Accepted Manuscript

Feasibility of indirect determination of blast induced ground vibration based on

support vector machine
Mahdi Hasanipanah, Masoud Monjezi, Azam Shahnazar, Danial Jahed
Armaghani, Alireza Farazmand
PII:
DOI:
Reference:

S0263-2241(15)00354-1
http://dx.doi.org/10.1016/j.measurement.2015.07.019
MEASUR 3472

To appear in:

Measurement

Received Date:
Revised Date:
Accepted Date:

24 March 2015
19 June 2015
13 July 2015

Please cite this article as: M. Hasanipanah, M. Monjezi, A. Shahnazar, D.J. Armaghani, A. Farazmand, Feasibility
of indirect determination of blast induced ground vibration based on support vector machine, Measurement (2015),
doi: http://dx.doi.org/10.1016/j.measurement.2015.07.019

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Feasibility of indirect determination of blast induced ground vibration based on support

vector machine
Mahdi Hasanipanaha, Masoud Monjezib*, Azam Shahnazar c, Danial Jahed Armaghanid, Alireza Farazmande

Young Researchers and Elite Club, Qom Branch, Islamic Azad University, Qom, Iran. E-mail:

mahdi_hasanipanah@yahoo.com.
b*

Department of Mining, Faculty of Engineering, Tarbiat Modares University, Tehran 14115-143, Iran.

E-mail: monjezi@modares.ac.ir. Tel.: +98 9122970377; Fax: +98 2182883381 (Corresponding Author).
c

Department of Computer engineering, Faculty of Engineering, Islamic Azad University, South Tehran

Branch, Tehran, Iran. E-mail: a_shahnazar211@yahoo.com.

d

Young Researchers and Elite Club, Qaemshahr Branch, Islamic Azad University, Qaemshahr, Iran. E-

mail: danialarmaghani@yahoo.com.
e

Head of geophysics and earthquake group, Moshanir power engineering consultants, Tehran, Iran. E-

mail: afarazmand@yahoo.com.

Abstract
Mines, quarries, and construction sites face blasting environmental problems due to high level of ground
vibrations. This phenomena can cause injury to both human and damage to structures in the blasting

environment. To estimate ground vibration, several empirical predictors have been established by various
researchers, while these predictors are not commonly enforceable beyond the particular conditions.
However, ground vibration prediction is a complicated issue in consequence of the fact that a large
number of influential factors are involved. In this study, a support vector machine (SVM) was applied and
developed to predict ground vibration in blasting operations of Bakhtiari Dam, Iran. To achieve this aim,
80 blasting works were investigated and results of peak particle velocity (PPV) as a vibration index,
distance from the blast-face and maximum charge per delay were measured and monitored to utilize in the
modelling. To demonstrate applicability of the SVM model for prediction of PPV, several empirical
equations were also employed and the relevant site constants were proposed. In the analyses procedure of
this study, 60 datasets were used for model development and remaining 20 datasets were applied to check
the performance capacity of the developed model. After comparing the results obtained from SVM and
empirical equations, it was found that the SVM method provides higher performance capacity in
predicting PPV compared to empirical equations.
Keywords: Blasting, Peak particle velocity, Empirical equation, Support vector machine.

1. Introduction
In blasting operations, only 20 to 30 percent of the produced energy is used for fragmentation purposes
and the rest of this energy is wasted to create unwanted blasting environmental issues like back-break, airoverpressure, yrock and ground vibration [1-6]. The wasted energy exposure creates problems for the
workers associated in the excavation process as well as the local in habitants in the nearby area [7].
Among environmental impacts of blasting, ground vibration is considered as one of the important blasting
environmental issues [8]. Although blasting vibration is short term transient phenomena, the residents in
the vicinity of operations feel that if vibration continues then its swelling may get damaged [9]. High
ground vibration resulting from blasting has undesirable effects on the structural integrity, groundwater,

