You are on page 1of 19

Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Contents lists available at ScienceDirect

Journal of Rock Mechanics and


Geotechnical Engineering
journal homepage: www.jrmge.cn

Review

Application of artificial intelligence to rock mechanics: An overview


Abiodun Ismail Lawal a, b, Sangki Kwon a, *
a
Department of Energy Resources Engineering, Inha University, Yong-Hyun Dong, Nam Ku, Incheon, South Korea
b
Department of Mining Engineering, Federal University of Technology, Akure, Nigeria

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

Article history: Different artificial intelligence (AI) methods have been applied to various aspects of rock mechanics, but
Received 29 January 2020 the fact that none of these methods have been used as a standard implies that doubt as to their generality
Received in revised form and validity still exists. For this, a literature review of application of AI to the field of rock mechanics is
28 March 2020
presented. Comprehensive studies of the researches published in the top journals relative to the fields of
Accepted 13 May 2020
Available online 13 November 2020
rock mechanics, computer applications in engineering, and the textbooks were conducted. The perfor-
mances of the AI methods that have been used in rock mechanics applications were evaluated. The
literature review shows that AI methods have successfully been used to solve various problems in the
Keywords:
Artificial intelligence (AI)
rock mechanics field and they performed better than the traditional empirical, mathematical or statis-
Rock mechanics tical methods. However, their practical applicability is still an issue of concern as many of the existing AI
Literature review models require some level of expertise before they can be used, because they are not in the form of
Statistical method tractable mathematical equations. Thus some advanced AI methods are still yet to be explored. The
limited availability of dataset for the AI simulations is also identified as a major problem. The solutions to
the identified problems and the possible future research focus were proposed in the study subsequently.
Ó 2021 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by
Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction equipment in and around such an excavation. Hence, the stability of


surface and underground excavations has always been a great
The quest for construction of heavy structures, underground concern to geotechnical engineers (Idris et al., 2016). The stability of
space, resources and security over the recent decades is an inevi- underground excavations, according to Idris et al. (2013), depends
table demand and trend of the modernization and development of on the physico-mechanical properties of the rock mass, the
the countries around the world. High-quality development and discontinuity parameters, the groundwater condition, and the
sustainable utilization of underground space have become an in- induced stresses. These parameters are generally determined using
ternational consensus. The utilization of the underground space experimental and in situ measurement techniques. However, it is
can basically be classified into two purposes, viz. civil engineering difficult to accurately estimate/predict the stability of underground
application and mining engineering application (Sterling and excavation owing to the complexity of the physical system and the
Nelson, 1982; Wallace and Ng, 2016). However, the most chal- efforts required in determining the input data associated with
lenging issue is the safety of the life of the workers, the users of the geotechnical parameters (Sakellariou and Ferentinou, 2005; Idris
facility located on the surface and the underground facility. One of et al., 2016). The majority of the engineering construction works
the key ways of achieving the safety of life and equipment in the and designs are carried out based on the experiences of the field
surface and underground excavations is to ensure its stability, engineers, but insufficient information may be provided concern-
because excavations can lead to disturbance and redistribution of ing the interaction between the host rocks and the structures.
the in situ stress field of the host rock (Mahdevari et al., 2017). It is acknowledged that many rock mechanics problems are due
Based on the degree of disturbance, surface and underground to the availability of limited data, resulting in insufficient infor-
excavations can pose a major threat to the safety of personnel and mation on the behavior of the subsurface. This scenario can be
described using the Holling (1978)’s classification as presented in
Fig. 1. In Fig. 1, there are four regions. The first region indicates
* Corresponding author. where there are good data but limited understanding. In region 2,
E-mail address: kwonsk@inha.ac.kr (S. Kwon). there are limited data and good understanding; while in region 3,
Peer review under responsibility of Institute of Rock and Soil Mechanics, Chi-
there are enough data and good understanding. In the fourth
nese Academy of Sciences.

https://doi.org/10.1016/j.jrmge.2020.05.010
1674-7755 Ó 2021 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 249

The AI method has recently gained ground in various fields of


sciences and engineering owing to its ability to handle complex
problems such as those encountered in rock mechanics. In AI, the
pattern of operation of the human brain, the social interaction
between the animals, formation, and reproduction of genes in the
human being are been emulated. Hence, the approach possesses
the ability to model complex problems utilizing a series of inter-
connectivity between the input and output parameters. AI has been
used in the rock mechanics field, in the area of structural rock
mechanics and rock comminution according to the classification of
the Rock Mechanics Committee of the American National Academy
of Sciences in 1963 (Cook et al., 1966). Hence, in this study, some of
the researches that have been conducted within the last decade on
the application of AI in the area of rock mechanics are reviewed.
The purpose of this study is to establish which of the AI approaches
is most suitable and commonly used in the field of rock mechanics,
and to identify the gap in the knowledge of rock mechanics that
have not been filled by the existing literature.

Fig. 1. Modified Holling (1978)’s classification of modeling problems. 2. Overview of artificial intelligence (AI)

region, there are neither enough data nor good understanding. AI is an aspect of computer science which deals with the
Therefore, the majority of the rock mechanics problems belong to simulation of human intelligence with the aid of a computer. It
regions 2e4 problems of Holling (1978)’s classification according to involves the design and building of machines that behave intelli-
Starfield and Cundall (1988), while other engineering mechanics gently. These machines solve new problems, for example, visual
problems falls within the regions 1 and 3. However, the major images manipulation and processing, symbols by algorithm and
problem is that the majority of the rock mechanics problems are heuristics from the previous learning and knowledge from expe-
modeled following the region 3 procedure, which is obviously rience. Although AI is basically related to computer science, it is also
inappropriate (Starfield and Cundall, 1988; Feng et al., 1997). In related to various fields of sciences such as mathematics, cognition,
addition, rock mechanics problems are not only limited to the un- philosophy and psychology, and biology, and it has been recently
availability of well-defined input data in modeling but also the incorporated into the field of engineering. According to Cawsey and
validation of such models and the capability to account for the Aylett (2009) and Carbonell (2003), tasks such as face recognition,
inherent variability in geomaterial (Starfield and Cundall, 1988). voice recognition, navigation of heavily crowded areas and lan-
The analytical equations for some rock mechanics problems are guage interpretation that are extremely difficult to automate are
not available, and most of the available models cannot accommo- resolved via AI. There are various branches of AI as shown in Fig. 2.
date all the influence parameters and therefore simplified as- Among the AI approaches, the most commonly used by the re-
sumptions (Feng et al., 1997) are generally used. When the design searchers in rock engineering and geotechnics is the artificial
engineers are supplied with incomprehensive information, their neural network (ANN). ANN establishes a reasonable relationship
design or judgment using prior experiences on similar geological between the input geotechnical and geometrical parameters (such
terrain would be limited. The advent of numerical modeling has as the relationship between uniaxial compressive strength (UCS)
abated the rock mechanics problems as it can simulate the me- and density, Schmidt rebound hardness, P- and S-wave velocities,
chanical behavior of the rock engineering structures, but there is a etc.) and the desired output based on the learning from the trained
huge disparity between the in situ measurements and the nu- data. The overviews of some of the AI approaches commonly used
merical predictions (Xue and Xiao, 2017). The disparity could be in the field of rock mechanics and rock engineering are presented in
attributed to the difficulty in defining suitable boundary conditions Fig. 2.
due to the lack of a comprehensive understanding of the rock
mechanics problems. 2.1. Artificial neural network (ANN)
Considering the inherent limitations in the traditional analytical
approaches in solving rock mechanics problems and the prediction ANN mimics the human brain in the way of taking input infor-
using the numerical methods, the use of data mining techniques mation, processing and interpretation of the information. The ANN
becomes indispensable. Many data mining methods, such as sta- method developed in the 1980s is a relatively new branch of AI. In
tistical methods, classification and regression tree (CART), symbolic recent years, ANN is considered as one of the intelligent tools that
learning, case-based learning, and artificial intelligence (AI), have can predict complex problems. It can generalize a solution from the
been used to obtain useful knowledge from the organization and pattern presented to it during training. The basic component of the
analysis of raw data in civil engineering (Arciszewski and Ziarko, ANN is characterized by its processing unit, learning rules, and
1992; Michalski et al., 1998; Leu and Lo, 2004). However, from connection formula (Almeida, 2002). A neural network must be
the Holling (1978)’s classification, the data mining methods such as trained so that a known set of inputs produces the desired outputs.
statistical methods, CART, symbolic learning, and case-based Once the network is trained with enough sample dataset, for a new
learning may not be reliable for rock mechanics problems since input of relatively similar patterns, predictions can be made based
they are only suitable for regions 1 and 3. Hence, the most on previous learning (Khandelwal et al., 2004). ANN consists of a
commonly used data mining approach in rock engineering and series of interconnected nodes. The processing ability of the ANN is
geotechnical engineering is AI. This is because many AI methods are stored in the weights interconnecting the nodes. The ways in which
hypothesis-free methods that do not require prior knowledge or the nodes are being connected is known as ANN architecture. ANN
specific experts or rules, unlike many statistical methods and the architecture can basically be classified into two categories, viz.
expert systems (Leu et al., 2001; Shahin et al., 2008). feed-forward network and feedback or recurrent network (Jain
250 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Fig. 2. Branches of artificial intelligence (AI).

