See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/308018215

Probabilistic analysis of crushed zone for rock blasting

Article  in  Computers and Geotechnics · December 2016

DOI: 10.1016/j.compgeo.2016.08.025

CITATION READS

1 158

2 authors, including:

Mahdi Shadab Far

Tongji University
11 PUBLICATIONS   11 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Mahdi Shadab Far on 08 March 2018.

The user has requested enhancement of the downloaded file.

 Computers and Geotechnics 80 (2016) 290–300

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research Paper

Probabilistic analysis of crushed zone for rock blasting

Mahdi Shadab Far 1, Yuan Wang ⇑,2
Hohai University, No. 1, Xikang Road, Nanjing, China

a r t i c l e i n f o a b s t r a c t

Article history: Transmission of explosion waves through a rock medium causes a severe vibration that stimulates the
Received 27 May 2016 mechanical behavior of rock mass. This stimulation imposes highly concentrated stresses on the ends
Received in revised form 30 July 2016 of existing fine joints and depending on the toughness of the rock, causes them to propagate rapidly.
Accepted 27 August 2016
Consequently, the propagation and joining of cracks form a crushed zone around the blast hole.
Several studies are available in the literature to estimate the radius of crushed zone, deterministically.
In this paper, however, a probabilistic approach has been adopted. This is because the initiation and prop-
Keywords:
agation of cracks have a probabilistic nature, and neither the initial state of the rock nor the explosion
Explosion load
Crushed zone radius
load could be expressed in a fully deterministic way. Thus, after generating random values for involved
Exceedance probability parameters, including explosive density, detonation velocity, dynamic Young’s modulus, dynamic
Reliability problem Poisson’s ratio, uniaxial compressive strength, and borehole radius, the Monte-Carlo sampling method
Probabilistic prediction was adopted to calculate the exceeding probability of the crushed zone radius from desired values.
The results showed that the exceedance probability for the growth of cracks falls sharply by the increase
in the crushed zone radius so that the probability of crushed zone radius longer than 0.5 m is less than
one percent. The results of this study, compared to the deterministic models, provide advantages in that
they are not only limited to a certain value for the crushed zone radius and show the probability of excee-
dance for any desired radius.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Due to the growth of fine cracks that subsequently join

Explosion is one of the conventional methods in tunnel and
mine excavations in rock environments. The main concern with
this method is the unwanted induced failures in surrounding rock
media that create unanticipated adverse effects, including instabil-
ity problems, the flow of water in joints, reducing the lifetime of
the structure, etc. [1]. More precisely, after the explosion, the
resulting pressure waves are released quickly and extremely
vibrate the rock environment. The resulting vibration, which
occurs very fast in a fraction of a second, will stimulate the
mechanical and dynamic properties of the rock mass. This stimula-
tion causes tensions and stresses in existing discontinuities, and
depending on the toughness of rock, would start opening and prop-
agation of the joint.
⇑ Corresponding author.
E-mail addresses: mahdishadabfar@yahoo.com (M. Shadab Far), wangyuan@
hhu.edu.cn (Y. Wang).
1
Ph.D. Student in Bridge and Tunnel Engineering, School of Civil and Transporta-
tion Engineering, Hohai University.
2
Professor of Geotechnical Engineering, School of Civil and Transportation
Engineering, Hohai University.
together, the range of discontinuities will increase, and some
crushed zones that are unstable will appear. On the other hand,
some of the cracks that receive greater impact from the explosion
shock, go beyond the crushed zone and radially penetrate through
the surrounding environment. Beyond the crushed zone and radial
cracks, the explosion effect is seen in the form of ground vibration.
These three parts are shown in Fig. 1.
Using the blast load in civil engineering projects has a long his-
tory, and consequently, the vibrations caused by the explosion
have been studied by various researchers [2–4]. However, investi-
gation of the different models for predicting the growth of the
crushed zone under the explosion in rock started some decades
ago [5,1]. Ash [6,7] was among the first researchers who tried to
predict the crushed zone around the blast-hole, based on the
experimental models. In his model, using the relative density of
the rock mass and the relative strength of explosives, the radius
of the affected area was calculated as a multiple of the blast-hole
radius. After Ash, Drukovanyi et al. [8], assuming a homogeneous
and elastic environment, studied the stress field characteristics
and crushing intensity of the rock under the one-dimensional
stress wave caused by the explosion. Szuladzinski [9], based on
the hydrodynamic theory in rock, suggested a relationship to

