Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Keywords: Although blasting is the most principal method of fragmentation in hard rock mining, the significance of
Backbreak the costs of blast induced rockmass damage in terms of mining efficiency and safety is becoming increas-
Regression model ingly recognized. Backbreak is one of the adverse phenomena in blasting operations that causes the insta-
Fuzzy model bility of mine walls, falling down of equipments, improper fragmentation, reduced efficiency of drilling,
Gol-E-Gohar iron mine
etc., and consequently increases the total cost of a mining operation. In this paper, predictive models
based on fuzzy set theory and multivariable regression have been developed for predicting backbreak
in Gol-E-Gohar iron mine of Iran. To evaluate performance of the employed models, the coefficient of cor-
relation (R2) and the root mean square error (RMSE) indices were calculated. It was concluded that per-
formance of the fuzzy model is considerably better than regression model. For the fuzzy and regression
models, R2 and RMSE were equal to 95.43% and 0.44 and 34.08% and 1.63, respectively. The fuzzy model
sensitivity analysis shows that the most effective parameters on backbreak phenomenon are stemming
length, hole depth, burden and hole spacing. Application of this model in the Gol-E-Gohar iron mine con-
siderably minimized backbreak and improved blasting efficiency.
Ó 2009 Elsevier Ltd. All rights reserved.
0957-4174/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2009.08.014
2638 M. Monjezi et al. / Expert Systems with Applications 37 (2010) 2637–2643
(UCS) of various rocks. Wu, Hao, and Zhou (1999) employed this oping fuzzy model rule base and the dataset used for examining
method to describe the damage threshold of a rockmass under dy- the results.
namic pressure of explosion. A methodology for slope stability
analysis using fuzzy logic was proposed by Dodagoudar and Venk-
3. Regression analysis
atachalam (2000). Klose (2002) described a simple approach for
geological interpretation of the seismic data utilizing fuzzy meth-
Multivariable regression is an extension of the regression anal-
od. Gokceoglu (2002) suggested a fuzzy triangular chart for the
ysis that incorporates additional independent variables in the pre-
prediction of UCS. Sonmez, Gokceoglu, and Ulusay (2003) applied
dictive equation. Utilizing this method, one can easily determine
fuzzy set theory to the geological strength index (GSI) to overcome
the relationship between the criterion (independent) variables
the uncertainties involved in the characterization of rockmasses.
and the predictive (dependent) variables (Eskandari, Rezaee, &
Den Hartog et al. (1997), Grima and Verhoef (1999) used the fuzzy
Mohammadnia, 2004; Jennrich, 1995). This method was utilized
logic approach to predict the performance and bit consumption of
in different mining fields (Finol, Guo, & Dong Jing, 2001; Gokceoglu
rock-cutting trenchers.
& Zorlu, 2004; Grima & Babuska, 1999). Here, backbreak is consid-
In this paper, a new fuzzy model was developed and applied to
ered to be a function of eight parameters i.e., burden, hole spacing,
predict backbreak in blasting operation in Gol-E-Gohar Iron mine
hole depth, specific drilling, stemming length, charge per delay,
of Iran. The simulation results were compared to the results of
rock density and powder factor. The above mentioned database
multivariable regression analysis on the basis of real practical data.
is used to predict backbreak in a blasting operation based on Eq.
(1). The regression model simulation result has been shown in
2. Case study Fig. 1.
The Gol-E-Gohar iron mine is located at 55 km southwest of Sir- Backbreak ¼ 10:72 þ 1:467B 0:112S þ 0:05907K
jan between 551150E and 551240E longitudes and 29130 N and þ 47:28SD þ 0:3328T þ 0:018525Cpd
29170 N latitudes. The deposit has been composed of 6 separate þ 0:2527D þ 3:105Pf ð1Þ
anomalies with an extension of approximately 10 km length and
4 km width. The total ore reserve of the Gol-E-Gohar mine is
1135 million tones.