and ecology of the nearby area [10]. Therefore, prediction of ground vibration is a significant criterion for
future blasting operations to minimize the blasting environmental problems.
Normally, ground vibration is recorded in terms of two different parameters namely peak particle velocity
(PPV) and frequency. Among them, PPV is considered as a vibration index, which is an important
indicator for controlling the structural damage criteria. Many vibration predictors have been develped
empirically to estimate PPV induced by blasting. Nevertheless, these methods considered only limited
numbers of effective parameters on PPV whereas, this phenomenon is also affected by other controllable
or non-controllable parameters [10, 11]. As a result, empirical methods are not accurate enough in many
cases, even though PPV prediction with high degree of accuracy is necessary to estimate the blast safety
area [12]. Apart from that, simple and multiple regression techniques in estimating PPV have drawn
attention [2, 13, 14]. Nevertheless, implementing the statistical predictive methods is not reliable if new
available data are different from the original ones as the form of the obtained equation needs to be
updated. Aside from that, feasibility of soft computing techniques like artificial neural network (ANN),
fuzzy interface system (FIS) and neuro-fuzzy system (ANFIS) in solving geotechnical engineering
problems [15-20] and more specific, for prediction of PPV resulting from blasting [10, 21-23] has been
reported in many studies. In the present study, support vector machine (SVM) is used to predict ground
vibration resulting from blasting operations in Bakhtiari Dam, Iran. For the sake of comparison, empirical
predictors are also employed to predict PPV.

2. Ground Vibration and Its Effective Parameters

When an explosive is detonated in a blast-hole, the explosive chemical reaction creates some gases with
high pressure. These gases pressure crush the rock adjacent to the blast-hole. A wave motion is created in
the ground by the strain waves conveyed to the surrounding rocks [24]. Due to various breakage
mechanism like, crushing and radial cracking the strain energy carried out by these strain waves

fragments the rock mass. During the propagation of the stress wave, high pressure gases extend
discontinuities such as fracture and joint [25]. These waves are identified as ground vibration.
Two main groups of parameters that affect ground vibration produced by blasting are included
controllable and uncontrollable parameters. The rst group consists of controllable or blast design
parameters like burden, spacing, sub-drilling, blast-hole depth, blast-hole diameter, number of blast hole,
bench height, stemming height and type and weight of the explosive. Some of the controllable parameters
of blasting are displayed in Fig. 1. Rock mass properties are considered as uncontrollable blasting
parameters [26, 27]. It is essential to optimize blasting design parameters to decrease ground vibration
based on the properties of rock mass which include rock strength, density, wave velocity, discontinuity
conditions [3, 8, 10]. In Indian Standard Institute [28], as a vibration index, PPV is introduced to control
the structure damage.
Several emperical equations were developed by some researchers or institutions to predict the PPV
induced by blasting. There are only two factors namely charge weight and distance from the blast-face in
these equations [28-30]. Table 1 shows some empirical predictors for prediction of PPV. In this table, W
is the maximum charge weight in kg, D is the distance from the blasting face (m), and K, a, b and n are
the site constants.
In addition to empirical predictors, artificial intelligent techniques have been extensively-used by several
researchers to predict PPV. Iphar et al. [21] utilized two different methods including simple regression
and ANFIS models to predict PPV. They used 44 PPV values obtained from blasting operations in
Turkey. The results indicated that the proposed model yields better results compared to regression
analysis. Khandelwal and Singh [33] used ANN and multivariate regression analysis (MVRA) techniques
to predict PPV and frequency. Finally, ANN results show closer agreement with the measured datasets in
comparison to MVRA prediction. An ANN model with four input parameters including distance from
blast-face, charge per delay, hole depth and stemming length was developed by Monjezi et al. [34] to
predict PPV. For this purpose, a number of 182 datasets was measured around the Kandovan tunnel in
4