http://dx.doi.org/10.1016/j.compgeo.2016.08.025
0266-352X/Ó 2016 Elsevier Ltd. All rights reserved.
 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300
Fig. 1. Crack formation around the blast hole.
predict the crushed zone radius around the blast-hole utilizing the
efficient energy of the explosive material. Esen et al. [10], using
several field-test explosions, developed a parameter called the
Crushing Zone Index (CZI) and suggested an empirical relation
for predicting the crushed zone radius. Iverson et al. [11] analyti-
cally calculated the triple areas around the blast-hole (Fig. 1) and
compared the results with the field-test explosions on concrete
samples. Hustrulid and Johnson [12] improved Ash’s model and
calculated the crushed zone radius based on the amount of explo-
sive energy and exerted stress on the blast-hole wall.
Apart from the research listed above, there are several other
research works done by analytical methods [13–15], software
modeling [16–18], and laboratory tests [19,20] available in the
related literature. However, as to the knowledge of the authors,
few if any studies have investigated the probability of the crushed
zone propagation under the explosion load. In fact, none of the
explosive load and bearing capacity of the rock environment are
fully known, and thus, they cannot be accounted as deterministic
variables. It is, therefore, necessary to develop a method, which
while considering the involved uncertainties, is able to examine
the problem probabilistically. In this regard, relying on previous
works done, and considering the involved parameters as random
variables, this paper tried to study the exceedance probability of
the crushed zone radius from any possible value.
2. Modeling
As explained in the introduction, this article focuses on the
probability analysis of explosion and crushed zone growth in rock.
For this purpose, we first need to set a deterministic model in
which we can express the crushed zone radius by a closed-form
relation. Then, defining the involved parameters as random vari-
ables, the problem can be transferred from a deterministic to a
probabilistic state, and it can be established as a reliability prob-
lem. Then, using one of the available approaches for reliability
problems, such as the Monte-Carlo sampling method that is
adopted in this paper, the established problem can be solved,
and the answer can be depicted in the form of exceeding probabil-
ity versus the given radius. The above process is shown schemati-
cally in Fig. 2.
The Esen’s model is chosen to be the primary deterministic
model. In this model, the crushed zone radius is expressed as a
291
Fig. 2. Probabilistic modeling process.
function of the rock mass and explosive characteristics. This
method is explained in the next section.
2.1. Esen’s model
Esen et al. [10] used concrete and a combination of other syn-
thetic materials to make 92 samples and study the mechanisms
of rock breakage by explosives. According to the crushing process
around the explosion point, they defined a Crushing Zone Index
(CZI) as follows (Eq. (1)):
P3b
CZI ¼ ; ð1Þ
K  r2c
where P b is the blast-hole pressure (Pa), K is the stiffness of rock
mass (Pa), and rc is the uniaxial compressive strength of rock
(Pa). The amount of Pb and K could be calculated from Eqs. (2)
and (3):
q0 D2CJ
PCJ ¼ ; ð2aÞ
4
P CJ
Pb ¼ ; ð2bÞ
2
Ed
K¼ ; ð3Þ
1 þ md
where P CJ is the ideal blast pressure (Pa), q0 is the unexploded
explosive density (kg=m3 ), DCJ is the ideal detonation velocity (m/
s), Ed is the dynamic Young’s modulus of rock (Pa), and md is the
dynamic Poisson’s ratio of rock. These relations are to approximate
the blast-hole pressure and stiffness. In cases that more accurate
values are available through direct measurements or numerical
modelings, they could be used instead of the provided relations.
After calculating the CZI, Esen et al. [10] demonstrated that
there is a power relationship between this factor and the radius
of the crushed zone. This relation is shown in Eq. (4):
r0
¼ 1:23  CZI0:219 ; ð4Þ
rc
where CZI is the crushing zone index, r c is the crushed zone radius,
and r 0 is the blast-hole radius. The crushed zone radius is then cal-
culated as follows (Eq. (5)):
292 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300
r c ¼ 0:812  r 0  CZI0:219 : ð5Þ
Although the rock samples were not directly used in Esen’s
model, the combination of concrete and other additives were used
so as to stimulate the rock blasting process [10]. Using cement-
based materials such as concrete samples to study the rock behav-
ior has been pointed out by other studies as well [21,22]. It should
also be noted that the existing discontinuities and fines around the
blast-hole have not been directly considered in Esen’s proposed
relations. In fact, it is assumed that the material within the crush-
ing zone is homogeneous and isotropic.
Esen et al. have not provided any scale factor for their model
application in real-scale projects. Instead, they compared their
models with single-blast full-scale projects and validated their pro-
posed relations by different rocks and explosive materials. They
mentioned that the full-scale projects acknowledged the general
trend of their proposed relation and had a good consistency with
its results. However, their model has a more accurate estimation
for high energetic explosive charges such as Emulsion than lower
energetic explosives such as ANFO. It is also mentioned that to gain
an accurate estimate, the stiffness of rock medium should be mea-
sured accurately because it could highly affect the proposed
crushed zone radius.
2.2. Expressing the deterministic model as a reliability problem
As already discussed, this paper tries to turn the development of
the crushed zone under the explosion load into a probabilistic
problem because deterministic methods are limited to a certain
answer, even for a problem which basically has a great deal of
uncertainties. In our study, this means that the deterministic mod-
els assume one hundred percent probability of having a specific
radius of crushed zone in their prediction, so it makes them ineffi-
cient and sufficiently unreliable. A more rational approach in this
regard is to calculate the probability of exceedance for any desired
radius. Indeed, it is correct to state that any expected value for the
radius of crushed zone corresponds to the probability of
exceedance.
For this purpose, it is assumed that the impact of the explosion
on the rock environment is Q, and the load-bearing capacity of the
rock mass against the explosion is R. Then, the Limit State Function
(LSF) can be defined as shown in Eq. (6). If the exerted load exceeds
the capacity of rock (i.e., R is less than Q), the model faces failure,
and if R is greater than Q, the model remains safe.
LSF ¼ R  Q : ð6Þ
When the rock is very resistant against explosion, the crushed
zone would be limited to a small area; however, when the rock
is weak against explosion the crushed zone would propagate in a
wider area. Therefore, the real crushed zone radius could be a good
representative to express the load-bearing capacity of rock. Thus,
we assume that Q is the radius of the crushed zone around the
blast-hole, which can be obtained from the Esen’s model, and R
is a certain radius of the crushed zone, for instance, 400 mm,3
which, in fact, is the damage border for the model. We try to calcu-
late how probable the crushed zone radius is to exceed this certain
amount of R. Then, the LSF can be established as follows:
LSF ¼ 400  0:812  r0  CZI 0:219
: ð7Þ
Our goal, as shown in Eq. (8), is to calculate the failure probabil-
ity or the exceedance probability of Q with respect to R:
PðfailureÞ ¼ PðLSF < 0Þ ¼ PðQ > RÞ: ð8Þ
3
The other possible radius will be considered later.
Many attempts have been made to calculate this probability
[23–25]. The Monte-Carlo sampling method, however, is used in
this paper [26–28]. In the next section, the use of the Monte-
Carlo method in our case study is explained.
2.3. Monte-Carlo sampling method to solve the reliability problem
To use the Monte-Carlo method, we first need to establish the
LSF according to the involved random variables. Then, for each ran-
dom variable, n random numbers are generated. Generated ran-
dom numbers should follow their probability distribution
functions, with their specific means and standard deviations. The
random numbers are then substituted in LSF so that LSF is available
for each random number. Thus, the result is n artificial values for
LSF. Then, using Eq. (9), the probability of LSF < 0 is calculated:
n
P¼ ; ð9Þ
N
where n is the number of samples less than zero, and N is the total
number of samples.
To ensure sufficient accuracy of the computed probabilities, one
possible way is to increase the number of samplings and recalcu-
late the exceedance probability until it is assured that the probabil-
ity is not sensitive to the number of samples, and it is proved that
the number of samples used to calculate the probability is enough.
Then, the probability of the crushed zone exceeding the desired
radius is calculated. The Monte-Carlo algorithm to calculate the
exceedance probability is shown in Fig. 3. In the next section, using
this algorithm, the under-studied probability is calculated, and the
results are discussed.
3. Model analysis and results
First, the input parameters, including explosive density, detona-
tion velocity, dynamic Young’s modulus, dynamic Poisson’s ratio,
uniaxial compressive strength, and borehole radius should all be
specified as random variables. Assuming theses parameters as ran-
dom variables enable us to consider the uncertainties and also
cover a quite wide range of possible load cases. It is assumed that
these variables follow the standard normal distribution for sim-
plicity. Their assumed characteristics are presented in Table 1 [26].
Using the assumed probability distribution function, the artifi-
cial values for P b could now be calculated. For this purpose, two
twenty-number groups of uniformly distributed random numbers
were generated that are shown in columns two (u1i ) and five (u2i )
of Table 2. Then, using the inverse CDF of standard normal distribu-
tion (Eq. (10)), the standard normal random variables were calcu-
lated corresponding to each generated random number. The results
are shown in columns three (z1i ) and six (z2i ) of Table 2.
zi ¼ U1 ðui Þ: ð10Þ
Next, with respect to the means and standard deviations listed
in Table 1, and using the Eq. (11), the generated standard normal
random values were turned into the normal variables as shown
in the fourth (q) and seventh (DCJ ) columns of Table 2:
R ¼ lR þ rR zi ; ð11Þ
where R is the assumed random variable, and lR and rR are the
means and standard deviations of R, respectively. Using the avail-
able random values, P b was then calculated based on Eqs. (2a)
and (2b). The results are presented in column eight of Table 2.
Using the described method, the random numbers correspond-
ing to K and rc could be calculated as well. The results for K and rc
are shown in Tables 3 and 4, respectively.
Having random values for all involved parameters, Eq. (1) was
adopted to generate the CZI. The results are given in the fifth col-
 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300 293

