You are on page 1of 10

A Simulation Approach to Predict

Blasting-Induced Flyrock and Size of


Thrown Rocks

Edy Tonnizam Mohamad


Associate Professor (Dr.), Universiti Teknologi Malaysia, Department of
Geotechnics and Transportation,Faculty of Civil Engineering, 81310 UTM
Skudai, Johor, Malaysia;e-mail: edy@utm.com

Danial Jahed Armaghani


PhD Student, Universiti Teknologi Malaysia, Department of Geotechnics and
Transportation,Faculty of Civil Engineering, 81310 UTM Skudai, Johor,
Malaysia; e-mail: danialarmaghani@yahoo.com

Mohsen Hajihassani
Researcher, Universiti Teknologi Malaysia, Department of Geotechnics and
Transportation,Faculty of Civil Engineering, 81310 UTM Skudai, Johor,
Malaysia; e-mail: mohsen_hajihassani@yahoo.com

Koohyar Faizi
Postgraduate Student, Universiti Teknologi Malaysia, Department of Geotechnics
and Transportation,Faculty of Civil Engineering, 81310 UTM Skudai, Johor,
Malaysia; e-mail: koohyar.faizi@yahoo.com
Aminaton Marto
Professor, Universiti Teknologi Malaysia, Department of Geotechnics and
Transportation,Faculty of Civil Engineering, 81310 UTM Skudai, Johor,
Malaysia; e-mail: aminaton@utm.my

ABSTRACT
Bench blasting is the most common method of rock excavation in quarries and surface mines.
Blasting has some environmental impact such as ground vibration, airblast, dust and fumes
and flyrock. One of the undesirable phenomena in the blasting operation is flyrock, which is a
propelled rock fragment by explosive energy beyond the blast area. Prediction of flyrock
distance and size of the thrown rocks is a remarkable step in reduction and controlling the
blasting accidents in blasting operations. Different empirical models have been developed to
predict flyrock distance. However, due to complex relationships between blasting parameters
and flyrock phenomena, empirical methods cannot take into account all of the relevant
parameters. Artificial neural networks have revealed as valid approaches to analyze
geotechnical problems and are mainly able to cover the limitation of the existing approaches.
This paper presents an approach based on artificial neural network to predict flyrock distance
and size of the thrown rocks in blasting operations. The obtained results demonstrate that
artificial neural approach is applicable to predict flyrock distance and size of the thrown rocks
in blasting operations.
KEYWORDS: Blasting, Flyrock, Thrown rocks, Artificial Neural Networks.

