Microsoft Word - 001.docx


 CHEMICAL ENGINEERING TRANSACTIONS  
 

VOL. 66, 2018 

A publication of 

 

The Italian Association 
of Chemical Engineering 

Online at www.aidic.it/cet 

Guest Editors: Songying Zhao, Yougang Sun, Ye Zhou 
Copyright © 2018, AIDIC Servizi S.r.l. 

ISBN 978-88-95608-63-1; ISSN 2283-9216 

Assessment of Blast-induced Ground Vibration using Different 

Predictor Approaches- A Comparison  

Prashanth Ragama, D.S. Nimajeb* 

a Research Scholar, National Institute of Technology, Rourkela, India  
b*Assistnat Professor, National Institute of Technology, Rourkela, India 

dsnimaje@nitrkl.ac.in 

Blasting is a mechanism of rock fragmentation. Usages of high explosive causes ground vibration, air noise, 

back breaks, over breaks as well as fly rock. Excessive blast-induced vibrations cause severe damage to nearby 

structures and residents in and around mine. So, it is essential to the precise prediction of ground vibration to 

reduce the ecological damage. Vibrations are expressed in terms of peak particle velocity (PPV). The aim of 

this study was to assess and predict the ground vibration at different explosive quantities and distances, 

applications of generalized regression neural network (GRNN) as well as empirical predictors were used. A total 

of fourteen blast events were collected at various strategic and valuable locations in the site, out of these ten 

blast events were considered for training and reaming for validate of the GRNN model. A three-layer GRNN with 

2-14-1 architecture was developed and trained with Levenberg–Marquardt algorithm. Five emphircal predictor 

equations were proposed by the United States Bureau of Mines (USBM), Ambraseys-Hendron, Langefors-

Kihlstrom, Central Mining Research Institute (CMRI) predictor and Bureau of Indian standard were applied to 

governing a relation between peak particle velocity and its influencing parameters. The obtained results reveal 

that the proposed GRNN model can predict the PPV accurately as compared to the other predictor models 

available. Obtained results were compared based on evaluation performance models such as coefficient of 

determination (R2) and mean square error (MSE) between monitored and predicted values of PPV. It was 

observed that GRNN approach provides high R2 (0.9988) and low MSE (0.0001) among all other empirical 

predictor approaches for accurate prediction of ground vibration. 

1. Introduction 

Mineral resources are the backbone of any industrial nation and industry needs metal and non-metals as raw 

material. These are extracted by both underground and open cast mining methods. In both cases, blasting 

mechanism was practiced to fragment the rock-mass. During the charging of holes with explosive, larger amount 

of energy was released in the form of pressure and temperature of about 50Gpa and 5000k respectively (Parida 

and Mishra, 2015). Only 20% of explosive energy was utilized for an actual fragment of the rock mass and 

remain dissipates in the form of ground vibration, fly rock, air noise, over break etc.(Cheng and Huang, 2006). 

Excessive intensity levels of blast-induced ground vibration can damage to or failure of structures. The intensity 

of ground vibration levels depends on various influencing parameters like distance, an explosive charge, 

spacing, burden and hole depth, etc. Ground vibrations are quantified by means of particle velocities at particular 

ground locations. Presently, the most widely considered term to measure the ground vibration is PPV and It is 

defined as the maximum speed at which each particle passes or moves in a ground to its inactive state. Thus, 

it is very important to evaluate and predict the blast-induced vibrations to know the safe pin point levels for 

reducing damage levels.  

Various scientists, academicians, and researchers were proposing a number of empirical predictors approach 

for the estimation as well as prediction of ground vibration due to blasting and are summarized in Table 1. 

Proposed predictors are able to predict the PPV based on two parameters such as a maximum explosive charge 

per delay and distance between the blasting sources to monitored point at the particular site. 