Iran. They demonstrated that ANNs are enforceable tools for prediction of PPV. ANN technique and
empirical equations were utilized by Monjezi et al. [3] for prediction of PPV. They compared the obtained
results from ANN model with the actual eld data obtained from Shur River Dam in Iran. Finally, they
concluded that the ANN is more accurate technique in predicting PPV in comparison with empirical
equations. Fisne et al. [35] employed FIS and regression model in order to estimate PPV considering
some blasting data obtained from Akdaglar quarry in Turkey. Similar to many studies, they used the
weight of explosive material and distance from blast-face as model inputs to predict PPV. They found that
FIS model can provide higher performance capacity in predicting PPV in comparison to statistical model.
Hajihassani et al. [23] proposed a new hybrid of artificial intelligence model namely imperialist
competitive algorithm (ICA)-ANN for prediction of PPV values obtained from Harapan Ramai quarry,
Malaysia. For comparison purpose, a pre-developed ANN model was also applied to predict PPV. It was
found that ICA-ANN predictive model can predict PPV values with higher level of accuracy in
comparison to pre-developed ANN approach. Verma and Singh [36] utilized three models including
MVRA, ANN and SVM to predict PPV induced by blasting. They utilized results of 137 blasting
operations obtained from Chhattisgarh site, India. They successfully showed that SVM model can be
performed to optimize PPV with greater degree of confidence due to its robustness compared to other
predictive models. In the other study of SVM, Mohamadnejad et al. [37] investigated the results of
ground-vibration measurements carried out in Masjed-Soleiman dam in Iran. They used two intelligence
techniques namely ANN and SVM and concluded that the SVM is a more precise and faster technique
than the ANN model. Table 2 shows several recently-investigations with their performances in predicting
PPV using soft computer techniques.

3. Support Vector Machine

Support vector machine (SVM) is defined as a technique that uses statistics theory. It is a machine
learning theory frame and general way, established on a set of finite samples [44]. To simplify the SVM

algorithm, a margin analogue is built in the space of the target values by using Vapniks e -insensitive
loss function as follows [45, 46]:

(1)
where

is Vapniks insensitive loss function,

and

are the targets and outputs, respectively. A linear

regression usually expressed by:

(2)
where w represents the weighting matrix, X is a vector and b is the bias term. If two margins are
considered for the regression line, then they have to minimize as below:
(3)

where C is the trade-off parameter to minimize the associated error and margin. A constrained
optimization problem is shown as below:

(4)
(5)

Subject to

(6)
(7)
where

and

are the distances between data points and regression margins which is called regression

error. The KarushKuhnTucker (KKT) conditions can be presented by the following equations:
(8)

(9)
(10)
(11)
where

and

are Lagrange multipliers.

Support vectors are defined by:

(12)
The value of b can be calculated as:
(13)
Finally, the output can be obtained from the following equation:
(14)

4. Case study
The Bakhtiari Dam is placed in the southwest of Iran. The dam lies geographically in latitude 32 57' 41"
N and longitude 48 46' 34" E, and is placed at 50 km of the Andimeshk city (see Fig. 2). The type of bed
rock is limestone and Mary limestone in this area. The features of Bakhtiari Dam are shown in Table 3.
Bakhtiari dam is constructed for several reasons namely hydroelectric energy generation, control of
floods, preventing annual destruction and generation of 3000 GW/h per annum. In order to provide an
access road between Bakhtiari dam and the nearest city, several tunnels were excavated using drilling and
blasting method. The railway tunnels are located on the side of access road, as shown in Fig. 3. Therefore,
ground vibration estimation as one of the environmental impacts of blasting, is of interest in the
mentioned site. To predict blast-induced ground vibrations in vulnerable locations in this site, 80 blast
events were monitored and the values of PPV in terms of mm/s, maximum charge per delay in terms of kg

and distance between blast-face and monitoring station in terms of m were measured carefully. To
measure the values of PPV, the MR2002-CE SYSCOM seismograph was used (see Fig. 4). The feathers
of the used seismograph are tabulated in Table 4. Not that, in the conducted blasting operations,
ammonium nitrate fuel oil (ANFO) was used as the primary explosive substance.
The distance between blast-face and monitoring station was measured using GPS (global positioning
system). Also, the value of maximum charge per delay was measured by controlling the blast hole charge
and based on blasting design. As suggested by Duvall and Petkof [24], distance from the blast-face and
maximum charge per delay are the most influential factors on PPV. Additionally, these parameters have
been extensively-used as predictors by several scholars as it can be seen in Table 2. Among all 80
collected blasting dada, results of 60 blast events were used to develop SVM technique and to determine
site constants of empirical equations, whereas remaining 20 datasets were applied to test the system and
also compare to empirical equations. Table 5 shows the range of input and output parameters used in this
study. An example of PPV result recorded by seismograph is shown in Fig. 5.

