However, with the general avail-
ability of faster desktop computers, the Monte Carlo
approach has gained wider acceptance due to its inher-
ent simplicity and the ability to use existing computer
programs with only minor modifications. Details of
this method can be read in Abramsom et al. (2002),
Baecher et al. (2003), and Tse (2009).

3 DATA COLLECTION & METHODOLOGY

Data gathering of the physical and mechanical prop-


erties of the lithology (coal, mudstone and sandstone)
of Tutupan coal mines, South Kalimantan, Indonesia
was carefully conducted. These properties include unit
weight, cohesion, and internal friction angle. The num- Figure 3. Overlay empirical and assumed theoretical CDF.
ber of data from each lithology consists of 10, 33 and
16 for coal properties data, mudstone properties data,
and sandstone properties data respectively. Table 3. The results of fitting test on cohesion of mudstone.
Furthermore, the fitting test of K-S method was
Statistical
done using Matlab software Ver.7.12.0. By using equa-
parameter Normal Lognormal Beta Gamma
tion [1], the empirical data representation can be done
by sorting the data from smallest to highest value, Mean 0.191 0.193 0.191 0.191
which produces a histogram in the form of CDF (See Dn 0.096 0.118 0.095 0.096
Figure 2). Critical value 0.237 0.237 0.237 0.237
In the K-S method, the empirical distribution func- Log-likelihood 46.111 42.632 45.543 45.147
tion was compared with the 4 assumptions in the AIC 44.111 40.632 43.543 43.147
theoretical distribution function, i.e. normal, lognor- BIC 44.593 41.113 44.024 43.629
mal, beta, and gamma. If the distribution function of
the data is not actually, then Dn has a considerable
value. Thus, with a significance level α = 5%, the null
hypothesis is rejected if Dn is greater than the√critical value smaller than the critical value, the hypothesis is
value. The critical value for α = 5% is 1.36/ n. For then accepted.
example, a fitting test was conducted on the cohesion Besides that, we also have to see the AIC and
distribution of mudstone (see Figure 3). The amount BIC values of the 4 assumed theoretical distribution
of data is 33 and α = 5%, the critical value is 0.237. functions. By using equation [3] and [4], AIC and
To declare the assumptions of the functions fit and the BIC values can be determined. The largest value of
hypothesis is accepted, the Dn value must be smaller that distribution is normal distribution, i.e. respec-
than the critical value. The results of the fitting test on tively 44.111 and 44.593. Thus, it can be said that the
cohesion distribution of mudstone indicates that the 4 most appropriate distribution function is the normal
assumed theoretical distribution functions have a Dn distribution (see Table 3).

215
Table 4. The result of fitting test in rock properties of Table 5. The results of failure probability.
tutupan coal mine.
Lithology Coal Mudstone Sandstone
Statistical c φ γ
parameter (MPa) (Deg.) (kN/m2 ) Slope Height (m) 15 15 15
Slope Angle (deg.) 70 70 70
Mudstone SF Deterministic 5.5 2.7 2.8
Mean 0.169 23.6 22.68 SF Mean 5.5 2.8 2.8
Variance 0.004 23.5 0.38 Cohesion (kPa) 241 169 191
Std. Deviation 0.061 4.8 0.62 Cohesion (kPa) 57 68 66
Dn 0.096 0.090 0.324 Back Analysis
Critical Value 0.237 0.237 0.237 Deterioration (%) 76.3 59.8 65.4
Log-likelihood 46.11 −98.38 −30.42 FP (%) 0 1.3 0
AIC 44.11 −100.38 −32.42 FP (%) Back Analysis 56.5 46.2 49.8
BIC 44.59 −99.82 −31.93
Distribution Normal Normal Normal
Sandstone
Mean 0.191 27.8 21.48 • The results of the characterization of the properties
Variance 0.003 33.1 – of 3 rock lithology produce all types of variables
Std. Deviation 0.057 5.8 – tested had a normal distribution, except for the
Dn 0.144 0.121 – cohesion of coal that has a gamma distribution.
Critical Value 0.340 0.340 – • Characterization results are used to estimate the
Log-likelihood 23.54 −50.20 – value of the safety factor following the data distribu-
AIC 21.54 −52.20 – tion, which would further strengthen the confidence
BIC 22.34 −51.40 –
Distribution Normal Normal –
level of the FK value.

Coal
Mean 0.241 34.1 12.36 ACKNOWLEDGEMENT
Variance 0.003 5.4 –
Std. Deviation 0.051 2.3 – The authors would like to thank the committee ISRM
Dn 0.190 0.120 –
Critical Value 0.430 0.430 –
which gives the opportunity to present this paper, and
Log-likelihood 15.77 −22.13 – PT Adaro Indonesia for the support in providing data.
AIC 13.77 −24.13 –
BIC 14.77 −23.13 –
Distribution Gamma Normal – REFERENCES
Abramson, L.W., Lee, T.S., Sharma, S. & Boyce, G.M. 2002.
Slope Stability and Stabilization Methods, second edition,
John Wiley & Sons.
Table 4 presents the summary of the characteriza- Ang, A.H-S., & Tang, W.H. 1984. Probability Concepts in
tion test of rock properties distribution on 3 lithology Engineering Planning and Design – Vol. II Decision, Risk
of rock. Only the cohesion of coalthat has gamma and Reliability John Wiley & Sons.
distribution, while others have normal distribution. Azizi, M.A., Kramadibrata, S., Wattimena, R.K. & Arif, I.
2010. Application of Probabilistic Approach on Slope
Stability Analysis, Annual Meeting of Indonesian Min-
4 DISCUSSION ing Professionals Association (PERHAPI) XIX 2010,
Balikpapan.
Azizi, M.A., Kramadibrata, S., Wattimena, R.K. & Arif, I.
Based on the characterization results using K-S 2011. Application of probabilistic approach on single
method, each fitted distribution was included in the slope stability analysis using limit equilibrium method,
calculation of the safety factor. Given any variation in National conference of Statistical 2011, organized by
the random variable that is used in the determination Statistical Department, Faculty of Science, University of
of SF, then FP was also be generated. Diponegoro, Semarang, Indonesia.
Characterization data can also be used to pre- Azizi, M.A., Kramadibrata, S., Wattimena, R.K., Suhedi &
dict failure probability when the slope failed. This is Basuki, S. 2011. Probabilistic approach on single slope
possible when there is a strength decrease (seeTable 5). stability analysis using Bishop Method, National Con-
ference of Mining 2011, organized by mining depart-
ment, faculty of engineering, University of Lambung
Mangkurat, South Kalimantan.
5 CONCLUSIONS Baecher, G.B., & Christian, J.T. 2003. Reliability and statis-
tics in geotechnical engineering. Wiley, Chichester, U.K.
Several points can be concluded from the results of Hoek, E., Factor of Safety and Probability of Failure, Chapter
this study, and is as follows: 8 – Rock Engineering.
Tse, Y.K. 2009. Nonlife Actuarial Models, Theory, Methods
• KS method is the most practical method to obtain and Evaluation, Cambridge University Press.
the most suitable distribution.

216

View publication stats

You might also like