et al., 1996). In the feed-forward networks, there is no loop. For 1993; Jain et al., 1996). The supervised learning is also known as
instance, in a multilayer perceptron, the input and output neurons associative learning; it is the learning in which the network is
are arranged into layers that have unidirectional connections be- trained by making the input and matching output patterns avail-
tween them. The feed-forward networks are static so that they able. The inputeoutput pairs can be made available by an external
produce a set of output rather than a sequence of output from a teacher or be self-supervised. It resembles the conventional sta-
given input, while the recurrent network is dynamic. Fig. 3 presents tistical models in a way that the model parameters are rearranged
different subdivisions of the two types of ANN architecture (Jain in the model calibration phase (training) in order to minimize the
et al., 1996). In either type of architecture, the learning process is error between model outputs and the corresponding measured
paramount. values for the given training set (Jain et al., 1996; Shahin et al.,
ANN network learns from examples or inputs presented and the 2008). ANNs perform best when they do not extrapolate beyond
output is obtained from the connecting weights and bias. Learning the range of the data used for the training (Flood and Kartam, 1994;
in the ANN palace according to Jain et al. (1996) can be expressed as Minns and Hall, 1996; Tokar and Johnson, 1999). Unsupervised
the problem of updating network architecture and connection learning is also known as self-organization; it is the learning in
weights so that a network can perform a given specific task effi- which an output unit is trained to respond to clusters of patterns
ciently. Often, the network must learn the connection weights within the input. In this archetype, the system is supposed to
obtained from the training patterns. The performance is improved discover statistically salient features of the input population. Unlike
by continuous updating of the network weights (Jain et al., 1996). the supervised learning archetype, there is no a priori definition of
The ability of the ANN network to automatically learn from ex- categories into which the patterns are to be classified but the sys-
amples makes it unique and attractive. tem must rather develop its own representation of the input
In order to design a learning process, there must be a model stimuli.
environment in which a neural network operates, that is, the in- Reinforcement learning is a variant of supervised learning. In
formation available to the network must be known. This phe- reinforcement learning, the learning machine does some action on
nomenon is termed a learning procedure. The learning procedure the environment and receives a feedback response from the envi-
can either be supervised or unsupervised or reinforced (Masters, ronment. The learning system then does the self-grading of its

Fig. 3. Classifications of neural networks (Jain et al., 1996).


A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 251

action; if the learning has emulated the environment very well, it of the network. Hence, there is no general rule in selecting the
will be graded good or rewarding and otherwise graded bad or number of hidden neurons and it is also problem-specific (Rafiq
unrewarding and then readjust its parameters (Jain et al., 1996; et al., 2001).
Shahin et al., 2008). There are also learning rules like the traditional
expert systems, but the rules in ANNs are not humanly defined. The 2.1.3. Radial bases function
learning rules that are used in ANN are error-correction, Boltz- Radial bases function (RBF) is another subset of the multilayer
mann, Hebbian, and competitive. In addition, there are learning feed-forward back-propagation algorithm in which each of the
algorithms designed or suitable for each of the ANN architecture. neurons in the hidden layer has a radial function, for example,
For single or multilayer perceptron which is the common ANN ar- Gaussian kernel as an activation function. The location of the kernel
chitecture in geoengineering problems, perceptron algorithms, function in the hidden layer is determined by the weight vector
back-propagation, Adaline, and Madaline are the associated associated with the neurons. Both the position as well as the widths
learning algorithms (Jain et al., 1996). The variants of the feed- of the kernel must be learned from the pattern of the training.
forward network are briefly discussed in what follows. Though there are many learning algorithms for the RBF network,
the basic one employs the double-step learning means or hybrid
2.1.1. Single-layer perceptron learning. At the first stage, the widths and the position of kernel
The mode of operation of the ANN, most specifically the feed- functions are first trained using an unsupervised clustering algo-
forward network trained with back-propagation of error, can be rithm and then a supervised least mean square algorithm for
expressed mathematically, as presented in Eq. (1). For a given computation of the connection weights between the hidden and
single-layer network (Fig. 3) consisting of a scaler input and weight, output layers. A supervised gradient-based algorithm can then be
the input is multiplied by the weight and then added to the bias used to refine the network parameters after obtaining the initial
which is more or less like weight, but it has a constant value. solution. One of the basic advantages of this network is that the
hybrid method used in the training of the algorithm converges
n ¼ q þ wX (1) faster than the back-propagation algorithm in the multilayer per-
ceptron. In terms of computation time, a multilayer perceptron is
where X is the input parameter, n is the argument within the better since the required number of hidden layers is far smaller
network, w is the connecting weight, and q is the bias. than those required in the radial based function (Haykin, 1994; Jain
Transfer or activation function (f) such as piecewise linear, sig- et al., 1996). This method is not commonly employed in the pre-
moid, or Gaussian is then introduced and the network output can diction models such as those encountered in the field of
be presented in Eq. (2). The sigmoid function is the most commonly geoengineering.
used in ANNs.
2.2. Fuzzy logic
Y ¼ f ðq þ wXÞ (2)
Fuzzy logic can be viewed as a technology for the development
where Y is the output of the network.
of intelligent control and information systems. Fuzzy logic accom-
plishes machine intelligence by providing a means for representing
2.1.2. Multilayer perceptron and reasoning about human knowledge that is imprecise by nature
The multilayer perceptron is the most common class of multi- (Gupta and Kulkarni, 2013). Zadeh (1965) proposed fuzzy sets as an
layer feed-forward networks, in which each computational unit extension of the traditional set theory. The fuzzy sets allow gradual
employs either the threshold function or the sigmoid function (Jain evaluation of the membership of elements in a given set described
et al., 1996). According to Minsky and Papert (1969), arbitrarily with the aid of a membership function valued in the real unit in-
complex decision boundaries can be formed by the multilayer terval [0, 1]. The membership functions are usually linguistically
perceptron and can as well represent any Boolean function. In a defined. The linguistic variable is individually associated with a
multilayer perceptron, the weights are determined using the back- fuzzy number characterizing the meaning of each generic verbal
propagation algorithm. This has made the multilayer perceptron term, since the linguistic variable is not directly mathematically
networks to be popular among researchers and users of neural operable (Chou and Chang, 2008). Setnes et al. (1998) mentioned
networks. The mode treatment of the input dataset in the multi- that the most attractive and interesting nature of the fuzzy models
layer perceptron network can be represented mathematically in Eq. in relation to other classical methods commonly used in geo-
(3) (Shahin et al., 2009; Lawal and Idris, 2020). sciences is that they can describe complex and even curvilinear
" # multivariable problems in a transparent way (Mishra and Basu,
X
N  
Yj ¼ f qj þ wji Xi (3) 2013).
i¼1 The fuzzy logic technique can be implemented to a real appli-
cation through the following three steps as suggested by Bai and
where qj is the bias at the hidden layer; wji is the connection weight Wang (2006) and is schematically represented in Fig. 4. The first
between the input variable and the hidden layer; Xi is the input stage of implementation is fuzzification, in which the fuzzifier
variable; Yj is the output variable; f is the transfer function; and N is maps crisp data into fuzzy data or membership functions. There are
the number of neurons in the hidden layer which is commonly different membership functions such as trapezium, triangle, bell,
determined using a trial-and-error approach, because there is no Gaussian, and sigmoid that can be used for the fuzzification. The
established rule to determine it (Rafiq et al., 2001). Although some second stage is the fuzzy inference process, in which the fuzzy IF-
authors related the number of neurons to the number of input and THEN rule (fuzzy implication) combines membership functions
output variables and the number of training patterns (Rogers, 1994; with the control rules to derive the fuzzy output. Finally, defuzzi-
Swingler, 1996; Jenkins, 1997a; 1997b), this rule cannot be gener- fication which involves the use of different defuzzifiers such as the
alized according to Rafiq et al. (2001). Some studies also mentioned center of gravity (COG) and middle of maxima (MoM) converts the
that the upper bound for the number of hidden neurons should be fuzzy outputs into crisp output. Fuzzy logic has been used in various
greater than twice the number of input parameters. According to decision making and it has also been used in solving many
Rafiq et al. (2001), this rule again does not guarantee generalization engineering-related problems such as quantification of
252 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Fig. 4. Block diagram of the fuzzy logic system.

environmental impact assessment by Shepard (2005), prediction of schooling of fish. It was first proposed by Kennedy and Eberhart
CO2 cutting parameters (Parthiban et al., 2014), and study of the (1995). The mode of operation of the PSO algorithm is based on
control of steel rod quenching after hot rolling (Giorleo et al., 1997). personal experience (Pbest), overall experience (Gbest) and present
It has also been successfully implemented in the geoengineering movement of the particles to decide their next positions in the
field in predicting the rock property and various aspects of blasting search space. In addition, the experiences are accelerated by two
operation. factors c1 and c2, and two random numbers (r1 and r2) generated in
interval of [0, 1], whereas the present movement is multiplied by an
inertia factor w varying in [wmin, wmax] (Alam et al., 2015; Alam,
2.3. Genetic algorithms (GAs)
2016). The general form of the PSO algorithm is written as
Genetic algorithms (GAs) were invented by Holland (1975). It is    
kþ1 k k k
in fact an extension and generalization of Darwin’s idea and others’
Vi;j ¼ wVi;j þ c1 r1 Pbesti;j  Xi;j þ c2 r2 Gbestjk  Xi;j
k
(4a)
ideas to assist in solving other real-world problems as it is inspired
by the principles of genetics and evolution in biological populations kþ1
Xi;j k
¼ Xi;j kþ1
þ Vi;j (4b)
(Coley, 1999). It also mimics the reproduction behavior observed in
them. The basic idea of GAs is that the offspring coming from the
where Pbestki,j means the personal best jth component of the ith
parents is expected to inherit a stronger character in the two par-
individual, while Gbestkj means the global best jth component of
ents, while the weaker traits in both parents will be suppressed
the best individual of the population up to iteration k. The search
which is survival of the fittest. The more fit the trait, the higher the
mechanism of PSO in the multi-dimensional search space is pre-
possibility of surviving to the next generation. Hence, over a
sented in Fig. 5. PSO has also been used to solve various problems in
number of generations (iterations), desirable traits will evolve and
the field of engineering. Some of the problems that PSO has been
remain in the genome composition of the population over traits
used to solve in the geoengineering field will be further explained
with weaker undesirable traits. Therefore, GAs are adaptive
in the next section.
computational methods based on the mode of operation of natural
genetic systems. They exploit the historical information to forecast
or speculate a new offspring with the expected improved perfor- 2.5. Support vector machine (SVM)
mance (Coley, 1999; Gupta and Kulkarni, 2013).
The GAs has been used extensively to solve complex design A support vector machine (SVM) proposed by Boser et al. (1992)
optimization problems as it can handle both discontinuous and is a discriminative classifier which is formally defined by a sepa-
continuous variables, nonlinear objective and constraint functions rating boundary line known as a hyperplane. SVM belongs to the
without the need for the gradient information. However, De Jong general category of kernel methods (Scholkopf and Smola, 2002;
(1993) stated that applications of GAs are far more than just a Shawe-Taylor and Cristianini, 2004). A kernel method is an
robust method for estimating a series of unknown parameters
within a model of a physical system. Nevertheless, GAs have been
employed mostly for optimization problems in the field of engi-
neering. Unlike other optimization tools or algorithms which begin
their search spaces with a single guess, GAs are initialized with a
population of guesses within the search space. The initializations of
GAs are usually random and spread throughout the search space. A
typical algorithm then uses three coded operators, selection,
crossover and mutation to direct the population iteratively over
generations towards the convergence at the global optimum. The
iteration steps are based only on the objective function information
and stochastic transition rules (Coley, 1999).