Fig. 3. Monte-Carlo algorithm for reliability analysis.

Table 1 umn of Table 5. Then, by generating random values for r0 and using
Random variables for Esen’s model. Eq. (5), the crushed zone radius was calculated for each random
Mean Standard deviation case and presented in the sixth column of Table 5. Having the
q ðg=cm3 Þ 0.95 0.2 LSF established (Eq. (7)), random numbers corresponding to the
DCJ ðm=sÞ 5000 750 LSF were finally generated, as shown in the seventh column of
E ðPaÞ 70  109 20  109 Table 5.
m 0.25 0.05 As can be seen, there is only one LSF less than zero in the last
rc (pa) 80  106 30  106 column of Table 5. Thus, the probability corresponding to LSF < 0
r 0 (mm) 80 30 could be calculated as follows (Eq. (12)):

Table 2
Generation of random values for Pb .

u1i z1i q ðkg=m3 Þ u2i z2i DCJ (m/s) P b (Pa)

1 0.81472 0.89544 1129.09 0.75127 0.678482 5508.86 4283131993

2 0.90579 1.31528 1213.06 0.25510 0.65854 4506.09 3078869389
3 0.12699 1.14075 721.85 0.50596 0.01493 5011.20 2265897505
4 0.91338 1.36184 1222.37 0.69908 0.52175 5391.31 4441203219
5 0.63236 0.33811 1017.62 0.89090 1.23135 5923.51 4463284170
6 0.09754 1.29569 690.86 0.95929 1.74252 6306.89 3435037551
7 0.27850 0.58731 832.54 0.54722 0.11863 5088.97 2695096726
8 0.54688 0.11779 973.56 0.13862 1.08652 4185.11 2131499478
9 0.95751 1.72246 1294.49 0.14929 1.03947 4220.40 2882150526
10 0.96489 1.81047 1312.09 0.25751 0.65105 4511.72 3338553285
11 0.15761 1.00432 749.14 0.84072 0.99741 5748.06 3093950247
12 0.97059 1.88958 1327.92 0.25428 0.66107 4504.19 3367554537
13 0.95717 1.71872 1293.74 0.81428 0.89380 5670.35 5199690748
14 0.48538 0.03667 942.67 0.24352 0.69501 4478.74 2363636129
15 0.80028 0.84262 1118.52 0.92926 1.47033 6102.75 5207223911
16 0.14189 1.07188 735.62 0.34998 0.38536 4710.98 2040739165
17 0.42176 0.19739 910.52 0.19660 0.85385 4359.62 2163201156
18 0.91574 1.37695 1225.39 0.25108 0.67108 4496.69 3097202036
19 0.79221 0.81410 1112.82 0.61604 0.29511 5221.33 3792257534
20 0.95949 1.74483 1298.97 0.47329 0.06701 4949.75 3978078979
294 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300

