You are on page 1of 11

Bull Eng Geol Environ

DOI 10.1007/s10064-017-1100-x

ORIGINAL ARTICLE

A new classification approach for prediction of flyrock throw


in surface mines
Turker Hudaverdi 1 & Ozge Akyildiz 1

Received: 19 December 2016 / Accepted: 10 June 2017


# Springer-Verlag GmbH Germany 2017

Abstract A novel classification approach was proposed for provide optimum rock fragmentation at minimum cost while
prediction of flyrock throw distance based on the site mea- considering environmental concerns. One of the most impor-
surements performed in a sandstone quarry. The classification tant adverse effects of blasting activities is flyrock. During
approach was created by using multiple discriminant analysis blasting, some rock pieces can be thrown and travel to very
(MDA) technique. The input parameters of discriminant anal- long distances. Usually this unexpected projection of rock
ysis are blast design parameters and a rock mass parameter. pieces is termed as flyrock (Bhandari 1997). Some conse-
The grouping stage of classification was performed consider- quences of flyrock are human response, serious injury or death
ing a well-known blast regulation for flyrock. Additionally, to employees, damage to properties and mine closure. Raina
multiple regression analysis technique applied to blast data et al. (2015) reported that, based on accident statistics, the 20–
to create a flyrock prediction equation. By this way, the capa- 40% of the blasting related accidents are due to flyrock.
bility and differences of the classification approach were in- Most important cause of flyrock is inappropriate blast de-
vestigated. Model validation was performed on the test blasts. sign. Researchers have been studied on prediction of flyrock
MDA model successfully estimated the flyrock throw dis- throw considering blast design parameters and some geotech-
tance. Unlike the classical prediction models, the MDA model nical factors. The most well-known investigation was
predicts severity of flyrock throws instead of a numerical val- achieved by the researchers of the Swedish Detonic
ue. MDA model can be easily practiced by the created terri- Research Foundation (Lundborg et al. 1975). They suggested
torial map. The model does not require any specific software a general equation for maximum flyrock throw length based
or training for usage and suitable for practical applications at on hole diameter. Also, size of the flying rock fragments was
mines. calculated. A few years later, a report was prepared for the
United States Bureau of Mines to create a model for the de-
Keywords Blasting . Flyrock . Classification . Discriminant termination of flyrock range as a function of shot conditions
analysis . Regression analysis (Roth 1979). Hillier et al. (1999) published another report for
limiting the instance of flyrock from quarry blasting opera-
tions. Some scholars focused on the basics of flyrock and tried
to understand effect of the blast design and geological param-
Introduction
eters on flyrock phenomena (Schneider 1997; Adhikari 1999).
Several researchers have approached the topic in terms of risk
Blasting is the main operation of excavation and production
management. These studies generally rely on analysis of his-
processes in mining. Blasting operation has to be designed to
torical flyrock accident data (Mishra and Mallick 2012;
Bajpayee et al. 2004). In recent years, various modeling tech-
* Turker Hudaverdi niques have been applied to predict flyrock throw distance.
hudaverdi@itu.edu.tr Monjezi et al. (2010, 2012), Rezaei et al. (2011) and Ghasemi
et al. (2012) applied artificial neural network and fuzzy logic
1
Department of Mining Engineering, Istanbul Technical University, techniques to predict flyrock throw. Amini et al. (2012) uti-
Maslak, 34469 Istanbul, Turkey lized a machine learning technique called support vector
T. Hudaverdi, O. Akyildiz