 

                                

 
 

 

 
   

                                                  
DOI: 10.3303/CET1866082 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Please cite this article as: Ragam P., Nimaje D., 2018, Assessment of blast-induced ground vibration using different predictor approaches- a 
comparison, Chemical Engineering Transactions, 66, 487-492  DOI:10.3303/CET1866082   

487



Table 1: Different empirical predictor equations 

Predictor name Equation 

USBM (Duvall and Fogelson, 1962)  𝑉 = 𝐾 [
𝐷

√𝑄𝑚𝑎𝑥
]

−𝐵

 

Amraseys-Hendron (Ambraseys and 

Hendron, 1968)  
𝑉 = 𝐾 [

𝐷

(𝑄𝑚𝑎𝑥)
1
3

]

−𝐵

 

Langefors-Kihstrom (Langefors and 

Kihstrom, 1963)  𝑉 = 𝐾[
√(𝑄𝑚𝑎𝑥/ 𝐷

2
3) ]

 𝐵

 

Bureau of Indian standard (Bureau of 

Indian standard, 1973)  𝑉 = 𝐾[(𝑄𝑚𝑎𝑥/ 𝐷
2

3)]B 

 CMRI (Roy, 1991)  𝑉 = 𝑛 + 𝐾 [
𝐷

√𝑄𝑚𝑎𝑥
]

−1

 

 

Where, 𝑉 is PPV (mm/s), Q is maximum charge per delay (kg), D is the distance between the blasting face to 

vibration monitoring point (m), K, B and 𝑛 are site constants and scaled distance 𝑆𝐷 =
Qmax𝑘1

𝐷𝑘2
 (k1 and k2 are 

predefined). The flaws of existing predictor approaches provide different predictor values of safe PPV vis-à-vis 

safe maximum charge per delay for the particular site and no uniformity in the predicted results. It is essential 

to evaluate and predict the PPV accurately. Therefore, many researchers are using soft computing techniques 

like artificial neural network (ANN), support vector machine algorithm (SVM), fuzzy theory, etc. for various 

applications.  

Ragam and Nimaje (2018a) developed a multilayer perceptron (MLP) based back propagation (BP) neural 

network model and to predict the ambiguous ground vibration at ACC Dungri lime stone mine, India. The 

observed results ensure the ANN model provides accurate results compared to other conventional models. 

Ragam and Nimaje (2018b) used an ANN model to estimate the PPV levels at Mine-A, Inida and the results of 

ANN was more close to observed PPV values among other models. Nimaje and Tripathy (2017) employed radial 

basis function neural network (RBFNN) to predict the fire risk of Indian coals and revealed that RBFNN model 

provides superior results over statisitcal models. Ahmad et al. (2017) were propsed multple neural network 

(MNN) to predict the combustion efficiency from the boiler. Liu and Wang (2016) introduced the support vector 

regression (SVR) model based on ant colony optimization for forecasting the concentration of pollutants in the 

air. 

In the current study, the blast experiment test was conducted at IDL Explosive Limited and recoreded fourteen 

blast events at different distances from the blast-source. The GRNN model along with five empirical predictor 

approaches are used for evaluation and prediction of the ground vibration levels and compared the results. 

2. Blast site description 

The ground vibration phenomenon due to blasting was studied at IDL Explosives Limited (IDL). The blast site 

location coordinates are 22°11'12.8"N 84°52'28.4"E and located at Sonaparbat area, Rourkela, India. IDL is the 

leading manufacturing of explosives and accessories located at the outskirt of Rourkela, which houses services 

for the production of a total range of packaged explosives and non-explosive emulsion matrix, an intermediary 

for delivery of bulk explosives along with accessories. The blast sitemap is depicted in Figure 1. Explosion clad 

plates are manufactured to use in various applications. The cladding is being conducted on the sand base 

surface which is spread uniformly and the plates are cladded with the help of explosives in form of a powder 

which is initiated by the remote device. The pressure released from this cladding by the explosives joins two 

different metal plates. This process generates sound and ground vibration due to blasting in the surroundings. 

3. Instrumentation and data collection 

Blasting events are monitored with the help of seismograph (make: Instantel. Inc, Canada- Miniamate plus along 

with two geophones and one microphone). In this study, total 14 blast events were recorded over a period of 

one month. Figure 2 depicts the monitoring of ground vibration with the seismograph instrument. The PPV data 

was monitored at various distances from the blast source with respect to the maximum charge per delay and 

the results are listed in Table 2. 