Table 3
Generation of random values for K.

u3i z3i E (GPa) u4i z4i m K (Pa)

1 0.10665 1.24453 45.11 0.54701 0.11811 0.26 35917875123
2 0.96190 1.77315 105.46 0.29632 0.53501 0.22 86215453410
3 0.00463 2.60199 17.96 0.74469 0.65788 0.28 13999748795
4 0.77491 0.75512 85.10 0.18896 0.88175 0.21 70570910737
5 0.81730 0.90514 88.10 0.68678 0.48673 0.27 69136141946
6 0.86869 1.12024 92.40 0.18351 0.90207 0.20 76691088331
7 0.08444 1.37584 42.48 0.36848 0.33587 0.23 34449407050
8 0.39978 0.25391 64.92 0.62562 0.32027 0.27 51280498005
9 0.25987 0.64374 57.13 0.78023 0.77296 0.29 44329480655
10 0.80007 0.84187 86.84 0.08113 1.39754 0.18 73583273712
11 0.43141 0.17278 66.54 0.92939 1.47123 0.32 50276829449
12 0.91065 1.34475 96.90 0.77571 0.75779 0.29 75235538063
13 0.18185 0.90835 51.83 0.48679 0.03311 0.25 41521420378
14 0.26380 0.63166 57.37 0.43586 0.16148 0.24 46191718021
15 0.14554 1.05576 48.88 0.44678 0.13379 0.24 39318260789
16 0.13607 1.09815 48.04 0.30635 0.50622 0.22 39223772829
17 0.86929 1.12305 92.46 0.50851 0.02133 0.25 73905770052
18 0.57970 0.20114 74.02 0.51077 0.02700 0.25 59154309665
19 0.54986 0.12531 72.51 0.81763 0.90636 0.30 55975566704
20 0.14495 1.05832 48.83 0.79483 0.82330 0.29 37821347382

Table 4 n 1
Generation of random values for rc . P¼ ¼ ¼ 0:05: ð12Þ
N 20
u5i z5i rc (Pa) It means that there is five percent probability for the crushed
1 0.08552 1.36890 38933101.57 zone area to go beyond 400 mm after the explosion. However, this
2 0.26248 0.63571 60928654.54 result is not still sufficiently reliable, because the sample’s number
3 0.80101 0.84525 105357527.77
was so small that the results might be sensitive to the sample size.
4 0.02922 1.89238 23228641.98
5 0.92885 1.46731 124019302.60 Therefore, to check the effect of sample size, a code was developed,
6 0.73033 0.61381 98414418.66 and the above calculations were repeated for larger sample num-
7 0.48861 0.02856 79143291.52 bers. The results are shown in Table 6 and Fig. 4.
8 0.57853 0.19812 85943649.82 As seen, for a small sample size, there is a huge change in excee-
9 0.23728 0.71507 58547966.48
dance probability, so that we cannot reliably estimate the probabil-
10 0.45885 0.10333 76899971.09
11 0.96309 1.78771 133631278.83 ity. However, by increasing the sample size, the change in
12 0.54681 0.11760 83527850.16 exceedance probability got smaller, the analyses reached an
13 0.52114 0.05300 81590134.39 acceptable accuracy, and the results converged to almost 2.985%.
14 0.23159 0.73361 57991813.62
So far, the probability of the crushed zone exceeding
15 0.48890 0.02783 79165015.34
16 0.62406 0.31616 89484849.15
R ¼ 400 mm was obtained. However, it is only one of the possible
17 0.67914 0.46528 93958485.35 cases, and the exceedance probability for R less and greater than
18 0.39552 0.26497 72050816.61 400 mm should also be calculated. In fact, our target is to obtain
19 0.36744 0.33865 69840495.45 the exceedance probability for any desired radius, not only exceed-
20 0.98798 2.25655 147696602.63
ing a certain value. For this purpose, instead of assuming R as a
constant value in LSF (i.e., R ¼ 400 mm), we define it as a variable,

Table 5
Generation of random values for r 0 ; CZI, and r c .

u6i z6i r 0 (mm) CZI r c (mm) LSF

1 0.05962 1.55798 33.26 1443.23 132.86 267.14
2 0.68197 0.47322 94.20 91.19 205.50 194.50
3 0.04243 1.72315 28.31 74.86 59.14 340.86
4 0.07145 1.46511 36.05 2300.53 159.46 240.54
5 0.52165 0.05429 81.63 83.61 174.73 225.27
6 0.09673 1.30041 40.99 54.57 79.91 320.09
7 0.81815 0.90833 107.25 90.72 233.72 166.28
8 0.81755 0.90606 107.18 25.57 176.99 223.01
9 0.72244 0.59010 97.70 157.56 240.27 159.73
10 0.14987 1.03701 48.89 85.52 105.17 294.83
11 0.65961 0.41139 92.34 32.99 161.24 238.76
12 0.51859 0.04663 81.40 72.75 169.01 230.99
13 0.97297 1.92643 137.79 508.61 438.01 38.01
14 0.64899 0.38260 91.48 85.00 196.53 203.47
15 0.80033 0.84280 105.28 573.00 343.52 56.48
16 0.45380 0.11607 76.52 27.06 127.94 272.06
17 0.43239 0.17029 74.89 15.51 110.86 289.14
18 0.82531 0.93581 108.07 96.75 238.85 161.15
19 0.08347 1.38210 38.54 199.75 99.82 300.18
20 0.13317 1.11153 46.65 76.30 97.89 302.11
 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300 295

Table 6 Table 7
Exceedance probability for different sample sizes. Exceedance probability for different values of decision variables.