5. Prediction of PPV Using Empirical Equations

Previously, several researchers developed site constants for predicting PPV (see Table 1). This section
presents determination procedure of site constants of PPV equations presented in Table 1. To do this, a
series of statistical analyses should be performed. By using scaled distance (SD) laws and also plotting
the graphs between PPV and SD values, site constants including K and n were determined. This process
was implemented using statistical software package, Microsoft Excel 2013. In fact, this software is able to
determine site constants considering the power regression analysis. As an example, the plotted actual
PPVs against SDs for USBM equation where SD =

, is shown in Fig. 6. As indicated in this figure,

values of 68.35 and 0.9 are obtained for k and n of USBM equation, respectively.

These site constants were achieved for model development (training) datasets as tabulated in Table 6.
Subsequently, using testing datasets (20 datasets), performance capacities of these site constants were
checked. Figs. 7 to 10 illustrate coefficient of determination (R2) of measured PPVs against PPVs
predicted by empirical equations for testing datasets using obtained site constants in Table 6. These
values were obtained as 0.557, 0.495, 0.001 and 0.684 for USBM [24], Davies et al. [32], Indian Standard
[28] and Ambraseys-Hendron [31], respectively. Generally, these values indicate low performance
capacity of the empirical equations in predicting PPV. More details regarding evaluation of the applied
empirical equations are given later.

6. Prediction of PPV Using SVM

The performance of SVM for regression depends greatly on the combination of several parameters. They
are capacity parameter C,

of

-insensitive loss function. C is a regularization parameter controlling the

trade-off between maximizing the margin and minimizing the training error. Selection of this parameter
depends on amount of avoiding misclassifying in each training set. Generally, larger values of C can be
utilized when correct classification of the training points and higher SVM performance are required.
Whereas, smaller values of C may cause a weak performance of SVM in solving engineering problems.
The optimal value for

which is usually unknown, depends on the noise present kind in the data. There is

the practical consideration of the support vectors numbers, if enough noise knowledge is available to
choose an optimal value for e. Selection of the appropriate value of

is critical based on literature. In this

paper, using trial-and-error method, the values of 1 and 0.1 were utilized for C and

respectively. It is

worth mentioning that in SVM, training datasets are used to learn the system and cannot access after
training. Therefore, only results of testing datasets are available after training. Fig. 11 shows the measured
and predicted PPV values for testing datasets. Obtained coefficient of determination of 0.957 for Fig. 11
shows that SVM can predict PPV induced by blasting with high degree of accuracy. More details for
evaluation of SVM results can be seen in the next section.
9

7. Analysis of the Results

To evaluate the accuracy and applicability of the SVM model as well as empirical equations, only results
of testing datasets (20 datasets) were considered and these results were compared to the measured PPVs.
Some performance indices i.e. variance account for (VAF), variance absolute relative error (VARE), root
mean square error (RMSE), median absolute error (MEDAE), mean absolute percentage error (MAPE),
nash and sutcliffe (NS) and R2 were applied to control the performance capacity of the used models.
These performance indices can be formulated as follows:

(15)

(16)

(17)

(18)

(19)

(20)
where xi is the measured value, xp is the predicted value, var is the sign for the variance and n is the
number of data sets. IF VAF value is 100 and values of VARE, RMSE, MAPE and MEDAE are zero, the
model will be excellent. Also, the optimal values for NS and R2 are 1. Results of these performance
indices for all mentioned methods using only testing datasets are tabulated in Table 7. According to this
table, the lowest values of VARE, RMSE, MAPE and MEDAE and also the highest value of VAF, NS
and R2 are obtained for SVM technique in comparison with the obtained performance indices results from

10

empirical equations. For instance, RMSE equal to 0.34 for SVM model indicates the robustness of this
technique for prediction of PPV, whereas they were achieved as 1.16, 1.15, 1.19 and 1.74 for USBM,
Ambraseys-Hendron, Davies et al. [32] and Indian Standard models, respectively. Based on above
discussion, SVM can predict PPV with higher degree of accuracy compared to empirical equations. Fig.
12 displays a comparison between measured and predicted PPVs by empirical equations and SVM model
for testing datasets. According to this figure, the proposed SVM model provides more accurate results in
comparison with empirical equations and it can be applied to predict PPV values in cases where high
degree of accuracy is required.

