You are on page 1of 12

Prediction of Ground Vibration by Statistical and Artificial

Intelligence Approaches - A Comparative Study

By Gyana Ranjan Tripathy and Prakash K. Palei, Vibration Technology Division, Central Water and Power
Research Station, Pune- 411 024, Maharashtra, India Corresponding Author’s E-mail ID: prakash.palei@gmail.com,
grtripathy@yahoo.co.in

Abstract
Ground vibration is an integral part of the rock blasting process which may cause damages to the
nearby structures located surrounding the blasting area. The peak particle velocity (PPV) associ-
ated with blast-induced ground vibration is the most suitable parameter for correlating damage
potential of blast-induced ground vibration. Thus, its prediction plays important role for smoothly
conducting the blasting operations. The PPV largely depends on quantity of charge weight per de-
lay, distance between the blast and observation points and the properties of the transmitting media.
In this research work, a comparative study has been carried out for prediction of safe charge weight
per delay for various distances by commonly used statistical equations, multiple linear regressions
as well as by ANN technique. The blast data collected from a basaltic region in an urban area has
been used for testing of the model. In the ANN model, the distance from the blasting site to the
monitoring stations, depth of hole and the maximum charge weight per delay were selected as the
input parameters of the constructed model, the output parameter was the PPV as the vibration
indicator. Results indicated that the ANN model with 10-5-1 configuration, sigmoid transfer func-
tion in hidden layer and linear transfer function in output layer gives the best correlation between
predicted and actual values, with overall correlation coefficient (R) of 0.95. It can predict the PPV
with acceptable mean -squared error (MSE) of 0.0029. The results in this study indicated that
ANN is a reliable technique to predict the PPV due to blasting.

Keywords: Ground Vibration, Intelligent technique, Empirical relation, Artificial neural network,
Vibration level

1. Introduction the acceptable level by suitably designing the blast-


Blasting is one of the most widely used methods ing pattern. Since excessive blast-induced ground
for rock excavation in mining and civil engineering vibration has potential to adversely affect the safety
projects. The previously reported results (Ebrahim and integrity impact on rock mass as well as nearby
et.al.; 2012) suggest that only 20–30% of explosive structures, accurate method for prediction of PPV
energy is utilized for breaking and moving the rock at different distances from known charge weight
while the rest causes ground vibration, air blast, per delay plays very significant role in completing
noise, fly rock, back breaks, etc. Ground vibration blasting operation safely in thickly populated ur-
is considered as the most important unwanted ef- ban areas.
fects of blasting having potential to cause damage
surrounding structures and creates annoyances to The PPV associated with blast induced ground vi-
people living nearby (Hakan et. al.; 2009). Though brations depends on many factors, which can be
during blasting ground vibrations cannot be com- broadly divided into two categories such as con-
pletely eliminated, but it can be controlled within trollable and uncontrollable parameters. Control-

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Study | 1
Blasting and Fragmentation
Vol. 11, No. 2, 2018