machine (SVM). Raina et al. (2011) used a probabilistic ap- dykes are intrusive in weak zones. Main rock structure is light-
proach for demarcation of flyrock danger zone in an open pit colored, thick − bedded sandstone. Rarely thin − bedded and
mine. Generally, all the researchers considered blast design laminated sandstone is observed in the upper zones of the
parameters as the input parameters of prediction models. quarry. Joint orientations vary by region. Main joints are char-
In this study, a new classification approach was proposed acterized by a NW-SE orientation. It has a NE-SW orientation,
for prediction of flyrock throw distance. The model was cre- more rarely. Because of the dense vegetation and structure of
ated based on multiple discriminant analysis (MDA). The aim Trakya formation, it is difficult to observe main joint system.
of the classification approach is to reduce the prediction error Mean fracture spacing in the blast faces is 1.15 m with a
and provide new viewpoints for evaluation of flyrock throw. standard deviation of 0.69 m. Flyrock measurements were
Unlike the existing models, classification approach estimates performed at the western benches of the Akdaglar quarry. In
severity of flyrock throws and uses a regulatory basis. the studied part of the quarry, there was no any significant
Formerly, Aler et al. (1996) used multiple discriminant anal- variation in the benches in terms of rock structure. The exces-
ysis (MDA) to create a blast fragmentation tool. The measured sive situations that may affect flyrock throw were strictly
blast data were classified into three groups based on the frag- monitored.
mentation efficiency. Castro et al. (1998) applied linear dis- The rock of the quarry is sandstone that contains 60 %
criminant analysis to quarry blasts to create a blasting index. silicium. The density of the rock is 2.70 g/cm3. The Mohs
The index helped the researchers to determine ease or diffi- hardness value is 5–6. The average porosity is 1.03%. The
culty of blasting operations. Hudaverdi et al. (2012) applied average compressive strength is 81 MPa and the average ten-
multiple discriminant analysis to evaluate blast fragmentation sile strength is 5.6 MPa. The Young’s modulus is 17 GPa.
in a quarry. The researchers compared the results of the dis- In blasting operations, Anfo (blasting agent) and emulsion
criminant analysis model and KuzRam fragmentation model. explosive (priming) were used. The diameter of the blast holes
In this research, blasting operations and flyrock incidents is 89 mm in the quarry. The blastholes are initiated by milli-
were observed in a sandstone quarry. The blast data was second electrical detonators. In each blast, between 2 and 4
grouped based on a well-known regulation about flyrock oc- rows, each of which has 10–40 holes, are detonated. The
currence in surface mines. Using multiple discriminant analy- number of the holes that are initiated instantaneously changes
sis (MDA), a predictive model was created for flyrock throw. between 3 and 6. Totally, 77 blast data were measured to
Also a prediction equation was developed by multiple regres- develop prediction models. A typical bench face in the studied
sion analysis. The results of MDA and regression analysis quarry was presented in Fig. 1.
were compared and the capabilities of MDA were discussed. All main blast design parameters were measured and re-
In the upcoming part of the paper, the modeling study is ex- corded to develop the flyrock prediction models. They are the
plained step by step. Burden (B, m), Spacing (S, m), Bench height (H, m),
Stemming (T, m) Hole diameter (D, m) and Powder factor
(Pf, kg/m3). The most critical and important parameter in sur-
Site investigation face blast design is burden (Hustrulid 1999). If the burden (B)
is too small, detonation gases escape to the atmosphere.
Site investigation was performed in Kemerburgaz Region lo- Escaped gases push the rock fragments and project them un-
cated in the western part of Istanbul. The regional geological controllably. Too large burden may also cause flyrock. The
structure is referred to Trakya formation. The formation sub- explosive in the hole cannot break the burden and the gases
stantially contains sandstone layers. Additionally, thin layered vent through the collar of the hole creating crater effect
shale and limestone are observed. Trakya formation consists (Olofsson 1990). Spacing (S) less than burden may induce
of carboniferous aged, fragmented sediments. It was de- premature splitting between holes and early loosening of the
formed, folded and fragmented by joint systems in different stemming. Both these effects cause rapid release of gases and
locations during Hercynian and Alpine Orogenesis period. excessive flyrock. Too close spacing may cause crushing and
The thickness of the sediments is between 5 and 50 cm. The cratering between holes and it also results in increase of
thickness reaches 50 to 100 cm in certain locations, especially flyrock incidents (Bhandari 1997). Bench height (H) divided
for sandstones (Tugrul and Undul 2006). by burden distance (H/B) indicates the stiffness ratio. Konya
There are four quarries operated in the region. Three of and Walter (1990) classifies the bench height based on H/B
them are neighbor quarries. This research was conducted in ratio. If the H/B ratio less than four and greater than one;
Akdaglar quarry. The production rate of the studied quarry is benches are considered as low benches. If the H/B is greater
4000 tons/day. The produced aggregate is mainly consumed than four, benches are classified as high benches. H/B ratio
by concrete and asphalt plants. Tectonic deformations are very should be more than two for good control of flyrock and
common in the studied quarry. Joints, folding, laminations and fragmentation. Insufficient stemming (T) can cause premature
thrust faults are observed in the benches. Andesite and basalt release of explosive gases and results in flyrock and poor
A new classification approach for prediction of flyrock throw

Fig. 1 A bench face of the


studied quarry

fragmentation. Conversely, excessive stemming results poor Since Ash’s (1963) suggestion about blast design ratios,
top breakage. That means large boulders in muckpile and several blasting researchers have considered blast design pa-
increase in amount of backbreak (Konya and Walter 1990). rameters as ratios. In this study, the blast design parameters of
The Powder factor (Pf) was added the model to reflect explo- all the blast data are also used as ratios. The ratio of bench
sive consumption per broken rock. Most blast fragmentation height to drilled burden (H/B), ratio of spacing to burden
formulas, like Kuznetsov (1973) and SVEDOFO (S/B), ratio of burden to hole diameter (B/D) and ratio of
(Ouchterlony 2003) equations, use powder factor as a funda- stemming to burden (T/B) are the blast design parameters used
mental blast design parameter. Excessive consumption of ex- for modeling. The apparent in-situ block size of the bench
plosive causes unnecessary overload. High powder factor faces was added to models to represent rock mass. In-situ
throws the flyrock to longer distances than low powder factor block size (XB) plays an important role in rock blasting oper-
(Bhandari 1997). The operational blast design parameters ap- ation and largely influences blasting performance. Rock
plied in the studied quarry are shown in Fig. 2, modified after blasting is considered to be a transformation from the state
Segarra et al. (2010). Generally, the values of the blast design of in situ block size distribution to the state of blasted block
parameters vary for each blast in the quarry. size distribution. The transformation energy is provided by