2.4. Particle swarm optimization (PSO)

Particle swarm optimization (PSO) is one of the nature-inspired


metaheuristic optimization tools based on the social behavior of Fig. 5. Search mechanism of PSO in multi-dimensional search space (Mohamed,
some living things in searching for its needs like flocking of bird or 2011a).
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 253

algorithm that relies on the data only through dot-products. Noble


(2006) defined SVM as a mathematical entity, an algorithm for
maximizing a mathematical function with respect to a given
collection of data. In order to grasp the essence of SVM, four basic
concepts, i.e. the separating hyperplane, the maximum-margin
hyperplane, the soft margin, and the kernel function, must be
thoroughly understood (Noble 2006). The mode of operation of the
SVM algorithm is based on locating the hyperplane that gives the
largest minimum distance to the training examples (Fig. 6). SVM
has the ability to model complex nonlinear decision boundaries,
which makes it widely embraced in different sciences and engi-
neering fields.

3. Application of AI to rock mechanics

In the rock mechanics field (e.g. structural rock mechanics and


rock comminution), various areas using AI are described below. Fig. 7. Environmental effects of blasting (Elevli and Arpaz, 2010).

3.1. Comminution aspect of rock mechanics


model (Eqs. (5) and (6)) for determination of the mean fragment
size and its equation was later modified by Cunningham (1987) by
Based on the classification of the rock mechanics by Rock Me-
integrating it with the size distribution function of Rosin and
chanics Committee of the American National Academy of Sciences
Rammler (1933) (Eq. (7)). The resulting equation is what is
in 1963 (Cook et al., 1966), drilling and blasting operation falls
currently known as the Kuz-Ram model. The Kuz-Ram model is a
under the category of comminution aspect of the rock mechanics.
famous model in predicting rock fragmentation size most impor-
In drilling and blasting operations, there are two expected out-
tantly prior to the actual blasting operation. Nevertheless, the
comes: desirable and undesirable. The desirable outcome is
model can give an unrealistic prediction of the number of fines
attributed to the quality of the fragmentation size, heave (throw),
generated during blasting (Ouchterlony, 2005). The Kuz-Ram
and muck pile shape obtained after blasting, while the undesirable
model has been modified by different studies (Kanchibotla et al.,
aspect is environmental problems (Hustrulid, 1999; Elevli and
1999; Ouchterlony, 2005; 2016; Gheibie et al., 2009) and there
Arpaz, 2010; Raina et al., 2015). The larger percentage of the
has been a significant improvement in prediction of the Kuz-Ram
explosive energy strays to the environment during blasting and
model that is most important for prediction of fines size. But the
causes a series of environmental problems ranging from ground
rock characteristics are not explicitly considered in the models. For
vibration, air overpressure, flyrock, toe and back-break, fumes and
the purpose of fragmentation size determination, the computer-
dust (Raina et al., 2014) (Fig. 7).
aided photographic techniques have been used in recent decades.
The quality of the fragmentation and the magnitude of the im-
The Wipfrag, FragScan, Split, and WIEP are available commercial
pacts of the blasting on the environment depend on the balance
software for prediction of the size distribution of fragments
between the design parameters known as controllable parameters
(Ozkahraman, 2006; Kabwe, 2018). The image analysis can be
(i.e. burden, spacing, hole depth, hole diameter, stemming length,
automated and could give a more accurate estimation of the frag-
and sub-drill) and geological parameters known as uncontrollable
mentation size. But their predictions are also subjected to errors
parameters (i.e. UCS, point load, P- and S-wave velocities, density,
that are inherent in image processing software (Maerz and Zhou,
porosity, and hardness). Researchers have made great efforts to
2000). A number of studies (e.g. Monjezi et al., 2009; 2010a;
quantify both the fragmentation sizes and the magnitude of the
Kulatilake et al., 2010; Asl et al., 2018; Mehrdanesh et al., 2018) have
effects of blasting on the environment using various methods such
used the soft computing/AI to predict the fragmentation sizes.
as experimental methods, mathematical models, empirical models
Different methods that are used in computing the fragmentation
and numerical methods. Kuznetsov (1973) proposed empirical
size is presented in Fig. 8.

1=6
Xm ¼ AðV0 =Qe Þ0:8 Qe (5)

1=6
Xm ¼ AðV0 =Qe Þ0:8 Qe ð115=SANFO Þ19=30 (6)

p ffiffiffiffiffiffiffiffiffiffi
Rm ¼ 1  eðX 0:693=Xm Þ
n
(7)

where Xm is the mean size (m); A is the rock factor; V0 is the volume
of rock to be blasted per hole (m3); Qe is the quantity of nitro-
glycerin based explosive used per hole (kg), i.e. relative weight
strength of explosive compared to ANFO; Rm is the Rosin and
Rammler’s distribution; and u is the uniformity index given in Eq.
Fig. 6. Structure of SVM on a two-dimensional feature space (Wang et al., 2018).
(8a) (for the same explosive in the bottom and column charge) and
Eq. (8b) (for different explosives in the bottom and column charge).
254 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Fig. 8. Methods of rock fragmentation size prediction.

Table 1 Table 2
List of some of the variants of scaled distance method. Existing AI models for fragment size prediction.

Source Proposed models Equation No. Source Method Predictor Number of CoD
datasets
Duvall and Petkof (1959) PPV ¼ kðR=Q Þb (10)
Langefors and Kihlstrom (1963) PPV ¼ kðQ =R2=3 Þb=2 (11) Monjezi et al. ANN D, HD, B/S, ST, N, PF, rR, C 250 0.98
Ambrseys and Hendron (1968) PPV ¼ kðR=Q 1=3 Þb (12) (2010a)
Nicholls et al. (1971) PPV ¼ 0:362ðSDÞ1:63 (13) Kulatilake ANN S/B, H/B, B/D, ST/B, PF, XB, E 91 0.94
IS 6922 (1973) PPV ¼ kðQ 2=3 =RÞ1:25 (14) et al.
 b
Ghosh and Daemen (1983) PPV ¼ k R=Q 1=2 or 1=3 eaR (15) (2010)
1:59
Ak et al. (2009) PPV ¼ 1:367ðSDÞ (16) Bahrami ANN B, S, ST, SD, PF, HD, C, BI, D, 220 0.97
Badal (2010) PPV ¼ 0:29ðSDÞ1:296 (17) et al. SMR
Mesec et al. (2010) PPV ¼ 0:508ðSDÞ1:37 (18) (2011)
Shi et al. SVM, H/B, S/B, B/D, ST/B, PF, E 90 0.96 (SVM),
Note: Q is the charge per delay, and R is the distance from the blasting point to the
(2012) ANN 0.94 (ANN)
measuring station.
Salimi et al. ANFIS DR, SC, ST, HD, BI, S/B 80 0.89
(2012)
Sayadi et al. ANN B, S, HD, SD, SC 103 0.85
(2013)
 