lable parameters can be changed while uncontrol- optimization since 1980s. Due to imprecision and
lable parameters cannot be changed. Various blast uncertainties in predicting real world problems,
design parameters like maximum charge weight ANN techniques have become increasingly useful.
per delay, charge per hole, depth of hole, distance ANN model is normally much faster and less com-
from blast site to the observation point, number of plex compared to most of the conventional simula-
holes in a blast, total delay in milliseconds, etc. are tion methodology models (Motevali et al., 2012).
considered as controllable parameters, whereas ge- It can model based on no assumptions concerning
ology of the site, topography, rock properties, etc. the nature of the phenomenological mechanisms,
are considered as uncontrollable parameters (Mo- and understand the mathematical background of
hamed, 2009). It is an established fact that ground problem (Fathi et al., 2011). This method is ca-
vibration is directly related to the maximum quan- pable of developing the relationship between the
tity of explosive used per delay and inversely to the inputs and outputs of a process. This method is
distance between the face of the blast to the moni- also computationally inexpensive and easy to use.
toring point (Khandelwal and Singh, 2009). ANN can also help to overcome the limits of the
empirical equations in predicting the ground vibra-
Peak particle velocity (PPV) has been mostly con- tion. The neural network approach could predict
sidered as the main parameter responsible for caus- the test data consistently with reasonable accuracy.
ing damage to the nearby structures. Most of the
available damage criteria are mainly based on the Recently, ANN has been extensively used to predict
PPV (mm/s) (Nateghi, 2012). Predicting the blast blast induced ground vibrations by several research-
induced ground vibration through the ground is an ers (Yong et. al., 2005, Khandelwal et. al., 2007,
uphill task due to the complex behavior of the rock. Mohammad, 2009, Monjezi et al., 2011, Deh-
In spite of these difficulties, reasonable assessments ghani and Ataee-pour, 2011, Amnieh et al., 2010,
of blast-induced ground vibration can be made 2012;). Yong et. al. (2005) studied the effect of var-
with some of the empirical, statistical and artificial ious parameters on ground vibration and predicted
intelligent techniques. Several empirical equations the blast-induced ground vibration by applying the
have been developed by various researchers (Duvall ANN technique. Khandelwal et. al. (2007)  used
and Fogleson, 1962; Duvall et al., 1963; Langefors ANN technique to predict PPV at a magnesite
and Kihlstrom, 1963; Birch and Chaffer, 1983; mine in India and compared their findings with
Davies et al., 1964; Ghosh and Daemen, 1983; commonly used predictors. They also studied blast
Ambraseys and Hendron, 1968; Bureau of Indian vibration and frequency using rock, blast design
Standard, 1973 and Siskind et al., 1980) for pre- and explosive parameters by ANN technique and
diction of PPV. All the predictors estimate the PPV compared their findings with multivariate regres-
by using the maximum charge per delay and the sion analysis (Khandelwal and Singh, 2006). Mo-
distance between blast face and monitoring point. hammad (2009)  used several ANN models and
There is no uniformity in the predicted result as observed that with the increase in number of input
they are based on different assumptions. Different data the capability of ANN to predict PPV can be
predictors give different values of PPV for same improved. Monjezi et al. (2011) developed an ANN
amount of charge per delays in the same operating model to predict PPV at Siahbisheh project in Iran,
area. Therefore in the past several years, soft com- using maximum charge per delay, distance from
puting technique, artificial neural networks (ANN) the blasting face to the monitoring point, stem-
have been used by the researchers for the prediction ming length and hole depth as input parameters
of ground vibration (Khandelwal, 2010). and compared their results with empirical models
and multivariate regression analysis. By sensitiv-
ANN is a branch of the artificial intelligence and ity analysis, they found that the distance from the
has been extensively used for the modeling and blasting face is the most effective and the stemming
2 | Gyana Ranjan Tripathy and Prakash K. Palei
Blasting and Fragmentation
Vol. 11, No. 2, 2018

is the least effective parameter on the PPV.  Deh- 2. Site Description and
ghani and Ataee-pour (2011) developed a model to Data Collection
predict PPV using dimensional analysis.  Monjezi Ground vibration data was collected for prediction
et al. (2013) proposed an ANN-based solution for of safe charge weight per delay for various distances
prediction of PPV at Shur River dam, Iran. Several during development of an integrated residential
other researchers also predicted PPV and frequency township in one of most thickly populated urban
based on ANN models (Amnieh et al., 2010, 2012; area in India, Mumbai. The township consists of
Alvarez-Vigil et al., 2012; Mohamadnejad et al., about 150 multistoried residential buildings. In or-
2012; Maryam  Amir et. al., 2016) and obtained der to provide underground utilities in this town-
better results compared to conventional methods. ship, trenches are required to be excavated in the
Hence, from the above literature survey, it can be close proximity of these buildings. The rock forma-
concluded the ANN is a very versatile and efficient tion at the site being hard basalt, blasting was used
method for the prediction of PPV. Though several for excavation of trenches. The nearest distance
studies have been conducted to predict the ground between the buildings and proposed trenches were
vibrations by ANN technique but they are mostly about 3.0 m and longest distances may be about
in the mining areas. However, there are hardly any 15.0 m. Fifteen test blasts were conducted at the
reports on the application of ANN technique for site using 33 mm diameter jack hammer holes with
prediction of ground vibration in the construc- depths varying from 1.2 m (4 ft) to 2.4 m (8 ft). 25
tion projects especially very close to the residential mm diameter and 20 cm long cartridge explosive
buildings where the ground vibration characteris- each weighing 0.125 kg was used with non-elec-
tics are significantly different from those observed trical delay detonators with in-hole delay of 450
at mining projects. This is due to the fact that the ms and surface delay of 25 ms for test blasts. The
ground vibrations observed from mining blasts are charge weights used per delay varied from 0.125 kg
characterized by small amplitude, low frequency to 0.438 kg and total charge per blast varied from
and long duration while those from construction 0.75 kg to 10.06 kg. The ground vibrations gener-
blasts have comparatively larger amplitude with ated due to the blasts were recorded near the build-
very high frequency content, shorter duration of ing as well as on exposed rock, at different distances
motion having little potential to produce structural varying from 3 m to 36.3 m. Parameters, such as
response amplifications (Tripathy et. al. 2016). hole diameter, hole depth, burden, spacing, charge
length and charge per delay were collected from the
Therefore, in the present study an attempt has blasting site. Distance from the monitoring point
been made to apply ANN technique for predicting to blasting face was also measured in the field prior
the PPV in a construction project where trenches to blast.
have been constructed for providing underground
utilities. The most widely used predictor equations
have also been used to predict the PPV and the
3. Empirical Methods for Prediction
estimated results have been compared with actual
field data. The same input–output data sets have of PPV
also been used for the prediction of PPV by ANN In order to compare the amplitude of the ground
technique. Here, the maximum charge per delay, vibration produced by blasting, scaled distance
depth of hole and distance from blast site to moni- (SD) laws were developed by several investigators
toring point has been taken as input parameters (Duvall and Fogelson, 1962; Langefors–Kihlstrom
and the measured value of PPV as output param- 1978, Ambraseys–Hendron 1968, Bureau of In-
eter. The main objective of the study was to test the dian Standard (BIS), (1973)). Table 1 shows the
suitability of the ANN model for the prediction of empirical blast-induced ground vibration predic-
PPV over the widely used predictors. tor equations proposed by various researchers. The