Fig. 2 The measured blast design


parameters in the studied quarry
T. Hudaverdi, O. Akyildiz

explosive material (Latham and Lu 1999). In addition to con- one discriminant function according to the number of
ventional measurements, WipJoint joint analysis software was groups. The first function maximizes the differences be-
used to determine the discontinuity properties and apparent in- tween the values of the dependent variable (groups). The
situ block size. WipJoint allows users to quantify bench char- second function is uncorrelated with it and maximizes the
acteristics such as joint spacing, orientation, and block size differences between values of the dependent variable, con-
(Hudaverdi et al. 2012). trolling for the first factor. The first function is the most
The on-site observations revealed that maximum flyrock powerful differentiating dimension, but later functions may
distance was due to face burst mechanism. Thus, the maxi- also represent additional significant contributions for dif-
mum horizontal distance between blast face and landed frag- ferentiation (Garson 2012). The discriminant analysis may
ments was considered as flyrock throw distance. Similar mea- be applied for (McLachlan 2004, Garson 2012):
surement technique was also applied by Ghasemi et al. (2012).
Modeling studies appearing in the literature also expect max- & Classification of the cases into the groups
imum throw length in front of the blast face (Richards and & Investigation of the differences between or among groups
More 2004). Flyrock throw distance was determined by using & Determination of the percent of variance in the dependent
tape measure and hand-held GPS (global positioning system). variable explained by the independents
The suggestions of the US Office of Surface Mining & Statistical pattern recognition
Reclamation and Enforcement (OSMRE) were followed dur- & Evaluation of the relative importance of the independent
ing GPS measurements (Eltschlager 2004). Only the frag- variables in classifying the dependent variable
ments which had the capacity to cause damage, injury or fa- & Analysis of classification accuracy of the grouped cases
tality were considered. Based on past observation in the stud-
ied quarry and the literature related to flyrock (Little 2007; Flyrock occurrence is controlled by several factors.
Ghasemi et al. 2012); these fragments have the approximate Therefore, a multivariate analysis should be performed to pre-
diameter of 10 cm. The descriptive statistics for the blast de- dict flyrock throw range. MDA is highly suitable for creation
sign parameters, apparent in-situ block size (XB) and mea- of a predictive model for flyrock. It is robust and flexible
sured flyrock throw distance (Ft) are presented in Table 1. technique. Flyrock occurrence may be grouped by severity
and MDA can be used to predict group membership. MDA
creates a practical territorial map. Unlike the existing soft
Multiple discriminant analysis (MDA) computing techniques appearing in the literature, MDA model
can be easily used by site engineers. The classification concept
Multiple discriminant analysis (MDA) is highly related to of MDA has a potential to provide new perspectives on the
multivariate analysis of variance (MANOVA) (Tabachnick topic of evaluation of flyrock in surface mines.
and Fidell 2013). MDA is sometimes also named as dis-
criminant factor analysis or canonical discriminant analysis Determination of group membership
(Garson 2012). MDA is performed to classify cases into
the values of a categorical dependent variable (groups for The first step of linear discriminant analysis is the creation of
our case). In discriminant analysis, the existence of the groups. Measured data is divided into groups (sets) and group
groups is known a priori (McLachlan 2004). A discrimi- membership of prospective measurements is determined by
nant function is created as a linear combination of discrim- the discriminant functions. The classification of data is per-
inating (independent) variables. MDA applies more than formed by categorical variables. Fuzzy logic technique also
utilizes data grouping approach based on categorical
(linguistic) variables. There are various approaches to deter-
Table 1 Descriptive statistics for modeling parameters mine group numbers and limits in discriminant analysis and
fuzzy logic. Intuition, inference, rank ordering, neural net-
Min Max Range Median Mode Mean Standard works, inductive reasoning etc. can be used for grouping
Deviation
(McNeill and Thro 1994). Researchers may also utilize spe-
S/B 1.00 1.28 0.28 1.19 1.20 1.18 0.05 cific regulations or limitations to define groups. In this study,
H/B 2.40 4.25 1.85 3.18 3.50 3.18 0.42 one of the most influenced and well-known flyrock regulation
B/D 20.22 28.09 7.87 24.72 28.09 25.17 2.39 was used to group the blast data. The flyrock regulation is a
T/B 0.91 1.67 0.76 27.72 1.00 1.09 0.13 part of the US Code of Federal Regulations Title 30 Mineral
Pf (kg/m3) 0.47 0.96 0.49 1.05 0.69 0.66 0.12 Resources (2016). Section 816.67 (Use of explosives: Control
Xb (m) 0.21 3.11 2.90 1.21 1.24 1.37 0.70 of adverse effects) determines allowable flyrock distance:
Ft (m) 28.0 142.0 114.0 74.0 30.0 79.2 35.7 Bc) Flyrock. Flyrock traveling in the air or along the
ground shall not be cast from the blasting site–
A new classification approach for prediction of flyrock throw