488



 

Figure 1: Site Map of IDL–Explosives Ltd 

Figure 2: Monitoring of ground vibration with minimate 

plus seismograph 

Table 2: Recorded PPV data at various distances at varying explosive charge 

S.No Distance (m) Max. Explosive per delay (kg) PPV (mm/s) 

1 354 540 1.48 
2 493 180 1.02 
3 500 300 1.78 
4 90 500 1.11 
5 50 564 0.64 
6 200 720 0.32 
7 195 777 0.57 
8 100 1020 0.32 
9 540 1320 0.25 
10 150 1500 0.19 
11 530 1520 0.38 
12 90 1850 0.44 
13 270 1990 0.32 
14 270 2750 0.19 

Note: Table 2 monitored data was listed based on day wise collection of results 

4. Data analysis 

4.1 Prediction of PPV using empirical predictor approaches 

All the empirical predictor approaches must have site-specific constants (Table 1). The specific constants are 

varied based on changes in ground vibration. The site constants (K, B, and n) are calculated by plotting graph 

between scaled distance and PPV on the log-log scale using the following Eq(1) and Eq(2) 

log 𝑣 = log 𝑘 − 𝐵 log 𝑆𝐷                                                                                                                                     (1) 

The general equation of the straight line is 

𝑦 = 𝑚𝑥 + 𝐶                                                                                                                                                        (2) 

Table 3: Calculated values of site constants 

Predictor equation K B 𝑛 

USBM 14.36 -0.80 -- 

Langefors-Kihlstrom 0.374 0.7766 -- 

Amraseys-Hendron 51.39 0.913 -- 

Bureau of Indian standard 0.374 0.3883 -- 

CMRI  Predictor (n=0) 14.361 -- -1 

 

Here, PPV and SD data should exhibit a straight line on a log-log graph. Therefore, intercept 𝐶 = log 𝑘 and slope 

−𝐵 = 𝑚. K, B, and n were determined by multiple regression analysis using SPSS-20 package. Figures 3- 7 

graphically represented the correlation between measured and predicted PPV by different SD laws. Table 3 

illustrates the site constants for different predictor models. The Amaresyes-Hendron and USBM have quite 

remarkable R2 while Langefors-khilstom and Bureau of Indian Standard are having less R2. 

489



4.2 Prediction of PPV using GRNN 

GRNN is a one-pass neural network with highly parallel construction. Specht (1991) proposed GRNN, is a 

variation of the radial basis function neural network (RBF).Basically, it is an algorithm based on function 

approximation (estimation) which is a statistical technique named as kernel regression (Xue and Yang, 2014). 

The training of GRNN is very quick and data needs propagated only once forwarded, unlike other neural network 

algorithms. A typical architecture of GRNN is depicted in Figure 9. The desired output is determined by taking 

an average of assigned weights of training output data set. The weight of each output is calculated using the 

Euclidean distance function between the training and testing data. If the Euclidean distance is more than the 

total weighs at the output is very less otherwise the additional weight should be assigned to output.  

 

Figure 3: USBM predictor 

 

Figure 4: Amraseys-Hendron Predictor 

 

Figure 5: Langefors-khilstom Predictor 

 

Figure 6: Bureau of Indian Standard Predictor 

 

Figure 7: CMRI Predictor 

 

Figure 8: GRNN model 

A GRNN contains four layers: an input layer, a pattern layer, summation layer and an output layer. The size of 

input neurons in input layer depends on the total number of the observed parameters. The input layer feeds the 

input to pattern layer and each neuron presents a training pattern and output. The main purpose of pattern layer 

is calculating the Euclidean distance along with activation function and forwarded to the summation layer 

(Alawode and Alawode, 2014). The summation layer has two sub-parts: numerator (N) and denominator (D). 

The numerator part consists of addition(summation) of the multiplication of training output data along with 

activation function and denominator part having the addition of all specified activation function. This summation 

layer feeds both the data of numerator part and denominator part to the output layer. The output layer is 

consisting of one neuron which determines the output by merely dividing the output of each numerator part (N) 

to each of denominator part (D), yielding the predicted output Y(x) to an unknown input vector x as 

490



Y(x) =
∑ Yi e

−(
di

2

2σ2
)

∑  e
−(

d
i
2

2σ2
)

                                                                                                                                                 (3) 

di
2 = (x − xi )

T(x − xi)                                                                                                                                        (4) 

Where x is the input sample and xi is the training sample and Yi is output sample. di
2 is the Euclidian distance 

from 𝑥𝑖  to 𝑥. e
−(

di
2

2σ2
)
 which is an activation function. Basically, activation function is the weight of the input data. 

Here, the unknown spread parameter is constant (𝜎). It can be adjusted by training process to an optimum, 

where the error should be minimized. The procedure of training is to find out the optimum of 𝜎. It varies between 

0.0001 to 1. The best practice is to minimize the MSE. All fourteen data sets were divided into training and 

testing datasets as per the thumb rule. The Training of the network was carried out on 70% data sets and the 

remaining data sets were used for testing and evaluation of the network using MATLAB 2015b. The prediction 

response of developed GRNN model is graphically represented in Figure 10. 