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Studyle | 3
Blasting and Fragmentation
Vol. 11, No. 2, 2018

Table1. Different predictors along with the values of K and β.

Figure 1. Log–log plots between PPV and scaled distance for various models.

4 | Gyana Ranjan Tripathy and Prakash K. Palei


Blasting and Fragmentation
Vol. 11, No. 2, 2018

Figure 2. Graphs between measured and predicted PPV values by various predictors.

values of site constants K and β are calculated by 4. Multiple Linear Regression


plotting the graph between PPV and SD on log– Analysis
log scale and are shown in table 1. Figure 1 shows Multiple linear regression (MLR) is an extension of
the log–log plots between PPV and different SDs. simple linear regression. MLR is a method which is
Figure 2 shows the relationship between measured mostly used to establish the relationship between a
and predicted values of PPVs by various SD laws. dependent variable and one or more independent
The higher coefficient of determination for USBM variables. In MLR, the model is fitted such that
(The United States Bureau of Mines) and Ambre- the sum of squares of differences of predicted and
seys–Henderson predictors indicates that these measured values are minimized. In the present case
equations have better prediction capability over MLR has been applied for the prediction of PPV.
Bureau of Indian Standard (BIS) and Langefors– MLR is given by the following equation (Scheaffer
Kihlstrom predictors. et al., 2011):

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Studyle | 5
Blasting and Fragmentation
Vol. 11, No. 2, 2018

5. Overview of Artificial Neural


Equation 1. Network
ANN is a mathematical model which is motivated
where Y is the predicted variable, Xi (i= 1, 2, . . ., by the biological neural networks. It is an informa-
n) are the predictors, β0 is called intercept, βi (i=1, tion-processing system which tries to mimicry the
2, . . ., P) is the coefficient on the ith predictor and human brain. Unlike classical Artificial Intelligence
e is the error associated with the predictor. The ap- (AI) systems that are aimed at directly imitating
plication of this method resulted in the following rational and logical reasoning. The information
equation: is processed in ANN by using simple intercon-
nected elements called neurons which are located
in separate layers of the network. Practically, ANN
Equation 2. provides accurate approximation for both linear as
well as non-linear functions (Hendrik et. al. 2005).
Where D is the hole depth and Q is the charge wt. A multilayer perceptron (MLP) is a feed forward
per delay and R is the distance between the ob- back propagation ANN model which maps input
servation point and blast face. The coefficient of data onto a set of apposite output data. An MLP
determination for predicted and measured values consists of at least three layers: input layer, output
of PPV is 0.0.31. Figure 3 shows the plot of mea- layer, and hidden layer(s) (Monjezi et al. 2012b)
sured and predicted PPVs by MLR. with each layer completely connected to the next
layer. The number of hidden layers and neurons in
each layer is decided by the complexity of the prob-
lem to be solved (Monjezi et al., 2011). The lines
between the nodes indicate the flow of information
from one node to the next.