(1). More than one-half the distance to the nearest dwelling other (Huberty and Olejnik 2006). The discriminant func-
or other occupied structure; tions are given by:
(2). Beyond the area of control required under
Section 816.66(c); or  .   . 
(3). Beyond the permit boundary.^ DF1 ¼ 16:155  S B −3:105  H B þ 2:376
 .   . 
The Clause 2 and 3 indicate general restriction for area  B D þ 8:560  T B þ 24:420  ðPf Þ
security. Clause 1 gives a restrictive and definite descrip-
tion for the limitation of flyrock. It is possible to group þ 0:338  ðXB Þ−95:035 ð1Þ
flyrock throw distance (F t) according to the Clause 1  .   . 
above. In the studied quarry, the distance between blasting DF2 ¼ 40:936  S B −4:045  H B þ 3:044
area and nearest occupied structure was around 220 m. If  .   . 
the flyrock throw distance exceeds 110 m (one-half the  B D þ 15:837  T B þ 55:311
distance to the nearest dwelling), the incident can be de-
fined as an excessive flyrock. In our case, measured  ðPf Þ−0:435  ðXB Þ−165:512 ð2Þ
flyrock values were grouped as low, high and extreme
flyrock considering the regulation and the measured actual Discriminant functions (DF) are similar to regression equa-
data (Table 2). Basic three group classification (low-medi- tions obtained by multiple linear regressions. It contains dis-
um-high, short-medium-tall, low-high-excessive etc.) is criminant coefficients, discriminating variables and constant.
very common in engineering and rock blasting literature. Unstandardized discriminant coefficients are used to form dis-
Hudaverdi et al. (2012) proposed a blast fragmentation criminant functions. The discriminant function coefficients
prediction model based on three group classification. The are partial coefficients, reflecting the individual contribution
researchers grouped the blast data as insufficient, moderate of each variable (blast parameters for our case) to the classifi-
and highly efficient fragmentation based on the site cation of the criterion variable (groups). In the 1st discriminant
measurements. A discriminant analysis model was function (DF1), the discriminant coefficient of the discriminat-
created to predict the fragmentation efficiency of ing variable S/B is 16.155. The constant of DF1 is −95.035.
measured blasts. Aler et al. (1996) also used three group The classification results are presented in Table 3. Results
classification to create a blast fragmentation model. Table 2 of a multiple discriminant analysis are often presented in the
summarizes the classification process for the measured form of a classification table. Briefly, a classification table
flyrock throw distance. explains how the discriminant functions perform (Landau
and Everitt 2004; Ho 2014). As seen in Tables 3, 28 blasts
out of 29 have been classified correctly for Group 1. Only 1
Development of the multiple discriminant analysis model blast was misclassified. The percentage of correct classifica-
tion in Group 1 is 96.6%. For Group 3, 19 blasts out of 21
After the determination of the groups, multiple discrimi- have been classified correctly. The percentage of correct clas-
nant analysis (MDA) was performed. Six blast design and sification for all 77 blasts is 90.9%. This classification accu-
rock parameters are the Bdiscriminating variables^ of the racy is the indicator of the success of discriminant analysis.
analysis. The three groups, which are determined based on The high success percentage is also a confirmation for the
flyrock throw distance, are called as Bcriterion variable^.
Discriminant analysis generates two discriminant functions
(DF). The number of discriminant functions is one less
than number of the groups. The discriminant functions Table 3 Classification results
(DF) are created to obtain the combinations of discriminat-
ing variables that maximally separate the groups from each Group Predicted Group Membership Total

1 2 3
Table 2 The classification of the measured flyrock throw distance
Count 1 28 1 0 29
Group no Category Flyrock range Number of
2 2 23 2 27
blasts
3 0 2 19 21
1 Low flyrock throw Ft ≤ 55 m 29 % 1 96.6 3.4 0.0 100.0
2 High flyrock throw 55 m < Ft ≤ 110 m 27 2 7.4 85.2 7.4 100.0
3 Excessive flyrock throw 110 m < Ft 21 3 0.0 9.5 90.5 100.0
T. Hudaverdi, O. Akyildiz

