Documente Academic
Documente Profesional
Documente Cultură
net/publication/263496825
Predicting the production rate of diamond wire saws using multiple nonlinear
regression analysis
CITATIONS READS
13 802
5 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Saleh Ghadernejad on 20 February 2015.
Geosystem Engineering
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tges20
To cite this article: Golsa Sadegheslam, Reza Mikaeil, Reza Rooki, Saleh Ghadernejad & Mohammad Ataei , Geosystem
Engineering (2013): Predicting the production rate of diamond wire saws using multiple nonlinear regression analysis,
Geosystem Engineering, DOI: 10.1080/12269328.2013.856276
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Geosystem Engineering, 2013
http://dx.doi.org/10.1080/12269328.2013.856276
Predicting the production rate of diamond wire saws using multiple nonlinear regression analysis
Golsa Sadegheslama*, Reza Mikaeila, Reza Rookib, Saleh Ghadernejada and Mohammad Ataeic
a
Department of Mining and Metallurgical Engineering, Urmia University of Technology, Urmia, Iran; bFaculty of Mining Engineering,
Birjand University of Technology, Birjand, Iran; cFaculty of Mining, Petroleum and Geophysics, Shahrood University of Technology,
Shahrood, Iran
(Received 28 June 2013; accepted 13 October 2013)
In this paper, the prediction of production rate of diamond wire saws was performed. Performance measurements of
diamond wire saws were conducted on 14 different carbonate rocks in mines located in Iran. Rock samples were collected
from the mines for laboratory tests. Elasticity modulus, quartz content, and uniaxial compressive strength (UCS) were
determined in the laboratory. Production values were correlated with the corresponding elasticity modulus, quartz content,
and UCS. The sawing speed and rock properties were evaluated using multiple nonlinear regression analysis, and the
developed model was validated by statistical tests and actual data of sawing speed for four new mines. It was concluded that
the sawing speed of carbonate rocks using diamond wire saws can reliably be estimated using the developed model.
Downloaded by [217.219.149.148] at 09:30 23 November 2013
Keywords: sawability; diamond wire saw; carbonate rocks; statistical analysis; neural network
Figure 1. Schematic presentation of a diamond wire-cutting layout and cross section of a diamond bead.
4. Laboratory studies
Sawing cost is controlled mainly by the sawing rate, which
is heavily influenced by the performance of the saw. A
range of factors affects the performance characteristics of
the saws. Factors that affect cutting efficiency in the
diamond wire cutting method are given in Table 3 (Özc elik,
1999). The effective rock parameters on the performance of
Figure 2. Mechanical interaction between saw and stone during diamond wire saw can be classified into three main groups:
sawing process (Tonshoff, Hillmann-Apmann, & Asche, 2002). (a) physical and mechanical parameters, (b) abrasive
Note: 1 – friction between swarf and matrix; 2 – matrix erosion
by swarf and chip; 3 – friction between stone and grain; 4 – characteristics, and (c) elasticity and plasticity character-
plastic deformation; 5 – elastic deformation; and 6 – primary istics. For assessing of rock sawability and prediction of
chipping zone. production rate, using some major rock properties is very
Geosystem Engineering 3
Table 2. Operational parameters of wire saw machine. weathering, texture, and matrix type. The UCS tests were
carried out on testing machine with a capacity of 2000 kN
Parameter Description
at a loading rate of 1 kN/s. Cylindrical specimens NX in
Main motor power (kW) 45 diameter with a length to diameter ratio of 2.5:1 were used.
Length of wire (m) 65 – 80
Linear speed (m/s) 30 – 35
Rotator diameter (cm) 60
Beads per meter 33 – 36 4.2. Equivalent quartz content
Beads type Special for soft rocks Quartz content is one of the most important factors to
evaluate the abrasiveness and hardness of rock. A typical
thin section belonging to each rock type was prepared for
petrographic analyses and the determination of textural
essential. But there is an important question; what are the rock characteristics. Primary and secondary minerals were
most important parameters? In this study, for prediction of identified. It is clear that tool wear and rate reduction are
wire saw’s production rate, one parameter from each predominantly a result of the mineral content harder than
mentioned family was selected. Uniaxial compressive steel (Mohs hardness ca. 5.5), especially quartz (Mohs
strength (UCS) as the most important physical and hardness of 7). To include all minerals of a rock sample,
mechanical properties of rocks and equivalent quartz the equivalent quartz content has been determined in thin
Downloaded by [217.219.149.148] at 09:30 23 November 2013
content as the most important abrasivity property and sections by modal analysis – meaning the entire mineral
elasticity modulus as the most important elasticity property content referring to the abrasiveness or hardness of quartz
have been selected and used in regression analyses (Equation (1)). Therefore, each mineral amount is
simultaneously. Rock blocks were collected from the multiplied with its relative Rosiwal abrasiveness to quartz
mines for laboratory tests. An attempt was made to collect (with quartz being 100) (Thuro, 1997).