The schematic diagram of the process at each


node is shown in figure 6. It is very much possible
for ANN to learn the behavior of the input data
and consequently make useful decisions despite
the presence of noise or incomplete information.
Typically, a neural network is trained to perform
Figure 3. Graph between measured and predicted PPVs by a particular function by adjusting weight values
MLR.
between the neurons as shown in figure 4. The

Figure 4. Schematic diagram of the ANN model.

6 | Gyana Ranjan Tripathy and Prakash K. Palei


Blasting and Fragmentation
Vol. 11, No. 2, 2018

neural network function is mainly determined by dicted output is compared to the actual measured
the connections between the elements. ANN can values and the error is calculated. The obtained er-
make accurate predictions by observing the trend ror is propagated back through the network to up-
of the data. Unlike MLR and regression method, date the individual weights, called backward pass.
the ANN technique has the ability to predict fu- In this process, both the training and the testing
ture trends. errors are reduced. This process is repeated until
the error is converged to a level defined by mean
The execution of the ANN model has been de- square error (MSE) or root mean squared error
scribed as follows. Prior to interpreting new infor- (RMSE) (Simpson 1990; Kosko 1994; Singh et
mation, the ANN model needs to be trained. Vari- al. 2004). This type of ANN technique has been
ous learning algorithms have been recommended successfully applied in various types of engineering
for training of neural networks, out of them Back problems (Tonnizam Mohamad et al.,2012; Jahed
Propagation Algorithm (BPA) is the most versatile Armaghani et al., 2014). However, an experimen-
and robust technique which provides the most ef- tal database including sufficient number of datasets
ficient learning procedure for MLP networks (Ar- is required to train the ANN models.
maghani et. al. 2014). BPAs are capable of solving
predictive problems which makes them very pop-
ular (Khandelwal and Singh, 2009). Rumelhart 6. Prediction of PPV by ANN
proposed back propagation algorithm (BPA) for The prediction performance of ANN method
training of the MLP networks and determination strongly depends on the architecture of the se-
of weights for the first time in 1986, (Bakhshan- lected network. Therefore, selecting the optimum
deh Amnieh et al. 2010). In this technique, the network architecture is very much essential in de-
strengths or weights of the inter neuron connec- signing ANN models. In ANN, for network con-
tions are adjusted based on the difference between struction, the number of hidden layers and the
the predicted and actual outputs. In a feed-forward number of nodes in each hidden layer should be
BP algorithm, the signals flow from input layer to determined. However, these numbers are related
the output layer, called forward pass, then the pre- to the number of input and output parameters. In

Table 2. Prediction of PPV by ANN method with one hidden layer.

Table 3. Prediction of PPV by ANN method with two hidden layers.

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Studyle | 7
Blasting and Fragmentation
Vol. 11, No. 2, 2018

the present case, depth of hole, charge weight per and 3. However, better performance was obtained
delay and distance between the monitoring point for the network with two hidden layers and one
and blast face were considered as inputs. For ANN output layer. In the hidden layers Sigmoid Tangent
modeling, all 68 datasets were divided into train- function has been used. A linear transfer function
ing, validation and testing datasets. In this regard, is used in the output layer. The Levenberg-Mar-
80% of the datasets (48 datasets) were assigned for quardt (LM) algorithm technique is more powerful
training purposes while the other 10% (10 datas- than the other techniques therefore; LM algorithm
ets) were used for testing and 10% (10 data sets) has been used in this network (Kisi & Uncuoğlu,
were used for validation of the of the network. It 2005). The layers array of this network is in the
is to be noted that, while developing the predic- form {10 5 1} showed improved performance. Fig-
tive model, the data were normalized to values be- ure 5 illustrates the correlation coefficient (R) for
tween -1 and 1. In the present study, the network the proposed ANN model for training, test, valida-
has been trained with various combinations of tion and overall data.
transfer functions as well as number of neurons in
the hidden layers. The results are listed in table 2

Figure 5. Coefficient of correlation for the training, validation, testing and overall data sets.