8. Sensitivity analysis
Sensitivity analysis is a technique to evaluate the most effective input parameter on output parameter. To
apply this, the cosine amplitude method can be utilized [47]. This method is illustrated in the following
equation:

(21)

where xi and xj represent input and output parameters, respectively and n is the number of all datasets.
The strengths of the relations between input and output parameters are shown in Fig. 13. Based on this
figure, maximum charge per delay is the most influential parameter on PPV.

9. Conclusion
In this study, a SVM technique was applied and proposed to estimate PPV resulting from blasting
operations in Bakhtiari dam, Iran. In this regard, 80 blasting works were investigated in the mentioned
site and relevant parameters were measured. For developing SVM model, results of distance from the

11

blast-face and maximum charge per delay were set as input parameters as recommended in literatures. In
the modelling procedure of SVM, 60 datasets were used for training the system and remaining 20 datasets
were applied to test the developed model. In addition, empirical equations were employed to predict PPV
using the same datasets of SVM section. For this purpose, site constants of these equations were
determined using statistical analysis. Eventually, in order to examine the applicability and accuracy of
the proposed model, an extensive comparison was made between measured and predicted PPVs obtained
from SVM and empirical equations using testing datasets. To do this, many performance indices such as
RMSE, VAF and VARE were selected and utilized. Based on obtained results of performance indices, it
was found that the SVM model provide higher performance capacity in predicting PPV in comparison
with used empirical equations. R2 and RMSE values of 0.957 and 0.34 for testing datasets of SVM model
reveal that this model can introduce as a practical technique for estimating PPV with high degree of
accuracy. SVM as an enforceable and accurate tool, can utilize in estimating PPV resulting from blasting.
Through the sensitivity analyses, it was found that the charge weight per delay is the most effective
parameters on PPV.

References
[1] A. Aghajani-Bazzazi, M. Osanloo, Y. Azimi, Flyrock prediction by multiple regression analysis in
Esfordi phosphate mine of Iran, Sanchidrian, J. A(Ed.) Rock Fragmentation by Blasting (pp.649-657).
London: Taylor and Francis Group, 2007.
[2] M. Khandelwal, M. Monjezi, M, Prediction of flyrock in open pit blasting operation using machine
learning method, Int J Min Sci Tech 23(3) (2013) 313-316.

12

[3] M. Monjezi, M. Hasanipanah, M. Khandewal, Evaluation and prediction of blast-induced ground

vibration at Shur River Dam, Iran, by articial neural network, Neural Comput Applic. 22 (2013) 16371643.
[4] D. Jahed Armaghani, M. Hasanipanah, E. Tonnizam Mohamad, A combination of the ICA ANN
model to predict air overpressure resulting from blasting, Eng Comput. 2015. Doi: 10.1007/s00366-0150408-z.
[5] R. Shirani Faradonbeh, M. Monjezi, D. Jahed Armaghani, Genetic programing and non-linear
multiple regression techniques to predict backbreak in blasting operation. Eng Comput. 2015.
doi:10.1007/s00366-015-0404-3.
[6] D. Jahed Armaghani, E. Momeni, SV. Alavi Nezhad Khalil Abad, M. Khandelwal, Feasibility of
ANFIS model for prediction of ground vibrations resulting from quarry blasting. Environ Earth Sci. 2015.
DOI 10.1007/s12665-015-4305-y.
[7] S. Bhandari, Engineering Rock Blasting Operations, Netherlands, A.A. Balkema publishers, 1997.
[8] M. Monjezi, M. Ghafurikalajahi, A. Bahrami, Prediction of blast-induced ground vibration using
artificial neural networks, Tunnel Undergr Sp Tech 26(1) (2011) 46-50.
[9] B.P.E. Hemphill Gary, Blasting Operations, McGraw-Hill Book Company, 1981.
[10] M. Khandelwal, T.N. Singh, Evaluation of blast-induced ground vibration predictors, Soil Dyn
Earthq Eng. 27 (2007) 116-125.
[11] T.N. Singh, V. Singh, An intelligent approach to prediction and control ground vibration in mines,
Geotech Geolog Eng 23 (2005) 249-262.
[12] E. Ghasemi, M. Ataei, H. Hashemolhosseini, Development of a fuzzy model for predicting ground
vibration caused by rock blasting in surface mining, J Vib Control 19 (2013) 755-770.