0:014B pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi L Esmaeili SVM, SC, ST, DR, D, S/B, BI 80 SVM (0.83),
n¼ 2:2  0:5 þ S=ð2BÞð1  W = BÞ (8a) et al. ANFIS ANFIS (0.89)
D H
(2014)
Enayatollahi ANN HD, PF, sd, BS, S/B, WD, ST, C, N, 70 0.98
  et al. RQD, st, B
0:014B pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n ¼ 2:2  0:5 þ S=ð2BÞð1  W = BÞðj2BCL =L  1j (2014)
D Shams et al. FIS B, S, HD, SH, JD, PF, ST 185 0.92
(2015)
L
þ0:1Þ0:1 (8b) Ebrahimi ANN B, S, ST, HD, PF 34 0.78
H et al.
(2015)
where B is the burden (m), S is the spacing (m), D is the hole Mehrdanesh ANN sc, UTS, Bri, Is(50), E, n, RQD, c, 432 0.98
diameter (m), W is the drilling standard deviation (m), L is the total et al. 4, B, S, ST, HB, D, XB, SH, rR, PF,
hole length (m), H is the bench height (m), and BCL is the bottom (2018) VP
Asl et al. ANN- B, S, HL, J, ST, C, PF, GSI 200 0.94
charge length.
(2018) FFA
On the other hand, mathematical, empirical and soft computing Murlidhar ANN, XB, RQD, C, PF, B/D, S/B, H/B, 111 0.82 (ANN),
models have been proposed for prediction of the negative effects of et al. ICA- ST/B 0.95
blasting operation. For instance, the scaled-distance approach was (2018) ANN (ICA_ANN)
proposed by the United States Bureau of Mining for prediction of Note: BS - bench slope; BI - blastability index; Bri - brittleness index; B/S - burden to
ground vibration in a form of peak particle velocity (PPV). The spacing ratio; B/D - burden to diameter ratio; C - charge per delay; DR - delay per
frequency and PPV are the two major indices commonly used for row; E - Young’s modulus; SMR - slope mass rating; XB - mean in situ block size; n -
Poisson’s ratio; sc - UCS; rR - rock density; SH - Schmidt rebound hardness; S/B -
assessing the blast-induced ground vibration. The scaled distance
spacing to burden ratio; H/B - height to burden ratio; ST/B - stemming length to
equation can be written as burden ratio; 4 - friction angle; c - cohesion; Is(50) - point load strength index; HB -
bench height; VP - P-wave velocity; WD - water depth; UTS - uniaxial tensile
strength; JD - joint density; J - sub-drill; st - tensile strength; HL - hole length; ST -
PPV ¼ kðSDÞb (9) stemming length; PF - powder factor; sd - specific drilling; HD - hole depth; N -
number of rows; SC - specific charge; CoD - coefficient of determination.
where SD is the scaled distance (m/kg1/2), and k and b are the field
constants.
The field constants in the scaled distance equation are usually which they are designed. This is due to the inherent variability in
computed by fitting the field measured value with the scaled dis- the rock materials which is not explicitly accounted for in Eq. (9).
tance which is the ratio of the distance from the point of mea- Since the cost of purchasing the seismograph or vibrometer and the
surement to the square-root of charge per delay. Eq. (9) is site- time required to perform field measurement coupled with various
specific and its performance is relatively good for the site in permissions from the owners of the structures near the quarry site
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 255

and the government policy are high, variants of Eq. (9) have been efforts have been made in the use of numerical modeling for pre-
proposed by the researchers such that some numerical values of diction of blast-induced ground vibration and this method is very
field constants have been made available. Some of the existing safe and can approximate the field measured values using the
variants of the scaled distance model are presented in Table 1. costly equipment. In addition, various factors influencing ground
A great deal of effort has been made (e.g. ISRM, 1992; Wu et al., vibration can also be incorporated, making this method more
1998; Ak and Konuk, 2008; Kuzu, 2008; Singh et al., 2008; Nateghi versatile.
2011; Kumar et al., 2016) to examine the effects of various rock Flyrock which is the uncontrolled traveling of rock fragments into
characteristics on the blast-induced ground vibration. The need for the region far from the blasting site has caused a large amount of
incorporation of different rock characteristics, such as density, UCS, damage ranging from severe to minor injury of the workers and the
rock quality designation (RQD), and geological strength index (GSI) dwellers of the surrounding building and loss to the mining com-
which were not considered in the few available models that panies (Jenkins and Floyd, 2000; Verkis, 2011). One of the areas
considered the rock characteristics, was stressed by Kumar et al. showing the influence of flyrock is presented in Fig. 9. It has been
(2016). In fact, this formed the basis of their studies. Neverthe- reported in the literature that flyrock is the primary cause of major
less, there are still many more rock parameters such as the rock injuries and fatalities among the negative impacts of blasting
hardness, slake durability, thermal property, porosity, Poisson’s (Jenkins and Floyd, 2000; Raina et al., 2014). Hence, the study of the
ratio, and tensile strength and many of the blast design parameters flyrock remains a subject of concern to the researchers since it has to
such as burden, spacing, stemming length, hole diameter, hole
depth, powder factor, and explosive properties which have not
been considered in the models. The AI approach, most specifically Table 4
ANN, has also been used by several researchers (see Table 4) in Recent researches on the use of AI for predicting blast-induced ground vibration.
recent decades for prediction of blast-induced ground vibration.
Source Method Predictor Number of CoD
More parameters have been incorporated in the soft computing datasets
models. The details of the soft computing application in blast-
Monjezi et al. ANN R, C, B/S, sc, rR, N 269 0.95
induced vibration will be chronicled in the next section. The
(2010b)
finite element method (FEM) has also been used by Torano and Monjezi et al. ANN HD, ST, R, C 182 0.95
Rodriguez (2003) and Torano et al. (2006) to predict the PPV by (2011b)
generating artificial vibrations to study the behavior of the actual Khandelwal et al. ANN R, C 130 0.92
complex waves. Lu et al. (2011) also simulated the particle vibration (2011)
Mohamed ANN, FIS R, C 162 0.94 (ANN), 0.9
velocities in the rock surrounding the Pubugou hydropower station (2011b) (FIS)
in China using an equivalent method of blasting vibration based on Fisne et al. (2011) FIS R, C 33 0.92
the dynamic FEM. They acknowledged the difficulties and costs Ghasemi et al. FIS B, S, ST, N, C, R 120 0.95
involved in simulating blasting vibration accurately using numer- (2012)
Monjezi et al. ANN C, R, TC 20 0.93
ical computation. Wu et al. (2004) and Li et al. (2017) used dynamic
(2013b)
finite element software to simulate the blasting vibration. Kumar Khandelwal et al. SVM R, C 170 0.95
et al. (2020) predicted single hole blast-induced ground vibration (2010)
using a dynamic FEM. It is, however, important to note that limited Khandelwal SVM R, C 150 0.96
(2010)
Armaghani et al. ANN- HD, S, B, ST, PF, C, R, N, 44 0.94
(2014a) PSO r R, J
Table 3 Armaghani et al. ANN, B/S, ST, PF, C, R, 166 0.77 (ANN),
Existing AI models for flyrock prediction. (2015b) ANFIS 0.94 (ANFIS)
Hajihassani et al. ICA- B/S, ST, C, R, VP, E 95 0.98
Source Method Predictor Number of CoD
(2015b) ANN
datasets
Hasanipanah et al. SVM R, C 80 0.96
Monjezi et al. (2010a) ANN D, HD, B/S, ST, N, 250 0.98 (2015a)
PF, rR, C Hasanipanah et al. PSO C, R 80 0.9 (linear), 0.94
Rezaei et al. (2011) FIS HD, S, B, ST, PF, J, 490 0.98 (2017a) (power)
rR, C Nguyen et al. ANN C, R 68 0.98
Monjezi et al. (2011a) ANN HD, BS, ST, PF, D, J, 192 0.97 (2019)
C, B Amiri et al. (2016) ANN, C, R 75 0.82 (ANN),
Monjezi et al. (2012b) ANN- HD, S, B, ST, PF, J, 195 0.89 ANN- 0.88 (ANN-
GA D, C, RMR KNN KNN)
Amini et al. (2012) SVM, HL, S, B, ST, PF, J, D 245 0.97 (SVM), Shang et al. (2020) SVM, S, B, R, C, PF 83 0.93 (SVM),
ANN 0.92 (ANN) FFA- 0.97 (FFA-ANN)
Armaghani et al. ANN- HD, S, B, ST, PF, C, 44 0.94 ANN
(2014a) PSO D, N, rR, J Gorgulu et al. ANN C, R, rres, VS, VP, NH, 36 0.95
Monjezi et al. (2013a) ANN HD, S, B, D, C 310 0.98 (2013) HD, ST, S, B, RMR
Khandelwal and SVM HL, S, B, ST, PF, J 187 0.95 Saadat et al. ANN C, R, ST, H 69 0.96
Monjezi (2013a) (2014)
Ghasemi et al. (2014) ANN, HL, S, B, ST, PF, C 230 0.94 (ANN), Alvarez-Vigil et al. ANN RMR, Bc/r, R, D, HL, S, 60 0.98 (PPV), 0.95
FIS 0.96 (FIS) (2012) B, IC, TC, NH, VoD (Frq)
Marto et al. (2014) ANN- rR, HD, B/S, ST, PF, 113 0.98 Armaghani et al. ANFIS C, R 109 0.97
ICA C, SH (2015c)
Armaghani et al. ANN, C, PF 232 0.92 (ANN), Mohammadnejad SVM C, R 20 0.94
(2015a) ANFIS 0.98 (ANFIS) et al. (2012)
Hasanipanah (2016) PSO S, B, ST, PF, rR 76 0.96 Ghasemi et al. ANFIS- B, S, ST, NH, C, R 120 0.96
Trivedi et al. (2014) ANN ql, B, ST, SC, sc, 95 0.983 (2016) PSO
RQD
Note: TC - total charge; DR - delay per row; VS - S-wave velocity; ICA - imperialist
Trivedi et al. (2015) ANN, ql, B, ST, SC, sc, 125 0.95 (ANN),
competitive algorithm; NH - number of holes; Bc/r - blast-control point relative
ANFIS RQD 0.98 (ANFIS)
arrangement; rres - resistivity; IC - instantaneous charge; VoD - velocity of deto-
Note: ql - linear charge concentration. nation; Frq - frequency.
256 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Fig. 9. Area of influence of flyrock (Raina et al., 2014).

do with the safety of lives and property. Therefore, various methods Raina et al. (2014). The diagrammatic representation of the sum-
such as empirical, mathematical, probabilistic, initial velocity and mary of the Raina et al. (2014)’s holistic review is presented in Fig.10.
range, and soft computing methods (Raina et al., 2014) have been In this study, the recent applications of AI approaches for predicting
used to predict the flyrock. However, despite the recent methods for the flyrock are reviewed in the next section.
predicting flyrock, the problem still remains, though there has been AI presented in the previous section has been widely used in the
a significant reduction in the incidents (Verkis and Lobb, 2007; field of mining engineering for prediction purposes. Various tech-
McKenzie, 2009; Rezaei et al., 2011; Stojadinovic et al., 2011; Verkis, niques of the AI that have been used in the fragmentation size
2011; Amini et al., 2012; Kricak et al., 2012; Raina et al., 2014). The predictions and the environmental effects of blasting operations
holistic review of the flyrock prediction methods can be found in are reviewed in this section.