Sample numbers Less-than-zero numbers Exceedance probability Decision variable (mm) Less-than-zero numbers Eceedance probability
20 1 0.05 10 19877 0.99385
50 2 0.04 100 16263 0.81315
100 4 0.04 200 6807 0.34035
500 10 0.02 300 1939 0.09695
1000 25 0.0250 400 597 0.02985
10,000 275 0.0278 600 115 0.00575
20,000 597 0.02985 800 44 0.0022
1000 25 0.00125

r d , which can take different values of radii greater than zero (Eq.
(13)):
LSF ¼ r d  0:812  r 0  CZI0:219 ;
where r d is the decision variable, and could be any desired value of
the crushed zone radius. Thus, by changing the rd , the above calcu-
lations were repeated for the new LSFs, and their corresponding
exceedance probabilities were calculated. The results are presented
in Table 7.
By taking a comparatively finer increment than Table 7, the
exceedance probability was plotted against the decision variable,
and a curve was fitted to them. The result is shown in Fig. 5. This
graph shows the exceedance probability for any desired value of
crushed zone radius and known as exceedance risk curve [29].
As can be seen in Fig. 5, by increasing the crushed zone radius,
the exceedance probability falls sharply, so that for the radii
greater than 500 mm, the exceedance probability is less than 1 per-
cent, indicating that the crushed zone with a radius of more than
500 mm is nearly impossible.
The importance of this curve is that it enables the geotechnical
and blast engineers to know how much the crushed zone is prob-
able to go beyond a certain limit, and provides the means for them
to reconsider or change their design if needed. On the other hand,
this graph shows the capacity of the rock mass to withstand the
explosion wave. If the curve is comparatively high, the cracks have
a greater chance to propagate, and the probability of failure is con-
sequently high. Conversely, if the curve is relatively low, the cracks
could not easily propagate, and the rock environment has a greater
capacity to accommodate the explosion effects.
It should also be noted that these results were obtained after
ð13Þ considering the uncertainties related to the mechanical properties
of both the environment and the explosive material through ran-
dom variables, and they are not the results of a determined model
with fixed characteristics. Hence, the results are more reliable and
practical than the deterministic models.
4. Discussion
4.1. Sensitivity analysis
As shown in Eqs. (1)–(7), LSF is a function of blast-hole radius
(r0 ), blast-hole pressure (Pb ), the stiffness of rock mass (K), and uni-
axial compressive strength of rock (rc ). This relation is rewritten in
Eq. (14) as follows:
!
P 3b
LSF ¼ 400  0:812  r0  : ð14Þ
K  r2c
To investigate the effects of these parameters in risk curve
(Fig. 5), a parametric study is needed to evaluate how the involved
parameters could change the results. Thus, as shown in Fig. 6, by
changing the mean of one parameter and keeping the other param-
eters constant, the analyses were repeated, and the results were
reported as exceedance probability versus crushed zone radius in
four graphs, one graph for each parameter.

Fig. 4. The sample number versus exceedance probability.

296 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300

Fig. 5. Exceedance probability versus crushed zone radius.

Fig. 6. Sensitivity study of exceedance probability to LSF’s parameters.

As seen, by changing the mean of variables the whole range of certain range of crushed zone radius. The blast-hole radius mostly
crushed zone radius would be affected. The exceedance probability affects the small crushed zone radius (50–150 mm), while the
increases by increasing ro and Pb , while it decreases by increasing K highest effect of blast-hole pressure is in the medium range of
and rc . However, the highest effects of each parameter happen in a crushed zone radius (150–250 mm). The stiffness and uniaxial
 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300 297

compressive strength of rock also have a greater effect on the As seen in the graph, in the range of short crushed zone radius,
longer crushed zone radius (200–350 mm). less than 210 mm, the Weibull distribution function shows lower
exceedance probability than the normal distribution function. It
4.2. Probability distribution function means that the Weibull distribution function gives a higher esti-
mate of rock resistibility against explosion for short crushed zone
The other important point is to investigate how the probability radius. Later, for longer radiuses, more than 210 mm, Weibull dis-
distribution function could affect the results. This is because the tribution function shows higher results, indicating less resistibility
type of probability distribution function shows the uncertainty in and more chances for the crushed zone to propagate. For the
random variables. Therefore, if the random variables follow a prob- crushed zone radius equal to 500 mm, the Weibull distribution
ability distribution function rather than Normal distribution func- estimates the exceedance probability as 8.86%, while it is esti-
tion, the results may change accordingly. To further investigate this mated as 1.15% by the Normal distribution. The difference between
issue, the Weibull distribution function, which has an extensive the results of two distribution functions gets smaller so that for the
application for modeling the probability distribution of rock radiuses longer than 1 m the graphs get tangent to each other.
mechanical parameters, was chosen. The Weibull cumulative dis-
tribution function is as follows (Eq. (15)): 4.3. Comparison with existing models
ðkxÞj
F X ðxÞ ¼ 1  e for x P 0; ð15Þ To further investigate the described algorithm, we tried to apply
it to other scenarios and compare the results. For this purpose,
where x is the random variable, and k and j are two Weibull distri-
three different models available in the literature, including Szu-
bution’s parameters. The relation between the k and j with mean
ladzinski [9], Djordjevic [30], and Kanchibotla et al. [31] were cho-
(lX ) and standard deviation (rX ) of x are as follows (Eqs. (16a)
sen. It should be noted that all these models address the crushed
and (16b)):
zone around the blast hole in rock media. A descriptive comparison
 