13

[13] A.K. Verma, T.N. Singh, Intelligent systems for ground vibration measurement: a comparative
study, Eng Comput 27 (2011) 225-233.
[14] T. Hudaverdi, Application of multivariate analysis for prediction of blast-induced ground vibrations,
Soil Dyn Earthq Eng 43 (2012) 300-308.
[15] G.M. Alvarez, R. Babuska, Fuzzy model for the prediction of unconned compressive strength of
rock samples, Int J Rock Mech Min Sci 36 (1999) 339-349.
[16] R. Singh, A. Kainthola, T.N. Singh, Estimation of elastic constant of rocks using an ANFIS
approach, Appl Soft Comput 12 (2012) 40-45.
[17] E. Momeni, R. Nazir, D. Jahed Armaghani, H. Maizir, Prediction of Pile Bearing Capacity Using a
Hybrid Genetic Algorithm-Based ANN, Measurement 57 (2014) 122131.
[18] D. Jahed Armaghani, M. Hajihassani, B. Yazdani Bejarbaneh, A. Marto, E. Tonnizam Mohamad,
Indirect Measure of Shale Shear Strength Parameters by Means of Rock Index Tests through an
Optimized Artificial Neural Network. Measurement 55 (2014) 487-498.
[19] E. Tonnizam Mohamad, D. Jahed Armaghani, E. Momeni, S.V. Alavi Nezhad Khalil Abad,
Prediction of the unconfined compressive strength of soft rocks: a PSO-based ANN approach, Bull Eng
Geol Environ (2014) doi:10.1007/s10064-014-0638-0.
[20] D. Jahed Armaghani, M Hajihassani, M Monjezi, ET Mohamad, A Marto, MR Moghaddam,
Application of two intelligent systems in predicting environmental impacts of quarry blasting, Arab J
Geosci. 2015. doi:10.1007/s12517-015-1908-2.
[21] M. Iphar, M. Yavuz, H. Ak, Prediction of ground vibrations resulting from the blasting operations in
an open-pit mine by adaptive neuro-fuzzy inference system. Environ Geol 56 (2008) 97-107.

14

[22] S. Ghoraba, M. Monjezi, N. Talebi, MR. Moghadam, D. Jahed Armaghani, Prediction of ground
vibration caused by blasting operations through a neural network approach: a case study of Gol-E-Gohar
Iron Mine, Iran, J Zhejiang Univ Sci A. 2015. doi:10.1631/jzus.A1400252
[23] M. Hajihassani, D. Jahed Armaghani, A. Marto, E. Mohamad, E. T. Ground vibration prediction in
quarry blasting through an artificial neural network optimized by imperialist competitive algorithm, Bull
Eng Geol Environ DOI 10.1007/s10064-014-0657-x
[24] W.I. Duvall, B. Petkof, Spherical propagation of explosion of generated strain pulses in rocks.
USBM, (1959) RI-5483.
[25] M. Khandelwal, Evaluation and prediction of blast induced ground vibration using support vector
machine. Int J Rock Mech Min Sci. 47 (2010) 509-516.
[26] M. Khandelwal, T.N. Singh, Prediction of blast induced ground vibrations and frequency in opencast
mine-a neural network approach. J Sound Vib. 289 (2006) 711-725.
[27] J.F. Wiss, P.W. Linehan, Control of vibration and air noise from surface coal mines, US Bureau of
Mines Report OFR (1978) p.103.
[28] Bureau of Indian Standard, Criteria for safety and design of structures subjected to underground
blast, ISI Bull IS-6922, 1973.
[29] W.I. Duvall, D.E. Fogelson, Review of criteria for estimating damages to residences from blasting
vibrations. R. I. 5968, US, Bureau of Mines, 1962.
[30] U. Langefors, B. Kihlstrom, The modern technique of rock blasting. Wiley, New York. 1963.
[31] N.R. Ambraseys, A.J. Hendron, Dynamic behaviour of rock masses. Rock mechanics in engineering.
Practices, Wiley, London, 1968, pp. 203207.