Fig. 10. Summary of the Raina et al. (2014)’s holistic review of flyrock.
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 257

3.1.1. Fragmentation size estimation Majid et al. (2016) also measured the performance of the
The positive outcome of the blasting operation is the desired blasting fragmentation by predicting the number of boulders
fragmentation size. The fragmentation size is important, because it generated after blasting using both ANN and the multi-linear
determines the efficiency of the downstream operations like regression (MLR). The results of 33 blasts recorded in Golegohar
loading, hauling and crushing (Hustrulid, 1999; Singh et al., 2005; iron ore mine in Iran were used for the model development. The
Singh and Narendrula, 2009). The amount of energy that will be input parameters in their model were joint spacing, density, UCS of
consumed in the downstream operation is significantly dependent the intact rock, burden, spacing, stemming length, bench height to
on the quality of fragmentation size. To achieve or obtain the burden ratio, and specific charge, while the ratio of boulder volume
desired fragmentation, the mine planers must strike balance be- to pattern volume was the only targeted output. In the study, MLR
tween those parameters (controllable and uncontrollable blasting was said to perform better than the ANN model, which is contrary
parameters). The quality of the fragmentation after blasting can, to the observations in many areas of application of ANN. Dhekne
therefore, be measured accurately via sieve analysis. However, due et al. (2016) also developed an ANN model for measuring the
to the large volume of run-off mine, this becomes practically number of boulders originating from the blasts in limestone
impossible. Hence, people resolve into the use of direct boulder quarries using 300 recorded blasting data. The input parameters in
counting, image analysis, and analytical models. While the image the model are hole diameter, burden, spacing, hole depth, stem-
analysis is expensive, time-consuming, and unsuitable to pre- ming length, number of holes per row, number of rows, type of
control the design of blasting, most of the field engineers rely on explosive and quantity of explosive. The predictions of the ANN
the analytical methods such as the Kuz-Ram model. However, this were also compared with that of the MLR analysis conducted, and
model is unreliable; hence, AI is used to predict the fragmentation the ANN was found to perform better than MLR as expected.
size after blasting operation. The targeted output for the proposed Based on the reviewed studies, there is a need to re-examine the
AI models a is generally obtained from the results of the mean application of ANN and compare the results with the other AI
fragmentation size from image analysis using WipFrag, SplitDesk- techniques (e.g. ANFIS, PSO-ANN, SVM-ANN, GA-ANN, and ICA-
top, GoldSize, etc. The input parameters are some of the control- ANN) to predict the number of boulders obtained from blasting
lable parameters only or a combination of some controllable and as means of assessing better fragmentation. This method is also
uncontrollable parameters. Monjezi et al. (2012a) predicted the relatively cheaper as compared with the image analysis.
rock fragmentation using an ANN in Sungun copper mine in Iran. In
the model, burden, spacing, powder factor, the number of rows,
hole diameter, the charge per delay, stemming length, and height of 3.1.2. The environmental aspect of blasting
the bench were the input parameters, while 20%, 50% and 80% Monjezi et al. (2012b) used ANNs to evaluate the effect of
passing were the outputs. The output of the model was compared blasting parameters on the flyrock in Sungun copper mine, Iran. In
with the multivariate regression analysis and ANN predictions were the model, the blasting parameters considered as the input pa-
found better than the regression analysis. However, the ANN ar- rameters are burden, spacing, charge per delay, hole diameter, hole
chitecture selected for the model is too complex and may be depth, stemming length, specific drilling, powder factor, and rock
difficult to transform the model into practice. The fragmentation mass rating (RMR), while the only targeted output was the flyrock.
predicted by Bahrami et al. (2011), Sayadi et al. (2013) and Ebrahimi Different ANN architectures and training algorithms were tried and
et al. (2015) using ANN also considered some of the blasting design the one that gave the best performance was selected. The ANN
parameters only. output was compared with the predictions of the traditional
Mehrdanesh et al. (2018) evaluated the effect of rock mass mathematical model by Lundborg (1974) and Lundborg et al. (1975)
properties on the fragmentation using ANN. In the proposed ANN and the MLR statistical model. ANN prediction was found to be
model, 19 parameters such as UCS, tensile strength, brittleness, closer to the field measured flyrock. In the model, the proposed
point load index, Young’s modulus, Poisson’s ratio, RQD, cohesion, ANN model was not presented in the form of a tractable equation
friction angle, burden, spacing, stemming length, hole depth, hole that can be used in the quarry with similar rock characteristics. In
diameter, mean in situ block size, Schmidt rebound hardness, addition, many of the rock properties that influence the formation
density, and powder factor were considered as the input parame- of flyrock such as P- and S-wave velocities, rock density, slake
ters, while the mean fragmentation size (X50) was the only targeted durability index, modulus of elasticity and Poisson’s ratio are not
output. The output of the ANN model was compared with the considered. Trivedi et al. (2014) also predicted the blast-induced
regression analysis and decision tree. ANN was found to perform flyrock in the Indian limestone mines using the ANN and the
better than the other two models. However, rock properties such as multivariate regression analysis. Blast design and geotechnical
P- and S-wave velocities, porosity, moisture content, and textural parameters, such as linear charge concentration, burden, stemming
properties were not considered in the model. In addition, the ease length, specific charge, UCS, and RQD, have been selected as input
of application of the model is questionable since it requires the parameters and flyrock distance used as an output parameter in
users to perform various laboratory experiments. Also, the ANN their model. Those parameters were also used as the dependent
model needs to be repeated since they are not in the form of a and independent variables in the multivariate regression analysis.
simple equation as in the case of a regression model. There is a high The results predicted using ANN were compared with the field
probability of not obtaining the same results if the ANN is repeated, measured, multivariate regression model predicted and the calcu-
since it is stochastic in nature. lated flyrock. ANN was found to outperform the other methods
SVM, fuzzy inference system (FIS), and hybrid AI such as adaptive compared.
neuro FIS (ANFIS), ICA-ANN, and ANN-firefly algorithm (FFA) have also Several studies (e.g. Monjezi et al., 2010a; Monjezi et al., 2013a;
been used for predicting fragmentation (Table 2). Nonetheless, there Ghasemi et al., 2014) have also used ANN for the prediction of
are other powerful AI techniques such as PSO, GA, and the hybrid form flyrock in various mines. A hybrid form of ANN such as ANFIS, ANN-
like ANN-PSO and ANN-GA, which can render easy simple models in PSO, ICA-ANN, and ANN-GA has also been used to predict flyrocks
the form of equations that have not been used to predict fragmenta- by many studies (e.g. Monjezi et al., 2012b; Armaghani et al., 2014a,
tion. The listed AI techniques can also be explored to ensure a sus- 2015a, b; Marto et al., 2014). Other AI techniques such as SVM, FIS,
tainable and robust solution to the problem of fragmentation and PSO have also been used by very limited researchers (Rezaei
prediction since the traditional Kuz-Ram model is unreliable. et al., 2011; Amini et al., 2012; Khandelwal and Monjezi, 2013a;
258 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Hasanipanah et al., 2016). The recent AI-based models developed vibration. However, most of the proposed models did not consider
for the flyrock predictions are presented in Table 3. the physico-mechanical properties (such as UCS, Young’s modulus,
For the case of blast-induced ground vibration, AI methods have Poisson’s ratio, density, porosity, Schmidt rebound hardness, slake
been used to predict the ground vibration in terms of PPV and durability index, and thermal conductivity) of the rock and some
frequency in various mining companies. Gorgulu et al. (2013) design parameters (such as burden, spacing, hole diameter, stem-
investigated the blast-induced ground vibrations in the Tulu Bo- ming length, bench height, sub-drill, explosive type, quantity, ve-
ron open-pit mine in Turkey using the ANN. The study indicated locity of detonation, and density of explosive), although the models
that the magnitude of the blast-induced ground vibration is have demonstrated the applicability of AI methods for prediction of
affected by parameters such as the physico-mechanical properties blast-induced ground vibration. The summary of the various AI
of the rock mass, characteristics of the explosive, and blasting approaches that have been used in blast-induced ground vibration
design parameters. The proposed model was based on distance predictions is presented in Table 4.
from the measuring station to the blasting point, the maximum Back-break, over-break, and air overpressure have also been
charge per delay, number of holes, hole depth, stemming length, investigated extensively in different studies (e.g. Monjezi et al.,
spacing, burden, and other parameters such as RMR, resistivity, and 2010c; 2012b; Khandelwal and Kankar, 2011; Mohamad et al.,
P- and S-wave velocities obtained from the site investigations. The 2012; Sayadi et al., 2013; Khandelwal and Monjezi, 2013b;
prediction of the ANN model was found to be closer to the Armaghani et al., 2015b; Ebrahimi et al., 2015; Hajihassani et al.,
measured PPV as compared to the traditional scaled distance 2015b; Hasanipanah et al., 2015b; Saghatforoush et al., 2016;
approach. The limitation of the proposed model is that it cannot be Hasanipanah et al., 2017b; Koopialipoor et al., 2019) using AI
applied to the mines with similar characteristics, since they were approach. A detailed review of blast-induced ground vibration and
not presented in the form of a tractable mathematical model flyrock has been discussed, and the AI techniques used for their
(Shahin et al., 2009; Lawal and Idris, 2020). In addition, rock predictions have similar procedures with different areas of
properties and design parameters such as Young’s modulus, Pois- applications.
son’s ratio, rock density, hardness, hole diameter, total charge, and
sub-drill depth were not included in the proposed model. 3.2. Structural rock mechanics
Saadat et al. (2014) predicted blast-induced ground vibration in
an iron ore mine in Iran using a four-layered feed-forward back- The structural aspect of rock mechanics according to Cook et al.
propagation multilayer perceptron ANN. In the proposed ANN (1966) is concerned with the stability of engineering structures in
model, the maximum charge per delay, distance from blasting face which the materials are predominantly rocks. Stability is a very
to the point of monitoring, stemming length, and hole depth were imperative issue in mining/rock engineering, because it concerns
taken as inputs, while the PPV was considered as output. The per- the safety of the life of the workers, the users of the facility located
formance of the proposed ANN model was compared with those of on the surface and the underground facility. One of the key ways of
the traditional scaled distance approach and the MLR model pro- achieving the safety of life and equipment in both surface and
posed in their study. It was found that ANN performed better than underground excavation is to ensure its stability. For instance, a
the other models. In addition, the physico-mechanical properties of threat can be posed to the surface mining workers, equipment or
the rock were not considered in the model and the intensity of the the building at the cliff of the surface excavations if slope fails. The
blast-induced ground vibration was attributed to the rock proper- assessment of the stability of underground mines and the slopes in
ties. The complex ANN architecture selected for the model may be surface mines is complex, because it depends on the rock charac-
unnecessary, because a three-layered ANN network is enough to teristics which are not well understood in most cases.
give a reasonable prediction (Bishop, 1995; Onifade et al., 2019; The rock characteristics play an important role in the planning
Lawal and Idris, 2020). of both open-pit and underground mines, design of support,
Alvarez-Vigil et al. (2012) used ANN to predict the blast-induced excavation strategy and production, coal bed methane production,
ground vibration in terms of PPV and frequency in an open-pit fracking for CBM extraction, and viability of CO2 sequestration (Roy
mine in Spain. The input parameters for the ANN are RMR, rela- and Singh, 2019). Hence, a fast and accurate assessment of these
tive arrangement of the blast and the control point, distance be- characteristics is required in both the field and laboratory. How-
tween the blast and control point, blast hole diameter, blast hole ever, it is very tedious, time-consuming, and costly to carry out the
length, spacing between blast holes, burden, instantaneous charge, analysis in the laboratory or in the field. In addition, a high level of
total charge, number of blast holes, and explosive detonation ve- expertise is also required. As a result, researchers have proposed
locity. The comparison showed that ANN performed better than the some empirical/mathematical models to estimate some of the
MLR model. In the model, some of the physico-mechanical prop- mechanical properties of the rock indirectly or via some simple
erties (Young’s modulus, Poisson’s ratio, density, porosity, P- and S- parameters that can be measured in situ or in the laboratory.
wave velocities, hardness, and slake durability index) of the rock However, the majority of these models are two-parameter models
were not considered and the models were not in the mathematical that have neglected the effects or contributions of other parameters
form. on the predicted rock properties, whereas those that considered
Khandelwal and Singh (2009), Mohamed (2011b), Gao et al. multiple parameters in their models used a simple regression
(2002), Amiri et al. (2016) and Nguyen et al. (2019a) have also method but the method cannot handle the inherent variability in
used ANN to predict the blast-induced ground vibration in terms of rock engineering materials, making their models site/case-specific.
PPV. Other AI methods that have been used in predicting blast- Hence, the use of the data mining approach, most specifically AI,
induced ground vibration are FIS (Mohamed, 2011b), SVM becomes imperative to handle the variability in the rock properties.
(Hasanipanah et al., 2015a; Mohammadnejad et al., 2012; Nguyen In addition, it enables the formulation of generalized models that
2019; Shang et al., 2020), and PSO (Hasanipanah, 2017a). The can be applied to other fields with similar properties once the
hybrid AI methods such as ANFIS (Iphar et al., 2008; Armaghani model has learned those data during the training (Onifade et al.,
et al., 2015b, c), PSO-ANN (Armaghani et al., 2014a; Hajihassani 2019; Roy and Singh, 2019).
et al., 2015a), FFA-ANN (Shang et al., 2020), ANFIS-PSO (Ghasemi This section presents the efforts that have been made by
et al., 2016), and ANN-KNN (Amiri et al., 2016) have also been different studies in the use of AI to predict the physico-mechanical
used by different studies to predict the blast-induced ground properties of rocks, and its application in other aspects of structural
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 259