accuracy and consistency of the classification of the blasts into at a significant level. It may be applied for statistical hypoth-
3 groups based on the flyrock throw distance. esis testing, prediction, construction of confidence intervals,
Test of equality of group means is presented in Table 4. and power analysis. Graphical visualization is used for clear
Wilks’ Lambda is a test statistic for multivariate analysis of depictions of the data and of the analytic results. Thus, multi-
variance (MANOVA). Wilks’ Lambda test is also used to ple regression analysis (MRA) is a versatile, all-purpose tech-
determine dominant discriminating variables on the occur- nique of analyzing the data over a wide range of sciences,
rence of the three groups. Lambda values vary between 0 technologies and engineering (Cohen et al. 2003).
and 1. The decrease of Wilks’ lambda value indicates the Multiple regression analysis (MRA) has been applied for
increasing contribution of the variable to the discriminant prediction of environmental adverse effects of rock blasting
functions (Garson 2012). As seen in Table 4, the lowest (Singh et al. 2008; Ghasemi et al. 2013; Saadat et al. 2014). In
Wilks’ Lambda values were obtained for B/D and Pf. The our case, MRA was performed to create a prediction formula
contribution of the B/D and Pf to the discriminant model is for flyrock throw distance. The aim of regression analysis is to
higher than the other blast design parameters. In other words, obtain a set of regression coefficients for the independent var-
the main differences among the groups are B/D, Pf, S/B and iables. By that way, predicted dependent variables from the
H/B, respectively. equation is bring as close as possible to the dependent vari-
The territorial map of discriminant analysis in presented in ables obtained by measurement (Tabachnick and Fidell 2013).
Fig. 3. Territorial map which is a graphical interpretation of Regression equation is formed by coefficents, independent
the analysis is plotted using discriminant function scores variables and constant. The six blast design parameters are
(SPSS Statistics Base 20 2011). The discriminant scores are the independent variables of the regression analysis.
obtained applying the discriminant function formulas (Eqs. (1 Measured Flyrock throw distance (Ft) is dependent variable.
and 2)) to the data belonging to a given blast. The symbol B⊕^ Regression equation is given by:
on the territorial map shows centroids of each groups. The
centroids of the groups may be used to see the mean discrim-  .   . 
inant scores of the groups on territorial map. The territorial Ft ¼ −33:045  S B þ 18:657  H B −17:664
map, also called canonical plots, is utilized to determine the  .   . 
group membership of the prospective blasts based on the dis-  B D −42:082  T B −105:316
criminant scores (Garson 2012). A random sample is present-
ed on Fig. 3 by dashed lines for the 14th measured blast data in  ðPf Þ−8:260  ðXB Þ þ 630:554 ð3Þ
the studied quarry. The measured S/B, H/B, B/D and T/B
values of Blast 14 are 1.14, 2.95, 24.72 and 0.91, respectively. Prediction formula is created by using unstandardized re-
Pf is calculated as 0.72 kg/m3. Measured XB value is 1.68 m. gression coefficients. In Eq. (3), the regression coefficient for
Two discriminant functions were applied to the Blast 14. The independent variable S/B is −33.045. The constant for the
calculated discriminant scores are DF1 = −1.08 and regression equation is 630.554. The constant indicates where
DF2 = −1.97, respectively. The intersections of the two scores the regression line intercepts the y axis (Value of Ft when all
reveal that the Blast 14 is belonging to the Group 2 (Fig. 3). the blast parameters are zero) (Tabachnick and Fidell 2013).
The model summary for MRA and Analysis of Variance
(ANOVA) are presented in Table 5. Obtained R2 value (coeffi-
cient of determination) is very high. Mathematically, R2 equals
Multiple regression analysis (MRA)
the regression sum of squares divided by total sum of squares in
the ANOVA table shown below (Garson 2014). High correlation
The multiple linear regression analysis is a powerful analytic
(R) also indicates the strength of the regression eq. F test is used
tool. It investigates whether a set of independent variables
to investigate significance of the model. F value in Table 5 is
explains a proportion of the variance in a dependent variable
obtained by dividing the mean square value of regression row by
Table 4 Tests of Equality of Group Means (MANOVA) the mean square of the residual row. Significance value is lower
than 0.05. In other words; the null hypothesis of no linear rela-
Wilks’ Lambda F df1 df2 Significance tionship of flyrock throw distance to the blast design parameters
is rejected (Dowdy et al. 2004)
S/B 0.714 14.822 2 74 0.000
H/B 0.717 14.594 2 74 0.000
B/D 0.163 190.312 2 74 0.000
T/B 0.894 4.410 2 74 0.016 Comparison of the models and the validation
Pf 0.231 122.915 2 74 0.000
XB 0.931 2.741 2 74 0.071 The discriminant analysis model (MDA) and the developed
regression equation were applied to 26 blast data to validate
A new classification approach for prediction of flyrock throw

Fig. 3 Territorial Map

and compare the models. In the first stage, 5 error measures 2011). In this study, a percentage error measure (Median
were used to examine the success of the regression equation Absolute Percentage Error, MdAPE) and a symmetric error
(Table 6). Generally, researchers use a few error measures to measure (Symmetric Mean Absolute Percentage Error,
evaluate accuracy of prediction equations in rock blasting. sMAPE) were also considered to provide additional
Absolute errors, especially Mean Absolute Error (MAE) and perspectives. Percentage errors are relative measures that ex-
Root Mean Square Error (RMSE), are very popular in the press errors as a percentage of measured actual data (Table 6).
blasting literature (Khandelwal 2010; Alipour and Ashtiani They provide information about magnitude of the errors.
T. Hudaverdi, O. Akyildiz

Table 5 Model summary and


analysis of variance for the Model Summary
regression analysis R R2 Adjusted R2 Standard Error Observations
0.918 0.843 0.830 14.7445 77
Analysis of variance (ANOVA)
Sum of Squares df Mean Square F Significance
Regression 81,762.054 6 13,627.009 62.682 0.000
Residual 15,218.024 70 217.400
Total 96,980.078 76

Symmetric errors were suggested to overcome the problem of classified Blats 8 into Group 1, low flyrock throw. The mea-
asymmetry of percentage errors and its possible influence by sured and predicted flyrock throw for Blast 19 are 117 and
outliers (Makridakis and Hibon 1995). Armstrong and 106 m, respectively (Fig. 4). Although regression equation
Collopy (1992) and Makridakis and Hibon (1995) discussed indicates high flyrock throw (55 m < Ft ≤ 110 m); MDA
error measures in detail. Contrary to the other error measures; correctly classified the Blast 19 into Group 3, excessive
the higher the VAF, the better the model performs. For in- flyrock throw (exceeds 110 m).
stance, a VAF of 100% means a perfect prediction for flyrock
throw has been achieved.
Figure 4 was created to examine the prediction capability A brief discussion on rock properties
of MDA and MRA models. Figure 4 also expresses the po-
tential of the classification approach. The model validation Variations in rock type and rock mass properties may affect
revealed that the MDA is able to accurately classify flyrock flyrock incidents. Variation in the grain size distribution of the
incidents. Correct classification percentage is %92.3. Only rock should be monitored in the studied quarry. Finer grain
two blasts were misclassified, Blast 9 and 17 (Fig. 4). size may increase possibility of the flyrock events compared
Particularly for the Blasts 3, 4, 5, 18 and 26, prediction error to coarser grain size. Bench face condition should also be
of the regression equation is lower than 5 m. However, for the considered in terms of blast damage and excessive backbreak.
blasts 10, 11 and 12, the prediction error exceeds 20 m. Damaged faces may increase severity of flyrock. Blasting
Classification approach (MDA) for flyrock distance is more damage is primarily achieved by stress wave interaction with
flexible than the prediction approach. Classification technique free surfaces. Compressive stress waves are reflected in ten-
does not try to predict an exact numerical value for flyrock sion upon encountering a free surface. Fractures occur as a
throw. MDA is more reliable particularly for the Blasts 8 and result of the reflected tensile stress waves (Yugo and Shin
19 (shown by blue arrow in Fig. 4). The measured Ft is 50 m 2015). If necessary, site engineers may create site-specific
for Blast 8. The regression equation calculated Ft as 59.6 m, indexes or charts to evaluate blast damage on bench faces.
indicating high flyrock throw. However, MDA was correctly The blast damage index proposed by Yu and Vongpaisal