8 | Gyana Ranjan Tripathy and Prakash K. Palei


Blasting and Fragmentation
Vol. 11, No. 2, 2018

7. Comparison of Results of using ANN for prediction of PPV is that it does


Figure 6 illustrates the comparison between the not follow the over fitting and under fitting law of
experimental values of PPV with the predicted one curves as in the case of vibration predictors. The
by different predictors, MLR and ANN method. use of any predictor without validation may invite
Here, the PPV values predicted by ANN closely further complication for smooth conduct of blast-
match with the measured PPV values whereas the ing operations. This study also highlighted that all
other predictors fail to predict the values with that predictor equations used in the paper are either
much accuracy. Hence, if the safe charge of explo- over estimating or under estimating the safe explo-
sive is estimated using these predictors, then this sive charge to keep the PPV under the safe limit.
is going to give more trouble in controlling the Predictions by using both predictor equations and
ground vibration. It may sometime under estimate multiple linear regression analysis are not appropri-
or over estimate the quantity of explosive. ate for the site where important residential build-
ings are very close to the excavation site. It has been
found that the ANN model with 10-5-1 configu-
ration, sigmoid transfer function in hidden layer
8. Conclusions and linear transfer function in output layer gives
In the present research work, peak particle veloc- the best correlation between predicted and actual
ity has been predicted by using ANN and various values, with overall correlation coefficient (R) of
widely used predictor equations. It has been ob- 0.95. The results in this study indicated that ANN
served that prediction capability of ANN is better can be applied to predict the PPV due to blasting
than the predictor equations. The most advantage operations.

Figure 6. Comparison of measured and predicted PPVs by various models.

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Studyle | 9
Blasting and Fragmentation
Vol. 11, No. 2, 2018

9. Acknowledgements R. Nateghi, Evaluation of blast induced ground


Authors are thankful to Dr (Mrs.) V. V. Bhosekar, vibration for minimizing negative effects on sur-
Director, Central Water and Power Research Sta- rounding structures, Soil Dynamics and Earth-
tion (CWPRS), Pune for the encouragement and quake Engineering, 43 (2012):pp.133–138.
support to publish the paper. We also thank Shri A.
K. Ghosh, Scientist – D and In-charge, Vibration A. Ghosh and J. K. Daemen, A simple new
Technology Division, CWPRS, Pune for the valu- blast vibration predictor (based on wave
able suggestions and guidance during the prepara- propagation laws). In: proceedings of the
tion of the manuscript. 24th US symposium on rock mechanics,
Texas, USA, 1983; 151–161.

10. References W. I. Duvall and D. E. Fogleson, Review of criteria


E. Ghasemi, M. Ataei and H. Hashemolhossei- for estimating damage to residences from blasting
ni, Development of a fuzzy model for predicting vibration, USBM-I, 1962; 5968.
ground vibration caused by rock blasting in surface
mining, J. Vib. & Contl. 2012;:19(5) 755–770. N. R. Ambraseys and A. J. Hendron, Dynamic be-
haviour of rock masses, Rock mech. Engg. Prac-
Hakan Ak, M. Iphar, M. Yavuz, A. Konuk, Evalu- tices, London: Wiley; 1968; 203–207.
ation of ground vibration effect of blasting opera-
tions in a magnesite mine, Soil Dyn. Earthq. Engg., U. Langefors and B. Kihlstrom, The Modern Tech-
2009;29: 669–676. nique of Rock Blasting, New York: Wiley, 1963.
Bureau of Indian Standard (BIS), Criteria for safety
G.R.Tripathy, R.R.Shirke and M.D.Kudale, Safety and design of structures subjected to underground
of engineered structures against blast vibrations: A blast, ISI Bull; IS-6922; 1973.
case study, J. Rock Mech. Geotech. Engg., 2016;
8: 248-255. C.H. Dowding, Blast Vibration Moni-
toring and Control, Prentice-Hall,
G.R. Tripathy and I. D. Gupta, Prediction of Englewood Cliffs, 1985; 297.
Ground Vibrations due to Construction Blasts in
Different Types of Rock, Rock Mech Rock Engg., M. Khandelwal and T. N. Singh, Evaluation
2002; 35: 195. of blast-induced ground vibration predictors,
Soil Mechanics and Earthquake Engineering,
M.T. Mohammad, Artificial neural network for 2007; 27(2):116–25.
prediction and control of blasting vibration in As-
siut (Egypt) limestone quarry, Int. J. Rock Mech. M. Monjezi, M. Hasanipanah and M. Khandel-
Min. Sci., 2009; 46(2):426–31 wal, Evaluation and prediction of blast-induced
M. Khandelwal and T. N. Singh, Prediction of ground vibration at Shur River Dam, Iran, by ar-
blast induced ground vibrations and frequency tificial neural network, Neural. Comput. Appl.,
in opencast mine: a neural network approach, J. 2013; 22:1637-1643.
Sound Vib., 2006; 289(4/5):711–25.
M. Ali, A. Ahmad, N. H. Gholam, M. Saeid and
M. Khandelwal and T. N. Singh, Prediction of G. Barat, Drying of jujube (zizyphus jujubemill)
blast-induced ground vibration using artificial fruit: comparison of prediction from mathematical
neural network, Int. J. Rock Mech. Min. Sci., models and artificial neural networks, AJCS 2012;
2009;46(7):1214–1222. 6(2):210-218.