4. Fuzzy sets
From the geological viewpoint, the mine is situated in the meta-
morphic rocks of Paleozoic that vertically consist of three parts i.e.,
The fuzzy theory started with the concept of fuzziness and its
lower, middle and upper. The lower part consist of successions of
expression in the form of fuzzy set introduced by Zadeh (1965).
gneiss, micaschist, amphibolite and quartzschist, while the middle
This theory is an extension of the concept of a crisp set. A crisp
part contains sequences of marble, micaschist, greenschist and
set only allows full membership or no membership to every ele-
graphitschist and finally the upper is composed of marble, dolo-
ment of a universe of discourse, whereas a fuzzy set allows for par-
mite and calcite.
tial membership (Finol et al., 2001). The process of generating
Blast design parameters of the mine are listed in Table 1. Also,
membership values for a fuzzy variable using membership func-
minimum and maximum values of the applied parameters in the
tions is defined as fuzzification. The shape of the membership func-
models and their respected symbols are given in Table 2. It should
tions can be either linear (trapezoidal or triangular) or non-linear
be noted that in the blasting operation, drilling cuttings are used as
depending on the nature of the problem to be solved (Acaroglu,
stemming material and delay time between the first and second
Ozdemir, & Asbury, 2008).
row is 80 ms where as it is 50 ms between the other rows.
The fuzzy set theory provides a systematic calculus to deal with
Database, collected from blasting operation of the Gol-E-Gohar
linguistic terms, and it performs numerical computation by using
iron mine, consists of two datasets viz., the dataset used in devel-
linguistic labels stipulated by membership functions. This theory
can also be used for developing rule-based models which combine
Table 1 expert knowledge and numerical data (Iphar & Goktan, 2006). Za-
Blasting pattern information of the Gol-E-Gohar iron ore mine. deh (1973, 1976) was the first to introduce the idea of analysis and
Parameter Description
system modeling using linguistic terms, and since then, it has been
the subject of considerable investigations (Mamdani & Assilian,
Explosive type ANFO
1975; Dubois & Prade, 1990; Dubois & Prade, 1991). Several
Blast hole pattern Staggered
Bench height 15 (m) rule-based fuzzy modeling methods have been proposed in the last
Hole diameter 0.203 (m)
Rows per blast 2–7
Hole per rows 10–20
Table 2
Description of the input and output parameters in the models.
two decades. According to the formation of the subsequent parts Gol-E-Gohar iron mine. Fuzzy model process is schematically
and the inference mechanism to compute the output of the model, shown in Fig. 2. To estimate backbreak, burden, hole spacing, hole
rule-based models can be classified into four main sets: fuzzy rela- depth, specific drilling, stemming length, charge per delay, rock
tional models (Pedrycz, 1984), linguistic models (Mamdani & Ass- density and powder factor are used as input parameters. Fig. 3
ilian, 1975), neural network based models (Jang, 1992; Lin & Lee, shows input and output variables of proposed fuzzy based predic-
1991), and Takagi–Sugeno–Kang (TSK) fuzzy models (Sugeno & tive model.
Kang, 1988; Takagi & Sugeno, 1985). There are several fuzzy infer- Normally fuzzification of input/output variables is fulfilled by a
ence systems that have been employed in various applications of membership function. In this paper, the most common types of
these models. The most commonly used are as follows: membership functions, triangular and trapezoidal, are used (Huang
& Siller, 1997). Therefore, they were applied for fuzzification the
Mamdani algorithm. input and output parameters of the proposed model.
Takagi–Sugeno–Kang fuzzy (TSK) algorithm. The membership functions of input and output parameters
Tsukamoto algorithm. were abbreviated and shown in the Figs. 4–12. In these member-
Singleton algorithm. ship functions, V stands for very, H for High, M for Medium, L for
Fig. 2. The proposed Fuzzy method block diagram for backbreak Prediction.
XI, Xr: Input variables.
AiI, Air, Bi: Linguistic terms (fuzzy sets).
Y: Output variables.
K: number of rules.
ð3Þ
where lC K ; lAK and lBK are the membership functions of output ‘‘z”
for rule ‘‘k”, input ‘‘x” and input ‘‘y”, respectively.
The rules are disjunctive so the aggregation operation max re- Fig. 3. Main structure of the fuzzy model.
sults in an aggregated membership function comprised of the outer
envelope of the individual truncated membership forms from each
rule. If a crisp value is needed for the aggregated output, some
appropriate defuzzification techniques should be employed to
the aggregated membership function (Ross, 1995). Defuzzification
is extraction of a representative crisp value from a fuzzy set. There
exist several defuzzification methods such as centroid of area
(COA), center of gravity, mean of maximum, smallest of maximum,
etc., from which the most commonly adopted is COA method (Gri-
ma, 2000; Hellendoorn & Thomas, 1993).