Table 6 Calculated error values


for the regression equation Error Type Formula Calculated Error Value

Mean Absolute Error n 9.92


MAE ¼ 1n ∑ jyi −xi j
i¼1
Root Mean Square Error rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n
ffi 12.01
RMSE ¼ 1n ∑ ðyi −xi Þ 2
i¼1
Median Absolute Percentage Error jyi −xi j
9.47
MdAPE ¼ 100 mediani¼1;n yi
Symmetric Mean Absolute Percentage Error n 13.72
jyi −xi j
sMAPE ¼ 200
n ∑ yi þxi
i¼1
VAF   89.05
ðyi −xi Þ
VAF ¼ 1− varvary 100%
i

yi = Measured flyrock throw (Ft, m), xi = Predicted flyrock throw (Ft, m), n = Case number
A new classification approach for prediction of flyrock throw

Fig. 4 Investigation of the accuracy of MDA and regression analysis

(1996) may be examined to understand important parameters prediction models. In the future, rock structure may changes
that influence blast damage. Yu and Vongpaisal (1996) created significantly in the different part of the studied quarry. The
a damage index considering the induced stress and damage developed MDA and MRA models should be updated de-
resistance. The induced stress is defined as dynamic tensile pending upon the time and progressing of the mine.
stresses, which are caused when compressive strain waves are Discriminant analysis and regression analysis are non-rigid
reflected at the free face. It is calculated by using vector sum of and very adaptive techniques. When the need arises, the blast
peak particle velocity, rock density and compressional wave database can be changed or renewed by adding new blast
velocity of rock mass. The damage resistance was expressed measurements. It is possible to add new parameters to the
in the form of the dynamic tensile strength of the rock mass developed models. MDA and MRA models presented in this
multiplied by the site quality constant. paper can be easily adapted to a particular situation.
Variation in rock mass or extreme conditions may increase
severity of flyrock incidents. Bhandari (1997) indicated that
zone of weakness and voids may cause excessive flyrock. An Conclusions
unexpected lack of resistance to drill penetration may be a
warning for a mud seam, a zone of weak rock or a void. A novel multiple discriminant analysis (MDA) model was
Recently, Raina et al. (2014) pointed out the importance of created considering blast design parameters and rock mass
presence of voids, incompetent beds in competent rock, parameter. The accuracy of the model was tested on the vali-
known or unknown anomalies in the strata. Incompetent zone dation blasts. MDA model was successfully estimated the
or voids in the bench may be stemmed instead of explosive group membership of the blasts. At the same time, a prediction
charging to prevent flyrock (Olofsson 1990). Persson et al. equation was created by multiple regression analysis. By this
(1994) recommended mapping extreme joints and fissures that way, the capability and differences of the classification ap-
can cause problem for drilling and blasting. proach was investigated.
Generally, blast researchers suggest revisions for the devel- MDA may estimate severity of the flyrock. It does not
oped ground vibration and airblast prediction models provide an exact numerical value for flyrock throw. This
(Kahriman 2004; Giraudi et al. 2009; Hudaverdi 2012 etc.). may be helpful for public relations. If a numerical value is
Revision is also important for the site-specific flyrock estimated, inaccurate predictions are noticed by the people
T. Hudaverdi, O. Akyildiz