- 365 -
Vol.18 [2013], Bund. B 366

INTRODUCTION
The rock is drilled, blasted, and extracted from quarries to produce crushed stone. Blasting is
the most common method to breakage rocks which is the controlled use of explosives. Blasting
has some environmental impact such as ground vibration, air blast, dust and fumes and flyrock.
As an environmental hazard flyrock causes serious injury to people, damage to equipment,
building and property. The main purpose of the blasting operation is to create desired fragment
size distribution leading to optimize the general mine economics (Monjezi et al. 2007). Blasting
affects the rock fragmentation, so there must be some beneficial measures to decrease undesirable
occurrences such as flyrock. Moreover, with diversity parameters which are effecting in the
blasting process, the simultaneous optimization of flyrock and rock fragmentation might not be an
easy task to perform (Singh and Singh 2005; Khandelwal and Singh 2006).
To develop a blast design, lowest possible cost should be considered to accomplish the
blasting operation. Moreover, it is necessary to satisfy the technical and safety requirements of
the blasting operations. However, variety of the parameters affect the operation’s out-comes that
some of them are out of designer’s control. These parameters can be categorized into three
groups; blast geometry, rock mass properties and explosive specifications (Zhu et al. 2008;
Thornton et al. 2002). Blasthole diameter, stemming, burden priming, spacing, and delay timing
are some of the parameters affecting the dynamics involved in the rock fragmentation.
One of the undesirable phenomena in the mining blasting operation is flyrock, which is
propelled rock fragments by explosive energy beyond the blast area (Rezaei et al., 2010). Flyrock
caused by mismatch between distribution of the explosive energy, mechanical strength of the rock
mass and charge confinement (Bajpayee et al., 2004). Flyrock is caused by inadequate burden,
insufficient stemming, inaccurate drilling, excessive powder factor, unfavorable geological
conditions, inappropriate delay timing and existence of loose rock in the blasthole collar
(Bhowmik et al. 2004; Fletcher and D’Andrea 1986; Massey and Siu 2003).
Different empirical models have been developed to predict flyrock distance and
fragmentation size. Relationship between mean fragment size and specific charge was developed
by Kuznetsov (1973). Cunningham (1983) presented a new model to predict rock fragmentation
based on Kuznetsov model. A fundamental empirical equation to predict flyrock distance based
on hole and rock diameter has been proposed by Lundborg et al. (1975):
Lm = 260 D 2/3
where Lm is maximum throw in meters and D is hole diameter in inches. This formula has
become the basis of other research presented by lundborg (1981), Adhikari (1999), Bajpayee et
al. (2004), and Kecojevic and Radomsky (2005).
Artificial Neural networks (ANNs) are used in advanced diversity applications of civil
engineering as a processor information (Hajihassani et al. 2011). Several research have been
conducted to predict flyrock distance using ANN as can be seen in Aghajani Bazzazi et al.
(2007), Monjezi et al. (2009), and Tonnizam Mohamad et al. (2012). This study presents the
prediction of flyrock distance and the size of the thrown rocks by means of ANN considering the
obtained data from a granite quarry site in Malaysia.

ARTIFICIAL NEURAL NETWORKS


An ANN is a mathematical model inspired by biological neural networks. A particular ANN
network defined by Simpson (1990) in three fundamental components; transfer function, network
architecture and learning law. Furthermore, to solve the problem these components have to be
defined. It consists of neurons with weighted connections in different cascaded layers including
an input layer, hidden layer(s) and an output layer with respect to a multilayer perceptron (MLP)
Vol.18 [20
013], Bund
d. B 367

feeddforward neu ural network. With prioriity of interprreting new innformation, the ANN model m has
to be trained. Foor training of neural netw works, differrent algorithm ms have beeen suggestedd, but the
backkpropagation n algorithm isi the most accomplished
a d and sturdy technique thhat provides the most
efficcient learninng proceduree for MLP neuraln netwoorks (Tawaddrous and Kaatsabanis, 20007). To
trainn an ANN model, an experimentaal database including suufficient num mber of daatasets is
requuired. Predicttions can be made for a new n input, while
w the trainning process is completedd dataset
(Taw wadrous and d Katsabanis,, 2005). The network is ppresented with a pair of patterns
p baseed on the
trainning processs: an input pattern
p and the correspoonding desirred output pattern.
p The network
commputes its acctual outputss by using th he given hisstorical inpu uts, weights and a mathhematical
funcction model threshold.
t Thhereafter, thee actual outpput is compaared with the historical ouutputs to
deteermine the output
o error (Demuth et al., 2007). Subsequently, this errorr is back proopagated
through the netw work to updatte the weightts. Thereforee, the generatted error wou uld be reduced in the
nextt iteration. Thhis iterated pprocedure is performed thhrough the trraining data. In this proceess, both
the ttraining and the testing errors
e are deccreased. The testing errorr starts to inccrease after a number
of itterations wh hich indicate overtrainingg; hence, thee training prrocess shoulld be stopped at this
junccture. Figure 1 shows the overall proccedure of AN NNs.