rock samples that were large enough to obtain all of the test
specimens of a given rock type from the same piece. Each
block sample was inspected for macroscopic defects so that X
n
Table 3. Factors affecting cutting efficiency in the diamond wire saw method (Özc elik, 1999).
Production (m2/h)
8
5. Statistical analysis 4
Production (m2/h)
P ¼ 23:2904LnðQcÞ þ 0:1551; R ¼ 0:78;
2
ð2Þ
Downloaded by [217.219.149.148] at 09:30 23 November 2013
8
20:3914E
P ¼ 15:21e ; R ¼ 0:60;
2
ð3Þ 6
4
P ¼ 210:567LnðUCSÞ þ 75:648; R 2 ¼ 0:53; ð4Þ
2
2
where P is the production (m /h), Qc is the quartz content 0
(%), E is the elasticity modulus (GPa), and UCS is the 0 10000 20000 30000 40000 50000 60000
uniaxial compressive strength (MPa). There is a strong Elasticity modulus (MPa)
correlation between quartz content, elasticity modulus, and
Figure 4. Production versus elasticity modulus.
production values (Figures 3 and 4). The correlation
between production and UCS values is weaker than two
out by using multiple regression methods. Multiple
other parameters relatively (Figure 5).
regression methods can be divided into two types as
linear and nonlinear methods. In this study, the twin-
logarithmic model, which is the one of the nonlinear
5.2 Multiple nonlinear regression analysis
methods, was used. The equation representing the model
Multiple nonlinear regression (MnLR) analysis was can be written in the following form:
performed with the expectation of obtaining more
significant relations than that of the simple regression. Y ¼ aX b11 X b22 · · ·X bnn ; ð5Þ
Because the sawability of rocks is affected by many
factors, sawability cannot be analyzed by using simple where Y is the predicted value corresponding to the
regression models. Therefore, the analysis must be carried dependent variables, a is the intercept, X1, X2, Xn are the
Mine name and location Production (m2/h) Qc (%) E (GPa) UCS (MPa) Wac (%) LA (%)
Godarsorkh/Esfahan 1.28 0.4 5.45 96 0.34 34.5
Hamzadeh/Esfahan 6 0.2 2 93.8 0.39 31.1
Joshghan qali/Esfahan 5 0.15 2.5 97.5 0.22 29.9
Laybid/Esfahan 7 0.15 2.5 87.1 0.5 42.2
Kandab/Esfahan 3 0.63 2.3 94.2 0.34 26.6
Jajarmi/Esfahan 1.5 0.45 5.2 96.9 0.36 34.5
Senesar/Kordestan 2.4 0.45 3 96.5 0.45 27.4
Shanoreh/Kordestan 10 0.075 2.3 59.0 0.32 27.4
Vihaje/Kordestan 10 0.05 2 59.15 0.26 33.3
Poshtbaskoli/Kordestan 6 0.25 3.7 64.0 0.33 35.6
Crysal/Kordestan 3.6 0.45 4 68.0 0.26 27.5
Sorti/Yazd 3.7 0.512 3.7 81.7 0.75 20
Baiatneyriz/Yazd 3.7 0.38 4.5 95.4 0.25 47
Angorakdehbid/Yazd 2.5 0.22 5.2 84.6 0.28 28
Geosystem Engineering 5
8
both variables are normally distributed and the obser-
6
vations are chosen randomly. The test compares computed
4 t-value with tabulated t-value using the null hypothesis. If
the computed t-value is greater than tabulated t-value, the
2
null hypothesis is rejected. This means that R is significant.