Fig. 5. Membership function of hole spacing. Fig. 9. Membership function of charge per delay.
Table 3
Samples of fuzzy if–then rules.
(COA) method (Grima, 2000; Hellendoorn & Thomas, 1993). Fig. 13 tion of backbreak can be made in any situations. The fuzzy model
shows the rule viewer and fuzzy reasoning mechanism in MATLAB simulation result data has been shown in Fig. 14.
environment.
The proposed fuzzy model can provide an accurate estimation of
backbreak when proper input data are used. For example, when 6. Model performance evaluation
input parameters are B = 5.5 m, S = 6.8 m, K = 17 m, SD = 0.028
m/m3, T = 6.2 m, Cpd = 87.2 kg/ms, D = 3.52 g/cm3 and PF = 0.29 To compare the performance of the regression model and the
kg/ton, model output (backbreak) would be 6.17 m (Fig. 13). Since new fuzzy model, two indexes were considered, coefficient of
the model has the ability of interpolating input parameters, predic- correlation R2 and root mean square error (RMSE) (Eqs. (4) and
2642 M. Monjezi et al. / Expert Systems with Applications 37 (2010) 2637–2643
Fig. 14. Comparison between the real and predicted backbreak for the fuzzy model.
Fig. 16. Strengths of relation (rij) between the backbreak and each input parameter.
Table 4
Performance indexes of the models. Table 5
Strengths value (rij) between the backbreak and each input parameter.
Index Fuzzy model Regression model
R2 95.43% 34.08% Pf D Cpd T SD K S B
RMSE 0.44 1.63 rij 0.882 0.866 0.741 0.903 0.761 0.899 0.884 0.885
X ¼ fX 1 ; X 2 ; X 3 ; . . . X m g ð6Þ
,vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u m
Fig. 15. Comparison of predicted backbreak with measured. X
m uX X m
rij ¼ xik xjk t x2 ik x2 jk ð8Þ
(5)) (Tzamos & Sofianos, 2006). The models performance indexes k¼1 k¼1 k¼1
were calculated and are summarized in Table 4. For testing the
models 115 data sets which were not incorporated in the model The strengths of relations (rij values) between the backbreak and in-
development were used. put parameters are shown in Fig. 16 and Table 5. As it is shown, the
2 32 most effective parameters on the backbreak are stemming length,
Pn
6 ðA ipred A pred ÞðA imeas A meas Þ 7 hole depth, burden and hole spacing.
R2 ¼ 1004qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
i¼1
ffi5 ; ð4Þ
2 Pn 2
i¼1 ðA ipred A pred Þ i¼1 ðA imeas A meas Þ
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 8. Conclusions
u n
u1 X
RMSEðAÞ ¼ t ðAimeas Aipred Þ2 : ð5Þ
n i¼1 The fuzzy inference system having sufficient application flexi-
bility seems to be a suitable measure to minimize the uncertainties
where, Aimeas is the ith measured element, Aipred is the ith predicted encountered during the blasting operations. In this study, a new
element, Aipred and Aimeas are the average of prediction and measured fuzzy model was established to predict backbreak in blasting oper-
sets, respectively, and n is the number of dataset. ation of the Gol-E-Gohar iron mine. The results of the model were
The real and predicted backbreak from both the fuzzy and compared with the results of multivariable regression analysis.
regression models for different patterns, are shown in Fig. 15. It was concluded that performance of the fuzzy model is consid-
The comparison shows the overall superiority of fuzzy based sys- erably better than the statistical model. For the fuzzy model R2 and
tem compared with the regression model. RMSE were equal to 0.95 and 0.44, respectively and for the regres-
sion method they were 0.34 and 1.63, respectively. Lower perfor-
7. Sensitivity analysis mance of the statistical method is probably due to linearity
assumption in this method. Also it was concluded that the most
To recognize the most sensitive factors affecting backbreak co- effective parameters on backbreak phenomenon are stemming
sine amplitude method (CAM) was utilized (Jong & Lee, 2004). To length, hole depth, burden and hole spacing. Application of the re-
apply this method, all of the data pairs are expressed in common sults of this study caused diminishing backbreak and increasing
X-space. The data pairs used to construct a data array X defined as: efficiency in blasting operation of the Gol-E-Gohar iron mine.
M. Monjezi et al. / Expert Systems with Applications 37 (2010) 2637–2643 2643