Figure 1:: Typical proocedure of AN


NNs

DATA
A COLLE
ECTION
N
In order to predict
p flyroock and size of the throwwn rocks a number
n of 399 blasting opperations
havee been invesstigated from m Ulu Tiram granite quaarry site in Malaysia.
M Figgure 3 shows overall
vieww of the blastting site. Uluu Tiram is loocated approxximately 18 km to the no orth of Johorr. During
dataa collection blasting
b paraameters incluuding ratio of
o burden to o spacing, steemming, holle depth,
holee diameter, powder
p factoor, rock density, subdrillling, numberr of rows, and a charge per p delay
weree carefully reecorded. In addition,
a flyrrock distancee and size off the thrown rocks
r were measured
m
in thhat particulaar site. In ordder to measu ure flyrock distance,
d thee bench surfaface was coloored and
threee video cameras were pllaced for pro ojection of flyyrock. Following the blaasting operattions, the
movvies recorded d by cameras were revieewed and thee maximum distance of the flied roccks were
deteermined. Usinng the measuuring tape th he flyrock distances and the diameterr of the throw wn rocks
weree determinedd.
Vol.18 [2013], Bund. B 368

Figure 3: Overall view of the blasting site

Kaolin Input and output parameters


In order to utilize the ANN method to predict flyrock distance and size of the thrown rocks
in blasting operations, Ulu Tiram granite quarry site in Malaysia were investigated. To train and
verify the ability and accuracy of the ANN method, 39 datasets were collected from this site.
Each dataset involves 9 input parameters including the most influential factors on flyrock
distance and size of the thrown rocks. Furthermore, each dataset consists of 2 output parameters
including the magnitude of flyrock distance and size of the thrown rocks measured during
blasting at the mentioned site. The range of the various input and output parameters considered
for developing neural network are shown in Table 1.

Table 1: Range of the measured blasting parameters


obtained from Ulu Tiram granite quarry site
Parameter Unit Description Min Max Mean

Hole Diameter cm Input 8.9 15 11.8


Hole Depth m Input 10 27 16.4
Subdrilling cm Input 24 50 36.5
Burden to Spacing ratio Input 0.63 0.86 0.76
Stemming m Input 1.75 3.6 2.26
Powder Factor kg/m3 Input 0.27 0.65 0.49
Number of Rows - Input 3 7 4.9
Charge per Delay kg Input 60 268 126
Rock Density g/cm3 Input 2.3 2.8 2.57
Flyrock Distance m Output 60 395 233.4
Fragment Size cm Output 2.3 6.5 4.15
Vol.18 [2013], Bund. B 369

MODELLING PROCEDURE
The modelling performance using ANNs is stoutly related to the architecture of the selected
network. Therefore, defining the optimum architecture of the network is a significant part of
ANNs designing. In ANN models, the number of nodes in the input and output layers are
determined by the relative input and output parameters, correspondingly. Therefore, the
architecture of networks including the number of hidden layers and the number of nodes in each
hidden layer should be determined. The optimum network architecture usually is determined
based on the mean squared error and regression values. In other words, using the trial and error
method several architecture are trained using input and output data and the network that produces
superior results is selected as the optimum network.
With the aim of determining the optimal network architecture to simulate blasting-induced
flyrock distance and size of the thrown rocks, following the trial and error method, several
networks were trained and tested. According to Hornik et al. (1989) an ANN with one hidden
layer can approximates any continuous function. Therefore, several networks with one hidden
layer and different number of nodes in hidden layer were trained and tested and the results were
compared. Assessments of the networks’ performance were conducted based on the obtained
correlation coefficients that indicated the correlation between actual and predicted outputs. Figure
4 shows the obtained correlation coefficients for training, validation and testing data, trained
using different ANN models. As can be seen in the figure the model with 20 nodes in the hidden
layer presents better correlation. The correlation coefficient for the selected model is 99 percent
for training datasets, and 97 percent for validation datasets and 95 percent for testing datasets.