10 | Gyana Ranjan Tripathy and Prakash K. Palei


Blasting and Fragmentation
Vol. 11, No. 2, 2018

M. Fathi, M. Mohebbi, and S. Razavi,. Applica- A.T. Edwards and T.D. Northwood Experimental
tion of image analysis and artificial neural network studies of the effects of blasting on structures, The
to predict mass transfer kinetics and color changes Engineer, 1960; 210, 538-546.
of osmotically dehydrated kiwifruit, Food and Bio-
process Technology, 2011; 4(8): 1357-1366. J.M. Esteves, Control of vibrations caused by blast-
ing Memoria 409 Laboratorio de Engenharia Civil,
Y. Lu , Underground blast induced ground shock Ministerio de Habitacao e Obras Publicas, Lisbon,
and its modelling using artificial neural network, Portugal (1978).
Computers and Geotechnics, 2005; 32( 3), 164-
178. U. Langefors, B. KihlstromThe modern technique
of rock blasting, John Wiley and Sons Inc., 1978;
M. Monjezi, , M.G. kalajahi, A. Bahrami,. Predic- 405.
tion of blast induced ground vibration using arti-
ficial neural networks, Tunnel. Under gr. Sp. Tech- D.E. Siskind, M.S. Stagg, J.W. Koop and C.H.
nol., 2011;26:46–50. Dowding, Structure response and damage pro-
duced by ground vibration from surface mine
H. Dehghani, M. Ataee-pour, Development of a blasting, US Bureau of Mines, 1980; 74.
model to predict peak particle velocity in a blasting
operation, Int. J. Rock Mech. Min. Sci., 2011; 48, H. L. Tolman, V. M. rasnopolsky,, D. V. Chalikov,
51-58. Neural network approximations for nonlinear in-
teractions in wind wave spectra: direct mapping for
H. B. Amnieh1, A. Mohammadi and M. Mozdian- wind seas in deep water, Ocean Modelling, 2005;
fard, Predicting peak particle velocity by artificial 8, 253–278.
neural networks and multivariate regression analy-
sis - Sarcheshmeh copper mine, Kerman, Iran, J. M. T. Hagan & M. Menhaj, Training Feed forward
Min. Environ., 2013; 4(2), 125-132. Networks with the Marquardt Algorithm. IEEE
Transaction on Neural Networks, IEEE Trans Neu-
A.E. Alvarez-Vigil, C. Gonzales-Nicieza, F. Lopez ral Networks, 1994; 5(6), 989-993.
Gayarre, M.I. Alvarez-Fernandez, Predicting blast-
ing propagation velocity and vibration frequency Ö. Kişi & E. Uncuoğlu, Comparison of three
using artificial neural network, Int. J. Rock Mech. back-propagation training algorithms for two case
Min. Sci., 2012; 55, 108-116. studies, Indian Journal of Engineering & Materials
Sciences, 2005; 12, 434-442.

D. J. Armaghani, M. Hajihassani, E. T. Mohamad,


A. Marto, S. A. Noorani, Blasting-induced flyrock
and ground vibration prediction through an expert
artificial neural network based on particle swarm
optimization, Arab J Geosci, 2014; 7, 5383–5396

M. Mohamadnejad, R.Gholami, and M. Ataei, ().


Comparison of intelligence science techniques and
empirical methods for prediction of blasting vibra-
tions. Tunnelling Under gr. Space Technol., 2012;
28, 238-244.

Prediction of Ground Vibration by Statistical and


Artificial Intelligence Approaches - A Comparative Studyle | 11
This page left intentionally blank.

You might also like