1 between the assumption and modeling process of these models are
lX ¼ k C 1 þ ; ð16aÞ
j provided in Esen et al. [10]. The details of these models are shown
     in Table 9.
2 1
rX ¼ k2 C 1 þ  C2 1 þ ; ð16bÞ There are two parameters, Pb and Pd in Djordjevic’s and Kanchi-
j j
botla’s models, respectively, that are conceptually different. P d is
where C is the Gamma function. Considering the Weibull distribu- the ideal detonation pressure, while P b is the pressure excreted
tion function for random variables by the characteristics listed in on the initial wall of the blast-hole. The borehole pressure is con-
Table 8, the established problem was analyzed again, and the sidered as half of the detonation pressure [10]. To perform the
results were depicted in Fig. 7. analyses, the existing variables should be defined as random vari-
ables with specific probability distribution functions, means, and
standard deviations. For simplicity, it is assumed that the variables
Table 8 are normally distributed. Their means and standard deviations are
Characteristics of random variables assuming the Weibull distribution function.
shown in Tables 1 and 10.
lX rX k j The new scenarios were analyzed by the Monte-Carlo sampling
q ðg=cm Þ 3 o.95 0.2 1.0523 1.4992 method using 20,000 samples. As shown in Fig. 8, the exceedance
DCJ (m/s) 5000 750 5635.7511 1.9355 probabilities and crushed zone radii obtained from all four models
E (GPa) 70 20 78.9120 1.9226 were then depicted in a graph.
m 0.25 0.05 0.2574 1.0787 As can be seen, the result of Esen’s model is lower than the
rc (MPa) 80 30 90.1848 1.9207
other models. This indicates that the Esen’s model has a more opti-
r0 (mm) 80 30 90.1817 1.9207
mistic estimate of the load-bearing capacity of rock mass com-

Fig. 7. Comparison between the Normal and Weibull distribution functions.

298 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300

Table 9 Pi  PEsen
Different models for estimating the crushed zone under rock explosion. Relative Difference ¼ ; ð17Þ
PEsen
Model Relation Parameters
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where Pi is the exceedance probability of Szuladzinski’s, Djordje-
Szuladzinski 2r20 q0 Q ef r 0 (mm): borehole radius,
rc ¼ F 0c vic’s, or Kanchibotla’s models, and PEsen is the exceedance probabil-
[9]
q0 ðg=mm3 Þ: explosive density, ity of Esen’s model.
Q ef (Nmm/g): effective energy of the Fig. 9 shows that for the radii less than 0.3 m, the results are not
explosive. It is assumed to be two third of the much different, and in fact, all models could predict the small
complete reaction heat (Nmm/g) [7],
cracks with the same accuracy. However, the maximum difference
F 0c (MPa): confined dynamic compressive
strength of the rock mass. It is assumed to be
in all three methods happened within the average radii, i.e.,
approximately eight times of unconfined between 0.3 and 1 m, which is the most likely range of failure in
static compressive strength, rc [7]. reality. The maximum differences between Szuladzinski and
Djordjevic r c ¼ pffiffiffiffiffiffiffiffiffiffiffi
r0 ffi r 0 (mm): radius of the blast hole, Djordjevic’s models compared to Esen’s model are roughly about
24T=Pb
[30]
T (Pa): tensile strength of the rock material,
5.74 and 29.5, respectively, while this value for Kanchibotla’s
P b (Pa): borehole pressure. model is around 169.55. This high difference shows that the results
qffiffiffiffi of Kanchibotla’s model, especially for the average range of radii, are
Kanchibotla rc ¼ r0 Pd r 0 (mm): borehole radius,
rc far from reality. For large radii, differences between Szuladzinski
et al. [31]
P d (Pa): detonation pressure,
and Djordjevic’s models compared to Esen’s model remain con-
rc (Pa): unconfined compressive strength of
the rock.
stant; however, the difference corresponding to Kanchibotla’s
model is greatly reduced. Finally, within the range of very large
radii, which are of course, not very probable, the results of all three
Table 10
models get close together again.
Required random variables in Szuladzinski’s, Djordjevic’s, and Kanchibotla’s models.

Parameter Q ef (Nmm/g) F 0c (Mpa) T (Pa) 4.4. Comparison with real-scale projects

Mean 4,000,000 640
Standard deviation 500,000 240
pared to the other three models. Moreover, Kanchibotla’s model is
much different from other models and offers a very high failure
probability of rock medium. This issue has also been pointed out
by Kanchibotla et al. [31] and Kojovic et al. [32]. Because the aim
of Kanchibotla’s model is to study the mine fragmentation and
optimize the mine excavation, it may not present very accurate
results for estimating crack propagation and crushed zone size.
To further investigate how far the results of the Kanchibotla model
are different from the other models, we used Eq. (17) to calculate
the relative difference between the results of the described meth-
ods. Then, we drew the relative differences together in Fig. 9. As it
is clearly seen, there is a sharp difference observed in Kanchibotla’s
model compared to other models.
5,000,000
2,000,000 To use the results of proposed risk curve and calculate the fail-
ure probability of real-scale projects, two single-blast full-scale
projects were selected from the literature. The characteristics of
these two projects are shown in Table 11.
Using the Esen’s model, the crushed zone radius was calculated
for each case as shown in the fifth column of Table 12. The failure
probability or probability of exceeding from the estimated crushed
zone radius was, then, extracted from the risk curve as 1.9%
and 0.23% for case 1 and case 2, respectively. The actual crushed
zone radiuses were, next, reported in the seventh column of
Table 12.
As seen in Table 12, the exceedance probability in both cases is
small values that show the low chance of crushed zone radius to
exceed the Esen’s prediction. However, if the target risk level for
these projects were available, we could argue if the explosion
design matches the expected risk level.