 

Figure 9: GRNN architecture 

 

Figure 10: The response of GRNN model 

5. Results and discussion 

The Figure 8 shows the correlation between the measured and predicted PPV using GRNN model. The analysis 
of all predictor models in terms of evaluation performance models i.e. R2 and MSE are summarized in Table 4. 
The graphical comparison of measured and predicted PPV of all proposed approaches are depicted in Figure 
11. It was observed that the GRNN predictor approach has highest R2 and lowest MSE as compared to other 
predictors. 

Table 4: Correlation coefficient and error parameter 

Predictor R2 MSE 

USBM 0.63257 0.0923 

Amraseys-Hendron 0.69816 0.7665 

Langefors-khilstom 0.31097 0.1916 

Bureau of Indian standard 0.31097 0.1916 

CMRI 0.63509 0.0926 

GRNN 0.9988 0.0001 

491



 

Figure 11: Comparison of measured and predicted PPV using different predictors 

6. Conclusions 

The blast-induced ground vibration evaluation analysis has been carried out using five established empirical 

predictors and GRNN model. The parameters such as distance and maximum explosive per delay were taken 

as input parameters and PPV as an output for all proposed predictors. The obtained results evident that GRNN 

model provides significant R2 (0.998) and less MSE(0.0001) as compared to other predicted models. Therefore, 

it shows more accuarate and better prediction results between recorded and predicted PPVs. 

References  

Ahmad Z., Bahadorib A., Zhangc J., 2017, Prediction of Combustion Efficiency using Multiple Neural Networks, 

Chemical Engineering Transactions, 56, 85-90. 

Alawode T.T., Alawode K.O., 2014, Prediction of substituent types and positions on skeleton of eudesmane-

type sesquiterpenes using generalized regression neural network (GRNN), African Journal of Pure and 

Applied Chemistry, 8, 102-109. 

Ambraseys N.R., Hendron, A.J.(Ed), 1968, Rock Mechanics in Engineering Practices, Wiley press, London. 

Bureau of Indian Standard, 1973, Criteria for safety and design of structures subjected to underground blast, 

ISI, India. 

Cheng G., Huang S.L., 2000, Analysis of ground vibration caused by open pit production blast Explosive and 

Blasting Technique, Holmberg(Ed), Balkema. 

Duvall W.I., Fogelson D.E., 1962, Review of criteria for estimating damage to residences from blasting 

vibrations, US Department of the Interior, Bureau of Mines, USA. 

Langefors U., Kihlström B., (2nd Ed), 1963, The modern technique of rock blasting, Wiley, New York. 

Liu J.F., Wang Q.M., 2016, Application of an Improved SVM Algorithm for Wireless Sensor Networks in the 

Prediction of Air Pollution, Chemical Engineering Transactions, 51, 337-342, DOI: 10.3303/CET1651057 

Nimaje D.S., Tripathy D.P., 2017, Fire risk assessment of some Indian coals using radial basis function (RBF) 

technique, Journal of The Institution of Engineers (India): Series D, 98, 49-58. 

Parida A., Mishra M.K., 2015, Blast vibration analysis by different predictor approaches- A comparison. Procedia 

Earth and Planetary Science, 11, 337-345. 

Ragam P., Nimaje D.S., 2018a, Monitoring of blast-induced ground vibration using WSN and prediction with an 

ANN approach of ACC dungri limestone mine, India, Journal of Vibroengineering, 20, 1051-1062. 

Ragam P., Nimaje D.S., 2018b, Evaluation and prediction of blast-induced peak particle velocity using artificial 

neural network: A case study, Noise & Vibration Worldwide, 49, 111-119, DOI: 10.1177/0957456518763161 

Roy P.,1993, Putting ground vibration predictions into practice, Colliery Guardian, 241, 63-67. 

Specht D.F., 1991, A general regression neural network, IEEE transactions on neural networks, 2, 568-576. 

Xue X., Yang X., 2014, Predicting blast-induced ground vibration using general regression neural network, 

Journal of Vibration and Control, 20, 1512-1519. 

492

http://dx.doi.org/10.3303/CET1651057
http://dx.doi.org/10.1177/0957456518763161