Figure 4: Correlation coefficient for training, validation and testing data trained using different
ANN modes
The correlation coefficient for all data trained using different ANN modes are shown in
Figure 5. According to the figure the network with 20 nodes in the hidden layer yields greater
correlation coefficient. Therefore the network with one hidden layer and 20 nodes in the hidden
layer was selected. Figure 6 shows the selected architecture of neural network containing three
layers for predicting blasting-induced flyrock and size of the thrown rocks.
Vol.18 [20
013], Bund
d. B 370

Figuree 5: Correlattion coefficieent for all data trained using differentt ANN modees

Figuree 6: Structuree of the seleccted networkk to predict fllyrock induceed by blastinng


Vol.18 [2013], Bund. B 371

RESULTS AND DISCUSSION


The ANN toolbox of MATLAB software was used for designing, training and testing the
models. The presented model was composed of 9 input parameters, one hidden layer with 20
nodes in it, and 2 output parameters. A total of 39 dataset were collected from Ulu Tiram granite
quarry site in Malaysia. To determine the optimal network architecture, 9 different models were
trained with 80 percent of datasets were utilized in training and evaluating the networks. The rest
of data were randomly selected to evaluate the performance of the networks.
The performance of the selected network was evaluated based on the correlation coefficient
obtained during training, validation and testing processes. Figures 7 and 8 show the linear
regression between predicted values obtained by proposed ANN model and measured values. The
results indicate high coefficient of correlation between measured and predicted values.

Figure 7: Performance of the selected model for training (left) and validation (right) datasets

Figure 8: Performance of the selected model for testing (left) and all (right) datasets
Vol.18 [2013], Bund. B 372

As can be seen in Figures 7 and 8, the predicted values are reasonably matched to the actual
values for training datasets, whereas a close relationship exists between actual and predicted
values for validation and testing datasets. The results reveal that the network has performed with
negligible error and noticeably desirable regression.

CONCLUSIONS
A backpropagation ANN model has been presented to predict the flyrock induced by blasting
and also size of the thrown rocks. In order to define an optimum model, a series of analyses were
conducted to determine the optimum architecture of ANN using the trial and error method.
Eventually, a model with 1 hidden layer and 20 nodes in hidden layer was selected. To train the
presented ANN model, 39 datasets were collected from Ulu Tiram granite quarry site in
Malaysia. It should be noted that the range of applicability of the presented model is limited by
the data used in the learning step. Therefore, outside of the trained range this network may not be
accurate enough. According to the adequate consistency observed between the predicted results
obtained from presented model and measured data, it can be concluded that this approach is
applicable to predict flyrock distance and size of the thrown rocks in blasting operations.

ACKNOWLEDGEMENTS
This research is being carried out using the Universiti Teknologi Malaysia (UTM) financial
support. The authors would like to thanks UTM for the research grant and facilities available in
the Department of Geotechnics and Transportation, Faculty of Civil Engineering. The authors are
also indebted to Syarikat Matrajaya Sdn. Bhd., for allowing this research to be carried out at their
premise.

REFERENCES
1. Adhikari, G.R. 1999 Studies on flyrock at limestone quarries. Rock Mechanics and
Rock Engineering, 32(4): 291-301.
2. Aghajani Bazzazi, A., Osanloo, M., and Azimi, Y. 2007 Flyrock prediction by
multiple regression analysis in Esfordi phosphate mine of Iran. Rock Fragmentation
by Blasting, 649-657, London: Taylor and Francis Group.
3. Bajpayee, T.S., Rehak, T.R., Mowrey, G.L., and Ingram, D.K. 2004 Blasting injuries
in surface mining with emphasis on flyrock and blast area security. Journal of safety
Research, 35(1), 47-57.
4. Bhowmik, S., Raina, A.K., Chakraborty, A.K., Ramulu, M., Sahu, P. B., and Haldar,
A. 2004 Flyrock prediction and control in opencast mines: a critical appraisal. Eng
Min J, 5(6):10.
5. Cunningham, C. V. B. 1983 The Kuz-Ram model for prediction of fragmentation
from blasting. In: Proceedings of first international symposium on rock fragmentation
by blasting. Lulea, Sweden, p. 439–53.
6. Demuth, H., Beal, M., and Hagan, M. 2007 Neural network toolbox user’s guide.
Natick, MA: The Mathworks.
7. Fletcher, L. R., and D’Andrea, D. V. 1986 Control of flyrock in blasting. In:
Proceedings of 12th conference explosives and blasting technique. Atlanta. p. 167–
77.
Vol.18 [2013], Bund. B 373