0 If the computed t-value is less than the tabulated t-value,
40 50 60 70 80 90 100 110 120
Uniaxial compression strength (MPa)
the null hypothesis is not rejected. In this case, R is not
significant. Since a 90% confidence level was chosen in this
Figure 5. Production versus uniaxial compression strength. test, a corresponding critical t-value 1.363 for the model
was obtained. As it is seen in Table 5, the computed t-
independent variables, and b1, b2, bn are the regression
values are greater than tabulated t-values for the best
coefficients of X1, X2, Xn.
models, suggesting that the models are valid. To test the
Taking logarithms of both sides of Equation (5)
significance of the regressions, analysis of variance was
converts the model into linear form as follow:
Downloaded by [217.219.149.148] at 09:30 23 November 2013
coefficient (R 2 adj)
neural networks. ANN has gained an increasing popularity
determination
in different fields of engineering in the past few decades,
Adjusted
because of their capability of extracting complex and
0.798
0.798
nonlinear relationships. Their mechanism is based on the
following assumptions (Hagan, Demuth, & Beale, 1996):
(1) Information processing occurs in many simple
elements that are called neurons (processing
coefficient (R 2)
elements).
Determination
0.94
tion links.
(3) Each connection link has an associated weight,
which, in a typical neural network, multiplies the
signal being transmitted.
(4) Each neuron applies an activation function
Tabulated
F-ratio
7.21
52.198
15.68
1.363
2 4.645
2 5.313
2 4.572
12.621
t-value
2.76
0.787
1.004
0.142
0.263
0.479
1.315
0.189
0.017
error
1.45
2 6.736
2 1.007
2 0.075
Constant
inputs.
UCS
UCS
Qc
Qc
Equation (9)
Table 6. The results of laboratory tests and production rate for test data.
Estimated production
(m2/h)
Mine name and location P (m2/h) Qc (%) E (GPa) UCS (MPa) MnLR MLR
Chahmorghidehbid/Yazd 2.7 0.932 5.2 84.0 2 21.2
Hasanabaddehbid/Yazd 3 0.1 4.2 114.3 3.13 3.12
Sobheneyriz/Yazd 4.2 0.61 2.4 80.0 4.1 4.07
Vahdatneyriz/Yazd 3.75 0.632 2.4 80.0 4.06 3.93
2
Pn
i¼1 Y i exp 2ðDi =2s Þ
2 2
YðXÞ ¼ Pn : ð11Þ
i¼1 exp 2ðDi =2s Þ
1 2 2
Table 7. The estimated and observed production values for all data.
14 14
12
12
Estimated production (m2/h)
10
10
8
8
6
6
4
4
2 Outlier
2
0
0 2 4 6 8 10 12 14
0
0 2 4 6 8 10 12 14 –2
Observed production (m2/h) Observed production (m2/h)
Figure 8. The estimated production values versus the observed Figure 9. The estimated production values versus the observed
production values (MnLR). production values (MLR).
6.2. Prediction of production rate using GRNN correlation coefficient and error between real
In this work, the inputs are Quartz content, elasticity value and prediction value in training data-set and
modulus, and compression strength, and the output is the test data-set. The results are given in Table 8. In
production rate. To predict the production rate, a GRNN this table, R is the correlation coefficient between
was designed in the MATLAB environment that this observed value and predicted value in the train
network has three layers (input, hidden, and output layer). data and test data. E is average relative error that is
The number of neurons in the hidden layer are equal with obtained as Equation (12):
number of training example. For achievement of the
optimum structure, the below steps were made:
(1) Scaling data (inputs and output) between the range 1X n
jti 2 yi j
E¼ ; ð12Þ
of 2 1 and 1 and selecting the number of train (14 n i¼1 ti
no.) and test data (4 no.).
(2) Training the network by different smooth factors where t is observed value, y is predicted value, and n is
to gain the optimum smooth factor according to number of samples in train data or test data.