Fig. 8. Exceedance probability against crushed zone radius for different models.
 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300 299

Fig. 9. Relative differences between different models.

Table 11
Characteristics of single blast full-scale projects.

No. Source q ðg=cm3 Þ DCJ (m/s) E ðGPaÞ m rc (MPa) r 0 (mm)

Case 1 Amnieh and Bahadori [33] 1.2 6437 12 0.2 11 38

Case 2 Slaughter [34] 1.2 5364 8.9775 0.35 20 80

Table 12
Exceedance probability of crushed zone radius for real-scale projects.

No. Esen’s model Exceedance probability (%) r actual

P b (GPa) K (GPa) CZI r c ðmmÞ
Case 1 6.22 10 198421.8 466.1817 1.90 250
Case 2 6.88 6.65 122322.1 844.9083 0.23 760

5. Conclusion ranges from small- to medium-sized cracks less than 0.5 m. It is

The main goal of this paper was to explain that since the exact
behavior of rock media and explosive materials are not fully under-
stood, the issue of using a deterministic approach to study the
environmental response under explosion, especially the growth
of the crushed zone, may not be sufficiently reasonable. On the
other hand, a deterministic estimation of the crack sizes around
the blast-hole is identical to assigning one hundred percent prob-
ability to a certain radius value, which is an issue that could be crit-
icized. Thus, the rock blast issue was defined as a reliability
problem in this paper. The Monte-Carlo sampling method was then
adopted to calculate the probability of a crushed zone radius
exceeding 400 mm. By increasing the number of samples, it was
assured that the calculated exceedance probability was not sensi-
tive to the sample size. The analysis was then repeated for other
values of radii, and a curve was fitted to the results so that for
any desired crushed zone radius, an exceedance probability was
obtained. This graph, which represents the failure risk for the
model, shows how probable the crushed zone is to exceed a certain
radius. Finally, the studied model was compared with three other
models, and their results were discussed.
This study showed that by increasing the crushed zone radius,
the exceedance probability drops sharply so that the most proba-
ble size of craushed zone, which might happen after the explosion,
also showed that the capacity of the rock mass to accommodate
the explosion effects has a meaningful relationship with excee-
dance probability. When the probability curve gets higher, there
would be more chances for cracks to propagate, which indicates
the low load-bearing capacity of the rock mass. Conversely, when
the probability curve falls lower, the probability associated with
growing cracks gets lower, and it shows that the rock mass is more
resistant against explosion.
The following conclusions can be drawn from the present
study:
 The probability of the crushed zone radius exceeding 0.5 m was
seen to be less than one percent using Esen’s model. It indicates
that crushed zone over 0.5 m size possibly may not occur. This
issue was also pointed out in Esen et al. [10] and Hustruli [35]
that the ratio between the crushed zone radius and the borehole
radius (r c =ro ) for ordinary cases is usually between 3 and 5. If
we think of borehole radius as 10 cm, the crushed zone radius
may not exceed 0.5 m. However, it does not mean there is no
chance for the crushed zone to go beyond 0.5 m; this is a simple
estimate for general cases.
 The sensitivity analysis showed that the blast hole radius
mostly affects the short crushed zone radiuses (50–150 mm),
while the blast-hole pressure has more effects on medium
300 M. Shadab Far, Y. Wang / Computers and Geotechnics 80 (2016) 290–300
ranges of radiuses (150–250 mm), and the stiffness and uniaxial
compressive strength of rock have a greater effect on the longer
crushed zone radiuses (200–350 mm).
 The four studied models showed that their prediction for short
and long cracks are not very different; however, the maximum
difference is within the range of 0.3–1 m, which is the medium
range of crack length.
 Esen’s model presents a more optimistic estimate of rock capac-
ity against the explosion load, and consequently, gives lower
failure probability.
 The results of Kanchibotla’s model seemed to be very different
from the other models, especially for the cracks between 0.3
and 1 m sizes. Thus, using Kanchibotla’s model for the average
range of cracks is not recommended.
Further research might investigate the closed-form solution for
the probabilistic estimation of the crushed zone. Because apart
from the random sampling methods, the established reliability
problem could also be solved using other reliability methods such
as importance sampling. This topic will be studied by the authors
in the future.
References
[1] Zhang Z-X. Rock fracture and blasting: theory and applications. 1st ed. Elsevier
Science; 2016.
[2] Lee FW, Thoenen JR, Windes SL. Earth vibrations caused by quarry blasting.
Report of investigations. Department of the Interior, United States Bureau of
Mines; 1936.
[3] Jimeno EL, Jimino CL, Carcedo A. Drilling and blasting of rocks. CRC Press,
Taylor and Francis; 1995.
[4] Zhao J, Li J, editors. Rock dynamics and applications - state of the art.