15

[32] B. Davies, I.W. Farmer, P.B. Attewell, Ground vibrations from shallow sub-surface blasts, vol 217.
The Engineer, London, 1964, pp. 553559.
[33] M. Khandelwal, T.N. Singh, Prediction of blast-induced ground vibration using artificial neural
network, Int J Rock Mech Min Sci 46 (2009) 1214-1222.
[34] M. Monjezi, M. Ahmadi, M. Sheikhan, A. Bahrami, A.R. Salimi, Predicting blast-induced ground
vibration using various types of neural networks. Soil Dyn Earthq Eng. 30 (2010) 12331236.
[35] A. Fisne, C. Kuzu, T. Hdaverdi, Prediction of environmental impacts of quarry blasting operation
using fuzzy logic, Environ Monit Assess 174 (2011) 461-470.
[36] A.K. Verma, T.N. Singh, Comparative study of cognitive systems for ground vibration
measurements, Neural Comput Applic 22 (2013) 341-350.
[37] M. Mohamadnejad, R. Gholami, M. Ataei, Comparison of intelligence science techniques and
empirical methods for prediction of blasting vibrations, Tunnel Undergr Sp Technol 28 (2012) 238244.
[38] M. Khandelwal, D.L. Kumar, M. Yellishetty, Application of soft computing to predict blast-induced
ground vibration, Eng Comput 27(2) (2011) 117-125.
[39] M.T. Mohamed, Performance of fuzzy logic and artificial neural network in prediction of ground and
air vibrations. Int J Rock Mech Min Sci 48 (2011) 845-851.
[40] E. Ghasemi, M. Ataei, H. Hashemolhosseini, Development of a fuzzy model for predicting ground
vibration caused by rock blasting in surface mining, J Vib Control 19 (2013) 755-770.
[41] M. Khandelwal, P.K. Kankar, S.P. Harsha, Evaluation and prediction of blast induced ground
vibration using support vector machine, Min Sci Technol (China) 20 (2010) 64-70.
[42] M. Khandelwal, Blast-induced ground vibration prediction using support vector machine, Eng
Comput 27 (2011) 193-200.

16

[43] D. Jahed 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 (2013) DOI 101007/s12517-013-1174-0.
[44] V.N. Vapnik, Statistical learning theory. John Wiley and Sons Inc. 1998.
[45] D.V. Sanchez, Advanced support vector machines and kernel methods, Neuro computing. 55 (2003)
520.
[46] Y. Yang, D. Yu, J. Cheng, A fault diagnosis approach for roller bearing based on IMF envelope
spectrum and SVM, measurement 40 (9-10) (2007) 943-950.
[47] Y. Yang, O. Zang, A hierarchical analysis for rock engineering using artificial neural networks,
Rock. Mech. Rock. Eng. 30 (1997) 207222.

List of Tables:
Table 1 Empirical PPV predictors
Table 2 Some studies of PPV prediction using soft computing methods
Table 3 Some features of Bakhtiari Dam
Table 4 Features of MR2002 seismograph
17

Table 5 Input and output parameters and their range

Table 6 Calculated values of site constants
Table 7 Performance results of different approaches used in this study

List of Figures:
Fig. 1 Several controllable parameters of blasting
Fig. 2 Location of Bakhtiari Dam
Fig. 3 Locations of access road and railway tunnel
Fig. 4 A view of seismograph
18

Fig. 5 Particle velocity from a blast event, measured by MR2002

Fig. 6 Plotted actual PPVs against SD values for USBM equation
Fig. 7 Correlation of predicted PPVs by USBM [24] against measured PPVs
Fig. 8 Correlation of predicted PPVs by Davies et al. [32] against measured PPVs
Fig. 9 Correlation of predicted PPVs by Indian Standard [28] against measured PPVs
Fig. 10 Correlation of predicted PPVs by Ambraseys-Hendron [31] against measured PPVs
Fig. 11 Correlation of predicted PPVs by SVM against measured PPVs
Fig. 12 Comparison of measured and predicted PPV values by utilized techniques
Fig. 13 Strengths of relation (rij) between input and output parameters