8. Hajihassani, M., Marto, A., Namazi, E., Alavi, S.V., and Karimi Shahrbabaki, M.
2011 Prediction of Surface Settlements Induced by NATM Tunnelling Based on
Artificial Neural Networks. Electronic Journal of Geotechnical Engineering (EJGE),
Vol 16, pp. 1471-1480.
9. Hornik, K., Stinchcombe, M., and White, H. 1989 Multilayer Feedforward Networks
are Universal Approximators, Neural Networks, 2, 359-366.
10. Kecojevic, V., and Radomsky, M. 2005 Flyrock phenomena and area security in
blasting-related accidents . Safety Science, 43, 739-750.
11. Khandelwal, M, and Singh, T. N. 2006 Prediction of blast induced ground vibrations
and frequency in opencast mine—a neural network approach. J Sound Vib,289:711–
25.
12. Kuznetsov, V. M. 1973 The mean diameter of the fragments formed by blasting rock.
Sov Min Sci, 9(2):144–8.
13. Lundborg, N. 1981 The probability of flyrock damages. Swedish Detoni Research
Foundation, Stockholm, D.S. 5, 39 pp.
14. Lundborg, N., Persson, P.A., Ladegaard-Pedersen, A. and Holmberg, R. 1975
Keeping the Lid on Flyrock in Opencast Blasting. Engineering and Mining Journal,
pp 95-100.
15. Massey, J. B., and Siu, K. L. 2003 Investigation of flyrock incident at ClearWater
Bay Road on 6 June. Civ Eng Dept, Govt Hong Kong Special Admin Region, Hong
Kong. p. 49.
16. Monjezi, M, Bahrami, A. and Yazdani Varjani, A. 2009 Simultaneous prediction of
fragmentation and flyrock in blasting operation using artificial neural network.
International Journal of Rock Mechanics & Mining Sciences, 47, 476-480.
17. Monjezi, M., Dehghan, H. and Samimi Namin, F. 2007 Application of TOPSIS
method in controlling fly rock in blasting operations. In: Proceedings of seventh
international science conference SGEM. Sofia, Bulgaria, p. 41–9.
18. Rezaei, M., Monjezi, M. and Yazdian Varjani, A. 2010 Development of a fuzzy
model to predict flyrock in surface mining. Safety Science journal, 49, 298-305.
19. Simpson, P. K. 1990 Artificial neural system: foundation, paradigms, applications
and implementations. New York: Pergamon.
20. Singh, T.N., and Singh, V. 2005 An intelligent approach to predict and control
ground vibration in mines. Geotech Geol Eng,23: 249–62.
21. Tawadrous, A. S, and Katsabanis, P. D. 2005 Prediction of surface blast patterns in
limestone quzarries using artificial neural networks. Int J Fragmentation
Blasting,10:233–42.
22. Tawadrous, A. S., and Katsabanis, P. D. 2007 Prediction of surface crown pillar
stability using artificial neural networks. Int J Numer Anal Methods Geomech,
31(7):917–31.
23. Thornton, D., Kanchibolta, S. S, and Brunton, I. 2002 Modelling the impact and blast
design variation on blast fragmentation. Int J Fragmentation Blasting, 6(2): 169–88.
24. Tonnizam Mohamad, E., Jahed Armaghani, D., Noorani, S.A., Saad, R., and Alavi
Nezhad Khaili Abad, S.V. 2012 Prediction of Flyrock in Boulder Blasting by Using
Artificial Neural Network. Electronic Journal of Geotechnical Engineering.Vol. 17,
Bund.R, 2585-2595.
Vol.18 [2013], Bund. B 374

25. Zhu, Z., Xie, H., and Mohanty, B. 2008 Numerical investigation of blasting-induced
damage in cylindrical rocks. Int J Rock Mech Min Sci., 5:111–21.

© 2013 ejge

You might also like