rock mechanics such as stability analysis and rockburst analysis has Table 5
been discussed in detail in what follows. Recent researches on the use of AI for predicting mechanical properties of rock.

Source Method Predictor Number of Predicted CoD


3.2.1. Predictions of mechanical properties of rock using AI datasets
Roy and Singh (2019) predicted the deformation properties Dehghan et al. ANN VP, Is(50), 30 sc, E 0.86 (sc), 0.77 (E)
(Young’s modulus and Poisson’s ratio) of Indian coal using ANN and (2010) SH, n
a hybrid form of ANN known as ANFIS. They have used UCS, tensile Majdi and Beiki GA- rR, RQD, 120 E 0.89
(2010) ANN n, Nj, GSI
strength, shear strength, and P-wave velocity as the input param-
Rabbani et al. ANN n, rB, WS 5000 sc 0.96
eters in the models, while the Young’s modulus and Poisson’s ratio (2012)
were the output parameters. The performances of their proposed Singh et al. ANFIS rR, Is(50), 95 E 0.66
ANN and ANFIS models were compared with the traditional MLR (2012) WA
model that was also proposed in their study. ANFIS gave the best Manouchehrian ANN Vo, Cfc, 44 sc 0.87
et al. (2012) Ca, Mi
performance followed by ANN and then MLR. It is obvious that the Monjezi et al. GA- rR, n, SH 93 sc 0.9589
input parameters used in developing the proposed models are (2012c) ANN
insufficient as parameters like density, porosity, shear modulus, Rezaei et al. FIS SH, rR, n 93 sc 0.95
and shear wave velocity were not included. In addition, the limited (2014)
Ceryan et al. ANN n, Id, VP, 840 sc 0.88
data used in developing the models may not be enough for the
(2012) ne, PSV
generalized model. The models developed were not in the form of a Beiki et al. GA rR, n, VP 5000 sc, E 0.83 (sc), 0.67 (E)
mathematical/empirical model that can be used by the intending (2013)
user(s) for estimation of the deformation properties. Torabi-Kaveh ANN VP, rR, n 105 sc 0.95
Torabi-Kaveh et al. (2015) predicted the UCS and Young’s et al. (2015)
Kainthola et al. ANFIS VP, rR 597 sc, st, c, 4 0.97 (sc), 0.95 (st),
modulus of limestone samples obtained from different dam sites (2015) 0.95 (c), 0.95 (4)
using ANN. In the model, the P-wave velocity, density, and porosity Sharma et al. ANN, VP, rR, Id 94 sc 0.95 (ANN), 0.98
were the input parameters, while the UCS and E were the targeted (2017) ANFIS (ANFIS)
output parameters. The prediction of the ANN model was Mohamad et al. PSO- rD, MC, 38 sc 0.957
(2018) ANN VP, Is(50),
compared with the regression model conducted in the study. ANN
Id
has the highest coefficient of determination as compared with the Momeni et al. ANN- Rn, r, VP, 66 sc 0.97
regression model. Roy and Singh (2019) indicated that there are still (2015) PSO Is(50)
many parameters that have a relationship with UCS and E that were Mishra and Basu FIS BPI, Is(50), 240 sc 0.98
not considered in the proposed model. In addition, though the ANN (2013) SH, VP, n,
rR
model performed better than the MLR model, the practical appli- Mishra et al. FIS, BPI, Is(50), 44 sc 0.97 (FIS), 0.92
cation of the ANN prediction by the field users may not be possible (2015) ANN, SH, VP (ANN), 0.98 (ANFIS)
since it is not in the form of the equations like the regression model. ANFIS
Dehghan et al. (2010), Manouchehrian et al. (2012), Rabbani Sarkar et al. ANN VP, Is(50), 120 sc, s 0.98 (sc), 0.98 (s)
(2010) Id, rR
et al. (2012) and Torabi-Kaveh et al. (2015) have used ANN to pre-
Armaghani et al. PSO- rD, Is(50), 230 4, c 0.966, 0.944
dict the UCS and Young’s modulus of rocks. Besides ANN, other AI (2014b) ANN sBt, UV,
methods such as fuzzy logic, GA, and hybrid form of ANN like ANFIS, SH
GA-ANN, and PSO-ANN have been used (e.g. Singh et al., 2012; Armaghani et al. ANN, n, VP, 71 sc, E 0.804 (ANN), 0.916
Rezaei et al., 2014; Mohamad et al., 2018) to predict the UCS and (2016) ICA- Is(50), SH (ICA-ANN); 0.643
ANN (sc), 0.713 (E)
Young’s modulus. These parameters are considered important in Singh et al. ANFIS rR, n, VP 90 sc, st, 0.9885 (sc), 0.7607
the rock engineering design and different models have been pro- (2017) Is(50), E (st), 0.9323 (Is(50)),
posed in the literature for their predictions considering various 0.8839 (E)
parameters. Hence, it may be difficult to compare the models’ Roy and Singh ANN, sc, st, s, 69 E, n 0.96 (E), 0.94 (n)
(2019) ANFIS VP (ANN); 0.99
performances since they were established using different input
(E),0.98 (n) (ANFIS)
parameters and different rock types. Tripathy et al. ANN sc, Is(50), 105 CAI, PR 0.97 (CAI), 0.96 (PR)
All the models have shown that AI approaches are more (2015) VP, E
powerful than traditional regression analysis. However, in terms of Kaunda and ANN RT, g, VP, 136 BI 0.84
the ease of practical use, a larger percentage of the models pre- Brian (2016) VS, n, E, l
Hussain et al. PSO, SH, VP, 79 BI 0.95 (PSO), 0.94
sented in Table 5 did not provide a tractable mathematical formula (2019) ICA rR, Is(50) (ICA)
that will eliminate the need to perform the simulation of the AI
Note: WS - water saturation; Id - slake durability index; ne - effective porosity; q -
when such a model is required for the rock with similar charac-
quartz content; Nj - number of joints per meter; n - porosity; Mc - moisture content;
teristic. This issue requires urgent attention; otherwise, the soft CAI - Cerchar abrasion index; PR - penetration rate; RT - rock type; g - unit weight;
computing models would be a mere academic exercise with little or l - Lame’s constant; PSV - petrography study values; BPI - block punch index; rB -
no practical application. Some of the models developed are pre- bulk density; rD - dry density; WC - water content; WA - water absorption; Cfc -
sented in Table 5. Feroan calcitic cement content; Ca - argillaceous cement content; Vo - void ratio;
Mi - mica content; UV - ultrasonic velocity; s - shear strength; sBt - Brazilian tensile
Kaunda and Brian (2016) also used ANN to predict the brittle- strength.
ness of limestone, metamorphic and igneous rocks using the rock
type (igneous, metamorphic or sedimentary), unit weight, P- and S-
wave velocities, dynamic Poisson’s ratio, dynamic Young’s considered in the model. Hussain et al. (2019) also predicted rock
modulus, and Lame’s constant as the input variables. The model brittleness using PSO and ICA. The Schmidt hammer rebound
showed a better fit when compared with the existing study and the number, wave velocity, density, and point load index (Is(50)) were
MLR. However, the model may not attract practical attention, selected as the independent (input) parameters and the BI values
because the ANN model was not presented in the form of a simple are considered as the dependent (output) parameter. The linear and
mathematical model like the existing model and not all the pa- power models of PSO performed better as compared with the ICA.
rameters that influence the investigated rock property were Hussain et al. (2019)’s model shows better performance when
260 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