19

Table 1. Empirical PPV predictors

Empirical models

Equation

USBM [24]

Ambraseys-Hendron [31]

Davies et al. [32]

Indian Standard [28]

20

Table 2 Some studies of PPV prediction using soft computing methods

Technique

Input Parameter

No. of Dataset

R2

ANFIS

DI, C

44

R2 = 0.98

Monjezi et al. [34]

ANN

DI, C, BS, UCS, DR, N

269

R2 = 0.95

Monjezi et al. [8]

ANN

HD, ST, DI, C

182

R2 = 0.95

Khandelwal et al. [38]

ANN

DI, C

130

R2 = 0.92

Mohamed [39]

ANN, FIS

DI, C

162

Fisne et al. [35]

FIS

DI, C

33

R2 = 0.92

Ghasemi et al. [40]

FIS

B, S, ST, N, C, DI

120

R2 = 0.95

Monjezi et al. [3]

ANN

C, DI, TC

20

R2 = 0.93

Khandelwal et al. [41]

SVM

DI, C

170

R2 = 0.95

Khandelwal [42]

SVM

DI, C

150

R2 = 0.96

Jahed Armaghani et al. [43]

ANN-PSO

HD, S, B, ST, PF, C, D, N, RD, SD

44

R2 = 0.94

Hajihassani et al. [23]

ICA-ANN

BS, ST, C, DI, Vp, E

95

R2 = 0.98

SVM

DI, C

80

R2 = 0.96

Reference
Iphar et al. [21]

Present study

R2ANN = 0.94
R2 FIS = 0.90

Distance from the blasting face (DI); spacing (S); burden (B); stemming (ST); powder factor (PF); charge per delay (C); hole diameter (D); hole
depth (HD); rock density (RD); burden to spacing (BS); number of row (N); particle swarm optimization (PSO); sub-drilling (SD); total charge
(TC); unconfined compressive strength (UCS); delay per rows (DR); support vector machine (SVM); Young modulus (E); P-wave velocity (Vp);
imperialist competitive algorithm (ICA); coefficient of determination (R2).

21

Table 3. Some features of Bakhtiari Dam

Dam specification

Power plant specification

Dam Type

Volume of
reservoir

Width

Crest
length

Height

Type

capacity

Double curvature
arch concrete dam

4.8 billion m3

50 m

434 m

315 m

Underground

1500 MW

22

Table 4. Features of MR2002 seismograph

Dynamic Range

96 dB, up to 130 dB

Frequency Range

1 to 315 Hz

Vibration Velocity

0.001 to 115mm/s

Vibration Acceleration

0.0001 to 10 g

Number of channels

3 (X,Y,Z) data channels

23

Table 5. Input and output parameters and their ranges

Parameter
Maximum charge per delay (kg)
Distance from the blast-face (m)
PPV (mm/s)

24

Category

Range

Input
Input
Output

5.9-72.5
30.8-190
0.9-15.4

Table 6. Calculated values of site constants

Site constant
Equation
k

USBM [24]

68.35

-0.9

Ambraseys-Hendron [31]

104.99

-0.91

Davies et al. [32]

53.54

-0.88

0.51

Indian Standard [28]

5.42

0.65

25

Table 7. Performance results of different approaches used in this study

VARE

RMSE

MAPE

MEDAE

R2

NS

VAF

USBM

2.18

1.16

27.7

0.99

0.557

0.34

35.01

Ambraseys-Hendron

2.26

1.15

23.89

0.73

0.684

0.35

39.56

Davies et al.

2.55

1.19

29.45

0.98

0.495

0.3

31.02

Indian Standard

16.4

1.74

48.94

1.77

0.001

-0.47

-46.89

SVM

0.3

0.34

6.36

0.15

0.957

0.94

94.24

Model/Equation

26

Highlights

We measured data of 80 blasting operations in Bakhtiari Dam, Iran

We proposed a SVM model for predicting PPV resulting from blasting

Empirical equations were employed to predict PPV

A comparison was made to demonstrate capability of the models

27