Proceedings of the first international conference on rock dynamics and
applications. Taylor and Francis Group, CRC Press; 2013.
[5] Mohanty B, editor. Rock fragmentation by blasting. Proceedings of the fifth
international symposium on rock fragmentation by blasting. Montreal
(Quebec, Canada): Taylor and Francis Group, CRC Press; 1996.
[6] Ash RL. The mechanics of the rock breakage (part 1). Pit Quarry
1963;56:98–100.
[7] Ash RL. The mechanics of rock breakage (part 2), standards for blasting design.
Pit Quarry 1963;56:118–22.
[8] Drukovanyi MF, Komir VM, Myachina NI, Rodak SN, Semenyuk EA. Effect of the
charge diameter and type of explosive on the size of the overcrushing zone
during an explosion. J Min Sci 1973;9:500–6.
[9] Szuladzinski G. Response of rock medium to explosive borehole pressure. In:
Proceedings of the 4th international symposium on rock fragmentation by
blasting (FRAGBLAST-4), Vienna, Austria. p. 17–23.
[10] Esen S, Onederra I, Bilgin HA. Modelling the size of the crushed zone around a
blasthole. Int J Rock Mech Min Sci 2003;40:485–95.
[11] Iverson SR, Hustrulid WA, Johnson JC, Akbarzadeh Y. The extent of blast
damage from a fully coupled explosive charge. In: Proceedings of the 9th
international symposium on rock fragmentation by blasting. FRAGBLAST-
9. CRC Press, Taylor and Francis; 2010. p. 459–68.
View publication stats
[12] Hustrulid W, Johnson J. A gas pressure-based drift round blast design
methodology. In: 5th international conference and exhibition on mass
mining Sweden, Sweden, Lulea, Lulea University of Technology. p. 657–69.
[13] Bahadori M, Amnieh HB, Khajezadeh A. A new geometrical-statistical
algorithm for predicting two-dimensional distribution of rock fragments
caused by blasting. Int J Rock Mech Min Sci 2016;86:55–64.
[14] Yilmaza O, Unlub T. An application of the modified holmbergpersson approach
for tunnel blasting design. Tunn Undergr Space Technol 2014;43:113–22.
[15] Trivedi R, Singh T, Raina A. Prediction of blast-induced flyrock in indian
limestone mines using neural networks. J Rock Mech Geotech Eng
2014;6:447–54.
[16] Yilmaz O, Unlu T. Three dimensional numerical rock damage analysis under
blasting load. Tunn Undergr Space Technol 2013;38:266–78.
[17] Wang Z, Konietzky H, Shen R. Coupled finite element and discrete element
method for underground blast in faulted rock masses. Soil Dyn Earthquake Eng
2009;29:939–45.
[18] Wang Z, Konietzky H. Modelling of blast-induced fractures in jointed rock
masses. Eng Fract Mech 2009;76:1945–55.
[19] Chen W, Ma H, Shen Z, Wang D. Experiment research on the rock blasting
effect with radial jet cracker. Tunn Undergr Space Technol 2015;49:249–52.
[20] Sun C. Damage zone prediction for rock blasting. PhD thesis. The University of
Utah; 2013.
[21] Dick R, Fourney W, Wang X, Young C. Results from instrumentedsmall scale
tests. In: Proceedings of the fourth international symposium on rock
fragmentation by blasting-fragblast-4, Vienna, Austria. p. 47–54.
[22] Stimpson B. Modelling materials for engineering rock mechanics. Int J Rock
Mech Min Sci Geomech Abstr 1970;7:77–121.
[23] Rackwitz R. Reliability analysisa review and some perspectives. Struct Safety
2001;23:365–95.
[24] Rebba R, Mahadevan S. Computational methods for model reliability
assessment. Reliab Eng Syst Safety 2008;93:1197–207.
[25] Madsen HO, Krenk S, Lind NC. Methods of structural safety. New York (United
States): Dover Publications Inc.; 2006.
[26] Far MS, Wang Y. Approximation of the monte carlo sampling method for
reliability analysis of structures. Math Probl Eng 2016;2016.
[27] Kroese DP, Taimre T, Botev ZI. Handbook of Monte Carlo methods. John Wiley
and Sons; 2011.
[28] Padmanabhan D, Agarwal H, Renaud JE, Batill SM. A study using monte carlo
simulation for failure probability calculation in reliability-based optimization.
Optim Eng 2006;7:297–316.
[29] Oberkampf WL, Helton JC. Engineering design reliability. Taylor and
Francis: CRC Press; 2004.
[30] Djordjevic N. Two-component of blast fragmentation. In: Proceedings of the
sixth international symposium on rock fragmentation by blasting-fragblast
1999, South African Institute of Mining and Metallurgy, Johannesburg, South
Africa. p. 213–9.
[31] Kanchibotla SS, Valery W, Morrell S. Modelling fines in blast fragmentation
and its impact on crushing and grinding. In: Proceedings of Explo99 a
conference on rock breaking, volume 99 of 5. The Australasian Institute of
Mining and Metallurg y, Kalgoorlie, Australia. p. 137–144.
[32] Kojovic T, Kanchibotla SS, Poetschka NL, Chapman J. The effect of blast design
on the lump: fines ratio at marandoo iron ore operations. In: Proceedings of
the mine to mill conference, The Australasian Institute of Mining and
Metallurgy, Brisbane, Australia. p. 149–52.
[33] Amnieh HB, Bahadori M. Numerical analysis for effects of single blast hole in
mudstone rock-mass at Gotvand Olya dam. Iran J Geophys 2012;6:56–72 [in
Persian].
[34] Slaughter S. Investigation of coal fines. Technical report. Julius Kruttschnitt
Mineral Research Centre (JKMRC), Australia; 1991.
[35] Hustrulid WA. Blasting principles for open pit mining. CRC Press, Taylor and
Francis; 2005.