compared with Kaunda and Brian (2016) with respect to the coef- slope angle, friction angle, cohesion and unit weight were used as
ficient of correlation. The models are also in the form of simple the input parameters in the second architecture. The targeted
mathematical equations that make their applicability visible, outputs in the two ANN models were FoS values which were ob-
although the study also neglected some rock parameters in the tained from four well-known methods of Fellenius (ordinary),
proposed model. Yagiz and Gokceoglu (2010) also used an FIS to Bishop, Janbu and Spencer. It was observed that the predicted FoS
predict rock brittleness based on the combination of rock proper- values using the two architectures were very close to that
ties such as UCS, Brazilian tensile strength, and unit weight of the computed by each of the corresponding four widely used limit
rock. The obtained fuzzy predictions were compared with the equilibrium methods.
nonlinear regression analysis results. The predictions of nonlinear Ai and Zsaki (2017) assessed the stability of homogeneous
regression are closer to the measured value in their study. In gen- slopes under dynamic load through tracked cranes using an ANN
eral, limited work has been conducted to predict the brittleness of approach. The parameters for the construction and training of their
rock using AI contrary to the case of UCS and deformation prop- ANN were obtained from the slope stability models performed
erties. Hence, there may be needed to explore more AI techniques using Monte Carlo sampling. Their proposed ANN model is capable
such as SVM, ANFIS, ANN-PSO, and GA-ANN for more accurate and of predicting the FoS of a loaded slope and the location of the
robust models for the rock brittleness index prediction. critical failure surface with relatively low accuracy. Hence, the
applicability of the AI to slope stability prediction has been
3.2.2. Stability predictions using AI demonstrated.
The accurate estimation of the stability of rock or soil is a very Aside from the slope stability analysis, ANN has also been
difficult problem in rock engineering because of the behavior of applied to the stability analysis of underground mines. For instance,
interactive parameters (such as cohesion, shear strength, friction Idris et al. (2015) stochastically assessed the stability of the un-
angle, slope height, slope angle, water pressure, and seismic force) derground mine pillar at Laisvall mine using the ANN. Mahdevari
that is yet to be fully understood and the difficulty involved in et al. (2017) predicted the stability of gate roadways in longwall
determining these parameters. Hence, stability analysis is one of mining using the ANN. Katkuri et al. (2019) used a neural network-
the multivariate dynamic systems which always poses a chal- assisted analysis for longwall gate road stability based on measured
lenging and demanding problem (Sakellariou and Ferentinou, roof convergence data. ANFIS was also used by Farid et al. (2013) to
2005). For the slope stability commonly experienced along with predict the roof fall in a coal mine. Studies carried out have also
the road cuts or surface excavations, the traditional method of shown that the AI can be used in assessing the stability of slopes
assessing the stability of the rock or soil slopes is limit equilibrium and underground mines, but many of the variants of AI methods
approach. In assessing the slope stability using this approach, one have not been used in performing this task. Hence, a more robust
needs to compute the factor of safety (FoS) which is the ratio of assessment of the stability of the slopes in both surface and un-
resisting force to the driven force (Das, 2008; Abdalla et al., 2015). derground mines is required in relation to the safety of lives and
The traditional approaches used in estimating the slope stability property. Some of the recent AI applications to the stability prob-
are based on simplified assumptions to render their equations lems are presented in Table 6.
statically determined. Hence, the suitability of the limit equilibrium
approach will depend on how good the underline assumption is. On 3.2.3. Rockburst prediction using AI
the other hand, numerical methods (analytical methods, basic nu- Rockburst is a major issue of concern in mining engineering and
merical methods, FEM, boundary element method, distinct element rock engineering, most especially in a deep underground mine
method, hybrid methods, extended numerical methods, etc.) have because of various degrees of injury that can be sustained by the
also been used to assess the stability of slopes (Jing and Hudson, workers and the destruction of the equipment as a result of its
2002; Jing 2003; Cheng et al., 2007; Verma et al., 2016). However,
Table 6
according to Jing and Hudson (2002) and Jing (2003), all numerical
Recent researches on the use of AI for stability prediction.
methods are aimed at one-to-one mechanism mapping in the
model. Source Method Predictor Number of Predicted CoD
datasets parameter
The term one-to-one mechanism mapping according to
Sakellariou and Ferentinou (2005) depicts the direct modeling of Erzin and ANN 4, c, g, k 1200 FoS 0.99a
geometry and physical mechanisms, either specifically or through Cetin
(2011)
equivalent properties, but AI approaches are none the one-to-one Mahdevari ANN, Hob, GSI, RQD, scm, 60 dmax 0.94 (ANN)a,
mapping. This characteristic makes them suitable for solving et al. SVM stm, c, 4, E, UCS 0.99 (SVM)a
various problems in rock and soil mechanics. In view of this, various (2012)
studies have used AI, mostly ANN, to assess the slope stability and Farid et al. ANFIS CMRR, PRSUP, IS, 109 RFR 0.856
(2013) DOF, MH
stability of excavations in underground mines.
Abdalla et al. ANN b, 4, c/(gH) 160 FoS 0.997
Verma et al. (2016) assessed the stability of a road cut along the (2015) b, H, 4, c, g FoS 0.983
landslide-prone area using the FEM and ANN. The FoS for various Verma et al. ANN b, ru, c, 4 100 FoS 0.987
combinations of soil/rock parameters was computed with Plaxis 2D (2016)
software using the strength reduction method. Then the ANN Mahdevari ANN sc, st, c, 4, E, s, r, 96 dmax 0.911
et al., Id, RMR
simulation was performed in the model using slope angle, pore (2017)
pressure, cohesion and friction as the input parameters and the Ai and Zsaki ANN W, Lboom, bi, x, H, 150 FoS 0.929
obtained FoS via FEM as the targeted output. The obtained FoS via (2017) b, g, c, 4, e, G, ltm
the ANN was compared with that of the FEM and they were found Note: k - seismic coefficient; Hob - height of overburden; H - slope height; scm -
to be very close, indicating that ANN is suitable for assessing the unconfined compressive strength of rock mass; stm - uniaxial tensile strength of
slope stability. rock mass; b - slope angle; ru - pore pressure; W - load lifted; Lboom - length of a
Abdalla et al. (2015) also predicted the minimum FoS against the boom; bi - boom inclination; x - crane’s distance from the top of a slope; e - ec-
centricity; G - crane’s self-weight; ltm - length of timber mat; dmax - maximum roof
slope failure in clayey soils using the ANN. The slope angle, friction displacement; RFR - roof fall rate; CMRR - quality of roof rock; PRSUP - roof bolt
angle and stability number (c/(gH)) were used as the input pa- density; IS - intersections; DOF - depth; MH - mining height.
a
rameters in one of the two architectures of the ANN model, while CoD of the training data.
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 261