Geosystem Engineering 9
7. Conclusions
Diamond wire saws are one of the most important
Figure 10. Architecture of GRNN (Demuth and Beale, 2002). machines used in the stone processing plants. Performance
Note: R, no. of elements in input vector; Q, no. of neurons in layer prediction of these saws is important in the cost estimation
1; Q, no. of neurons in layer 2; Q, no. of input/target pairs. a1i is
ith element of a1 where IW1,1 is a vector made of the ith row of and the planning of the plants. A correct estimation of
IW1,1. sawability helps to make the planning of the stone sawing
projects more efficient. The prediction sawing perform-
ance of diamond wire saw is a new research for
According to Table 8, the optimum smooth factor (SF) determination sawability of carbonate rocks. In this
Downloaded by [217.219.149.148] at 09:30 23 November 2013
was selected 0.66. This network has three layers: input study, the possibility of estimating sawability of carbonate
layer with 3 neurons, hidden layer with 14 neurons, and rocks using multiple nonlinear and linear regression
out put layer with 1 neuron. By this smooth factor, the slab analysis and general regression neural network was
production was predicted in the train and test data-set. For investigated. For this purpose, performance measurement
compression of predicted value and real value in train of diamond wire saws and rock mechanics tests on rock
data-set and test data-set, the correlation coefficient of two samples were carried out. Rock quartz content, elasticity
sets are given in Figures 11 and 12. modulus, and UCS are suggested for the estimation of the
The results show that the correlation coefficients in train sawability of carbonate rocks. Then, regression analysis
data and test data-set are 0.97 and 0.93, respectively. Notice was carried out and stepwise regression method was used
that the network did not has see the test data-set in the to obtain the best model for estimating sawability. To
training step. These results show high ability of neural check the validity of the derived model and GRNN model,
Table 8. Correlation coefficient values and means relative error with different smooth factors in train and test data-set.
Smooth factor (SF) R (train data) R (test data) E (train data) E (test data)
0.12 1 0.409 0.004 0.291
0.15 1 0.409 0.007 0.291
0.18 1 0.407 0.011 0.290
0.21 0.999 0.403 0.018 0.288
0.24 0.998 0.397 0.030 0.282
0.27 0.996 0.392 0.047 0.273
0.30 0.994 0.395 0.065 0.259
0.33 0.991 0.413 0.085 0.243
0.36 0.988 0.451 0.103 0.224
0.39 0.985 0.509 0.120 0.204
0.42 0.982 0.583 0.134 0.184
0.45 0.980 0.663 0.146 0.163
0.48 0.978 0.739 0.158 0.143
0.51 0.976 0.802 0.169 0.124
0.54 0.975 0.849 0.179 0.106
0.57 0.974 0.883 0.188 0.089
0.60 0.973 0.907 0.198 0.073
0.63 0.973 0.922 0.209 0.065
0.66 0.972 0.933 0.220 0.064
0.69 0.972 0.940 0.232 0.063
0.72 0.972 0.945 0.244 0.063
0.75 0.971 0.949 0.256 0.067
0.78 0.971 0.951 0.268 0.075
0.81 0.971 0.952 0.280 0.082
0.84 0.970 0.953 0.291 0.089
10 G. Sadegheslam et al.
Acknowledgements
6.00
The comments received from and the enlightening discussions
4.00 with our anonymous reviewers improved the paper and are
appreciated.
2.00
References
0.00 Ataei, M., Mikaiel, R., Sereshki, F., & Ghaysari, N. (2012).
0 2 4 6 8 10 Predicting the production rate of diamond wire saw using
Observed production (m2/h) statistical analysis. Arabian Journal of Geosciences, 5,
1289– 1295.
Figure 11. The estimated production values versus the observed Biasco, G. (1993). Diamond wire for quarrying hard rocks.
production values (ANN). Industrial Diamond Review, 5, 252– 255.
Bortolussi, A., Ciccu, R., Manca, P. P., & Massacci, G. (1990).
Simulation and optimization of rock cutting with diamond
wire. In International symposium APCOM, Berlin
Downloaded by [217.219.149.148] at 09:30 23 November 2013
diamond wire cutting method. In G. N. Panagiotou & T. N. Thuro, K. (1997). Drillability prediction—Geological influences
Michalakopoulos (Eds.), 9th mine planning and equipment in hard rack drill and blast tunneling. Geol Rundsch, 86,
selection symposium (pp. 661–666). Rotterdam: Balkema. 426– 438.
Özc elik, Y., Kulaksız, S., & Çetin, M. C. (2002). Assessment of Tonshoff, H. K., Hillmann-Apmann, H., & Asche, J. (2002).
the wear of diamond beads in the cutting of different rock Diamond tools in stone and civil engineering industry:
types by the ridge regression. Journal of Materials Cutting principles, wear and applications. Diamond and
Processing Technology, 127, 392– 400. Related Materials, 11, 736–741.
Specht, D. F. (1991). A general regression neural network. IEEE Wright, D. N. (1992, October). Marble and granite quarries turn
Transactions of Neural Networks, 2, 568–576. to diamond wire. Stone World, 77 – 80.
Downloaded by [217.219.149.148] at 09:30 23 November 2013