more easily. For our case, even though high correlation coef- Amini H, Gholami R, Monjezi M, Torabi SR, Zadhesh J (2012)
Evaluation of flyrock phenomenon due to blasting operation by
ficients were obtained, some predictions of regression equa-
support vector machine. Neural Comput Applic 21:2077–2085.
tion deviate more than 10–20 m. It should be kept in view that doi:10.1007/s00521-011-0631-5
flyrock throw length may be presented using categorical var- Armstrong JS, Collopy F (1992) Error measures for generalizing about
iables instead of numerical values. forecasting methods: empirical comparisons. Int J Forecast 8:69–80.
doi:10.1016/0169-2070(92)90008-W
MDA model relies on a well-known legal regulation for
Ash RL (1963) The mechanics of rock breakage (part 2) – standards for
flyrock. US Code of Federal Regulations for Mineral blast design. Pit Quarry 56:118–122
Resources were used to determine the groups. By this way, Bajpayee TS, Rehak TR, Mowrey GL, Ingram DK (2004) Blasting inju-
safe and dangerous flyrock throw distances are decided by ries in surface mining with emphasis on flyrock and blast area secu-
rity. J Saf Res 35:47–57. doi:10.1016/j.jsr.2003.07.003
available regulations. Relying on legislations strengthens
Bhandari S (1997) Engineering rock blasting operations. A. A, Balkema
hands of miners. MDA is a very flexible technique. Number Castro JT, Liste AV, Gonzales AS (1998) Blasting index for exploitation
of the groups and definition of group limits may be adjusted of aggregates. In: Proceedings of the 7th Mine Planning and
according to local requirement. Miners may create new MDA Equipment Selection Calgary, pp. 165–168
Cohen J, Cohen P, West SG, Aiken LS (2003) Applied multiple
models considering regional regulations.
regression/correlation analysis for the behavioral sciences.
Another advantage of the both MDA and regression model Lawrence Erlbaum Associates Publishers, London
is ease of use. MDA model is easily practiced by a territorial Dowdy S, Wearden S, Chilko D (2004) Statistics for research − third
map and common spreadsheets. Discriminant scores of linear edition. John Wiley & Sons Inc., Hoboken
discriminant functions are calculated by using a spreadsheet. Eltschlager KK (2004) Blasting applications for GPS. Proceedings of the
Advanced Integration of Geospatial Technologies in Mining and
Then even a slight examination of territorial map may help Reclamation Atlanta, Georgia, In, pp 1–11
miners to estimate severity of possible flyrock incidents. Garson GD (2012) Discriminant function analysis −statistical associates
However, complex modeling techniques appearing in blasting blue book series. Statistical Associates Publishing, Asheboro
literature generally requires expertise. For example, a fuzzy or Garson GD (2014) Multiple regression − statistical associates blue book
series. Statistical Associates Publishing, Asheboro
neural network model cannot be used by untrained miners. Ghasemi E, Sari M, Ataei M (2012) Development of an empirical model
These models also require specific software. for predicting the effects of controllable blasting parameters on
Flyrock is a complex problem and influenced by several flyrock distance in surface mines. Int J Rock Mech Min Sci 52:
factors. Therefore, prediction of flyrock incorporating all the 163–170. doi:10.1016/j.ijrmms.2012.03.011
Ghasemi E, Ataei M, Hashemolhosseini H (2013) Development of a
relevant factors is an extremely challenging task. There are fuzzy model for predicting ground vibration caused by rock blasting
various suggested models appearing in the literature to evalu- in surface mining. J Vib Control 19:755–770. doi:10.1177/
ate flyrock throw. Each model contributes the literature and 1077546312437002
provides new perspectives. In the future, attempts may be Giraudi A, Cardu M, Kecojevic V (2009) An assessment of blasting
vibrations: a case study on quarry operation. Am J Environ Sci 5:
made to create new multiple discriminant analysis models 467–473. doi:10.3844/ajessp.2009.467.473
considering additional technological and geological parame- Hillier DE, Holywell PD, Jeffries RM, Scott IMB (1999) Limiting the
ters in various mining sites. Classification approach intro- instance of fly-rock from quarry operations, research report. WS
duced in this paper has shown the capability of developing Atkins Consultants Ltd., Warrington
Ho R (2014) Handbook of univariate and multivariate data analysis with
accurate prediction models for flyrock throw distance. IBM SPSS − second edition. CRC Press, Boca Raton
Huberty CJ, Olejnik S (2006) Applied MANOVA and discriminant anal-
ysis −, second edn. John Wiley & Sons, New Jersey
Acknowledgments This study was partly supported by the Research Hudaverdi T (2012) Application of multivariate analysis for prediction of
Fund of the Istanbul Technical University (project name: ‘The effects of blast-induced ground vibrations. Soil Dyn Earthq Eng 43:300–308.
the variations in blast design and initiation systems on blast induced doi:10.1016/j.soildyn.2012.08.002
ground vibrations. No: 38511). The authors are grateful to the Research Hudaverdi T, Kuzu C, Fisne A (2012) Investigation of the blast fragmenta-
Fund of the Istanbul Technical University for their financial support. tion using the mean fragment size and fragmentation index. Int J Rock
Mech Min Sci 56:136–145. doi:10.1016/j.ijrmms.2012.07.028
Hustrulid WA (1999) Blasting principles for open pit mining−Vol 1 gen-
eral design concepts. A. A, Balkema
References IBM SPSS Statistics Base 20 (2011) IBM Corp., Armonk
Kahriman A (2004) Analysis of parameters of ground vibration produced
from bench blasting at a limestone quarry. Soil Dyn Earthq Eng 24:
Adhikari GR (1999) Studies on flyrock at limestone quarries. Rock Mech 887–892. doi:10.1016/j.soildyn.2004.06.018
Rock Eng 32:291–301. doi:10.1007/s006030050049 Khandelwal M (2010) Evaluation and prediction of blast-induced ground
Aler J, Du Mouza J, Arnould M (1996) Evaluation of blast fragmentation vibration using support vector machine. Int J Rock Mech Min Sci
efficiency and its prediction by multivariate analysis procedures. Int 47:509–516. doi:10.1016/j.ijrmms.2010.01.007
J Rock Mech Min Sci Geomech Abstr 33:189–196. doi:10.1016/ Konya CJ, Walter EJ (1990) Surface blast design. Prentice Hall Int, USA
0148-9062(95)00055-0 Kuznetsov VM (1973) Mean diameter of fragments formed by blasting
Alipour A, Ashtiani M (2011) Fuzzy modeling approaches for the pre- rock. Sov Min Sci 9:144–148. doi:10.1007/bf02506177
diction of maximum charge per delay in surface mining. Int J Rock Landau S, Everitt BS (2004) A handbook of statistical analyses using
Mech Min Sci 48:305–310. doi:10.1016/j.ijrmms.2010.11.010 SPSS. Chapman & Hall/CRC, Boca Raton
A new classification approach for prediction of flyrock throw

Latham JP, Lu P (1999) Development of an assessment system for the Raina AK, Murthy VMSR, Soni AK (2015) Flyrock in surface mine
blastability of rock masses. Int J Rock Mech Min Sci Geomech blasting: understanding the basics to develop a predictive regime.
Abstr 36:41–55. doi:10.1016/S0148-9062(98)00175-2 Curr Sci 108:660–665
Little TN (2007) Flyrock Risk. In: Proceedings of the EXPLO 2007 Rezaei M, Monjezi M, Yazdian Varjani A (2011) Development of a fuzzy
Conference Wollongong, Australia. pp. 35–43 model to predict flyrock in surface mining. Saf Sci 49:298–305. doi:
Lundborg N, Persson PA, Ladegaard-Pedersen A, Holmberg R (1975) 10.1016/j.ssci.2010.09.004
Keeping the lid on flyrock in opencast blasting. Eng Min J 95–100 Richards AB, More AJ (2004) Flyrock Control – By Chance or Design.
Makridakis S, Hibon M (1995) Evaluating accuracy (or error) measures. In: Proceedings of the 30th Annual Conference on Explosives and
INSEAD Working Paper Series. No. 18/TM:1–31. https://sites. Blasting Technique New Orleans, Louisiana. pp.1–13
insead.edu/facultyresearch/research/doc.cfm?did=46875 Roth J (1979) A model for the determination of flyrock range as a func-
McLachlan GJ (2004) Discriminant analysis and statistical pattern recog- tion of shot conditions, final report contract no. J03872A2.
nition. John Wiley & Sons, New Jersey Management Science Associates, Los Altos
McNeill FM, Thro E (1994) Fuzzy logic: a practical approach. AP Saadat M, Khandelwal M, Monjezi M (2014) An ANN-based approach
Professional, Cambridge to predict blast-induced ground vibration of Gol-E-Gohar iron ore
Mishra AK, Mallick DK (2012) Analysis of blasting related accidents mine, Iran. J Rock Mech Geotech Eng 6:67–76. doi:10.1016/j.
with emphasis on flyrock and its mitigation in surface mines. In: jrmge.2013.11.001
Proceedings of the 10th International Symposium on Rock Schneider L (1997) Back to the basics, flyrock (part 2: prevention). The
Fragmentation by Blasting New Delhi, India. pp. 555–563 Journal of explosives engineering 14:1−14 quarry blasting. Appl
Monjezi M, Bahrami A, Yazdian Varjani A (2010) Simultaneous predic- Acoust 71:1169–1176. doi:10.1016/j.apacoust.2010.07.008
tion of fragmentation and flyrock in blasting operation using artifi-
Segarra P, Domingo JF, López LM, Sanchidrián JA, Ortega MF (2010)
cial neural networks. Int J Rock Mech Min Sci 47:476–480. doi:10.
Prediction of near field overpressure from quarry blasting. Appl
1016/j.ijrmms.2009.09.008
Acoust 71:1169–1176. doi:10.1016/j.apacoust.2010.07.008
Monjezi M, Mehrdanesh A, Malek A, Khandelwal M (2012) Evaluation
of effect of blast design parameters on flyrock using artificial neural Singh TN, Dontha LK, Bhardwaj V (2008) Study into blast vibration and
networks. Neural Comput & Applic 23:349–356. doi:10.1007/ frequency using ANFIS and MVRA. Min Technol 117:116–121.
s00521-012-0917-2 doi:10.1179/037178409X405741
Olofsson SO (1990) Applied explosives technology for construction and Tabachnick BG, Fidell LS (2013) Using multivariate statistics, Sixth edn.
mining. Applex, Sweden Pearson, New Jersey
Ouchterlony F (2003) Influence of blasting on the size distribution and Tugrul A, Undul O (2006) Engineering geological characteristics of
properties of muckpile fragments, a state-of-the-art review, MinFo Istanbul greywackes. In: Proceedings of the 10th International
project report P2000–10: energy optimisation in comminution. Association for Engineering Geology and the Environment
Lulea University of Technology, Sweden, Swebrec (IAEG) Congress Nottingham, United Kingdom. Paper no. 395
Persson PA, Holmberg R, Lee J (1994) Rock blasting and explosives US Code of Federal Regulations-Title 30 Mineral Resources (2016)
engineering. CRC Press, Boca Raton Office of the Federal register national archives and records admin-
Raina AK, Chakraborty AK, Choudhury PB, Sinha A (2011) Flyrock danger istration. U.S. Government Publishing Office, Washington, DC
zone demarcation in opencast mines: a risk based approach. Bull Eng Yu TR, Vongpaisal S (1996) New blast damage criteria for underground
Geol Environ 70:163–172. doi:10.1007/s10064-010-0298-7 blasting. CIM Bull 89:139–145
Raina AK, Murthy VMSR, Soni AK (2014) Flyrock in bench blasting: a Yugo N, Shin W (2015) Analysis of blasting damage in adjacent mining
comprehensive review. Bull Eng Geol Environ 73:1199–1209. doi: excavations. J Rock Mech Geotech Eng 7:282–290. doi:10.1016/j.
10.1007/s10064-014-0588-6 jrmge.2014.12.005

You might also like