occurrence. This term is commonly used to describe various levels For group 3 in the case of rockburst risk index, the depth, UCS,
of rock failures, which are commonly experienced in tunnels, deformability modulus, rockburst maximum stress, and stress ratio
shafts, caverns and mines (Dowding and Andersson, 1986; Kaiser were used, and for group 4, percentage of carbon was added. The
et al., 1996; Blake and Hedley, 2003; Li et al., 2017a, b; Zhou et al., results obtained using ANN and SVM were compared with that of
2018). Rockburst occurred in almost all the existing underground MLR analysis. SVM performed better than the remaining two
constructions and mines in the whole world (e.g. South Africa, models for groups 1 and 2 in the rockburst maximum stress pre-
China, Chile, USA, Canada, Norway, Russia, Sweden, Switzerland diction. Whereas in the rockburst risk index, SVM performed better
and Korea) (Zhou et al., 2018; Pu et al., 2019). Due to the complex in group 3, while ANN performed the best in group 4. The study
nature of the rockburst, the phenomenon is yet to be understood indicated that only MLR provides an equation relating to the output
due to its dynamic and nonlinear multivariate nature (Hoek and and the input variables. Hence, the practical applicability of the best
Brown, 1980). In general, rockburst methods of evaluation can be performing model is uncertain.
categorized in two ways: rockburst classification and rockburst This review has shown that AI has been used in predicting the
potential prediction (Zhou et al., 2018). Due to the lack of detail rockburst which is a major catastrophic hazard in underground
information and noisy data with regard to studies on rockburst, the constructions and mines. More details on the application of ma-
traditional means of assessing rockburst cannot capture the model chine learning in rockburst applications can be found in Zhou et al.
and parameter uncertainty. This has motivated researchers to (2018) and Pu et al. (2019). Notwithstanding, there are other AI
explore the use of AI to model/capture the uncertainty in the techniques, especially the hybrid ones, that have not been used in
rockburst prediction parameters (Zhou et al., 2016; Pu et al., 2018). solving the problem of rockburst. The performances of the AI
Some of the recently published work on rockburst prediction is techniques have been established in different studies, but the
discussed as follows. models developed are not presented in a simple equation form like
Jia et al. (2013) predicted rockburst using the PSO algorithm and the MLR. Some of the recent researches that employed AI in pre-
the general regression neural network (GRNN) model. In their dicting rockburst are delineated in Table 7.
approach, ANN was used to build the regression model using the
existing rockburst database and the PSO was used to optimize the
input parameters of the network to minimize the adverse effect of a 4. Discussion, conclusion and future direction
human-induced factor in the model construction. There are four
input parameters in the proposed PSO-GRNN model, including Researches have shown that rock mechanics/geoengineering
tangential stress on the boundaries of tunnels or caverns, UCS, problems are highly characterized with limited data and knowl-
uniaxial tensile strength of the rock, and elastic energy index of the edge, and site-specific. Therefore, AI is a promising tool to simulate
rock. Those parameters were obtained based on energy theory and such a relationship, because it does not require any prior knowl-
data obtained from 26 practical cases. The proposed PSO-GRNN edge about the kind of relationship that exists between the pre-
applicability was tested in the Cangling tunnel and Dongguashan dictors and the predicted variables. This is one of the main
copper mine and was found to be successful. advantages of AI over traditional empirical and statistical methods.
Zhou et al. (2012) used the SVM to evaluate long-term predic- However, there are still some problems with the application of AI in
tion of rockburst. In the model, buried depth H, maximum rock mechanics problems.
tangential stress of rocks, UCS and uniaxial tensile strength of rocks,
stress coefficient, rock brittleness coefficient, and elastic energy Table 7
index were the input parameters. To further enhance the perfor- Recent researches on the use of AI for rockburst prediction.
mance of the SVM, GA and PSO were used to optimize the hyper-
Source Method Predictor Number of Predicted Performance
parameters of the SVM. A total of 132 datasets were used to eval- datasets parameter
uate the rockburst. It was found that GA and PSO optimized pa-
Zhou et al. GA-SVM D, sq, sc, st, sq/sc, 132 RB 0.9494a
rameters can enhance the search rate of SVM parameter
(2012) PSO- sc/st, Wet 1
optimization. The proposed model is robust and could serve as an SVM
important model in rockburst evaluation. Zhao and Gross (2017) GSM- 0.8861
also predicted rockburst using the SVM based on the case history SVM
data. The proposed model can be used in predicting the rockburst in Adoko et al. FIS sq, sc, st, Wet 174 RB 0.458 (VAF)
(2013) ANFIS 0.92
the mine with similar rock characteristics (Pu et al., 2019). He et al. ANN D, sc, E, Cb, K, sRB 139 RB 0.9653
Zhou et al. (2016) evaluated the performances of various AI (2015) (CoD)c
approaches such as ANN, SVM, support vector regression, and SVM 0.8677
adaptive-neuro fuzzy in rockburst prediction. The performances of Zhou et al. ANN D, sq, sc, st, SCF, 246 RB 2.91b
(2016) SVMLinear sc/st, B2, Wet 7.27
the different methods were evaluated and compared. The algo-
SVMRadial 4.86
rithms and different rockburst indicators were used as the input Liu and BP sq, sq/sc, Wet 191 RB 0.7391a
features and their training sample sizes were different. Hence, the Hou PSO-BP 0.7826
bases of the comparison of the model performances may be faulty (2019) PNN 0.7826
since the models were not built based on the same parameters. PSO-PNN 0.8696
SVM 0.7391
He et al. (2015) also predicted the rockburst using ANN and SVM PSO- 0.8261
based on the laboratory tests database. The study predicted the SVM
rockburst maximum stresses and rockburst risk index using two
Note: sq - maximum tangential stress of rocks; sq/sc - stress coefficient; Wet - elastic
groups of data, each for the rockburst maximum stresses and energy index; sc/st - brittleness index; B2 - (scst)/(scþst); D - depth; De - critical
rockburst risk index. The depth, UCS, deformability modulus, depth; SCF - stress concentration factor; sRB - rockburst maximum stress; Cb -
loading, and unloading face horizontal in situ stresses and vertical percentage of carbon; K - stress ratio; RB - rockburst; BP - back-propagation
stress due to overburden are the data in group 1 used for the network; PNN - probability neural network; VAF - variance accounted for.
a
Performance evaluated using the percentage accuracy.
rockburst maximum stress prediction. While in the second group, b
Performance evaluated using the Friedman test ranking (the closer the value to
the percentages of clay, quartz, feldspar, calcite, carbon, and volume 1, the better the performance).
of the sample were added to the first group data. c
The presented CoD is the average of groups 3 and 4.
262 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

For the fragmentation analysis, ANN, SVM, ANFIS, FIS, ANN-FFA Acknowledgments
and ICA-ANN have been used to predict the fragmentation size
based on the controllable parameters only or with both uncon- This work was supported by Korea Research Fellowship
trollable and controllable parameters. Though the predicted frag- Program through the National Research Foundation of Korea
mentation size using AI is far better than the traditional Kuz-Ram funded by the Ministry of Science and ICT (Grant No.
model as evident in the literature, the practical applicability of the 2019H1D3A1A01102993).
AI model is still an issue of concern. This is because the AI-based
models are not in the form of simple equations that can be used
by the field engineers, while designing the blasting rounds thereby References
necessitates the simulation of AI for each blasting which is obvi-
ously impossible. In addition, the AI method has never been used to Abdalla, J.A., Attom, M.F., Hawileh, R., 2015. Prediction of minimum factor of safety
against slope failure in clayey soils using artificial neural network. Environ-
predict the distribution of the fragmentation size. There are still
mental Earth Science 73 (9), 5463e5477.
more AI methods that have not been used to develop models with Aladejare, A.E., Akeju, V.O., 2020. Design and sensitivity analysis of rock slope using
consideration of design and rock parameters in the existing works. Monte Carlo simulation. Geotechnical and Geological Engineering 38, 573e585.
This can further be investigated in future studies. Adoko, A.C., Gokceoglu, C., Wu, L., Zuo, Q.J., 2013. Knowledge based and data-driven
fuzzy modeling for rockburst prediction. International Journal of Rock Me-
In the case of flyrock prediction, many AI methods have also chanics and Mining Sciences 61, 86e95.
been used for its prediction as presented in Table 4. However, those Ai, X., Zsaki, A.M., 2017. Stability assessment of homogeneous slopes loaded with
models also have the same drawback as observed in the case of mobile tracked cranes - an artificial neural network approach. Cogent Engi-
neering 7 (1), 1e13.
fragmentation size. But PSO was used to predict the flyrock and its Ak, H., Konuk, A., 2008. The effect of discontinuity frequency on ground vibrations
prediction was reported in the form of the tractable mathematical produced from bench blasting: a case study. Soil Dynamics and Earthquake
model, although many of the design and rock parameters were not Engineering 28 (9), 686e694.
Ak, H., Iphar, M., Yavuz, M., Konuk, A., 2009. Evaluation of ground vibration effect of
included in their model. In addition, more robust AI that can handle blasting operations in a magnesite mine. Soil Dynamics and Earthquake Engi-
the complex variability in the input parameters should be explored neering 29 (4), 669e676.
and simplified into a simple form. This is also applicable to the case Alam, M.N., Das, B., Pant, V.A., 2015. Comparative study of metaheuristic optimi-
zation approaches for directional overcurrent relays coordination. Electric Po-
of blast-induced ground vibration. wer Systems Research 128, 39e52.
For the predictions of the mechanical properties of the rock, Alam, M.N., 2016. Codes in Matlab for particle swarm optimization. ResearchGate.
ANN, FIS and a hybrid form of ANN are mostly used. However, there https://doi.org/10.13140/RG.2.1.1078.7608.
Almeida, J.S., 2002. Predictive non-linear modeling of complex data by artificial
are many mathematical models that are available for the quick
neural networks. Current Opinion in Biotechnology 13 (1), 72e76.
estimation of the strength parameters, though they are not reliable. Alvarez-Vigil, A.E., Gonzalez-Nicieza, C., Gayarre Lopez, F., Alvarez-Fernandez, M.I.,
AI methods offer more accurate results, but their practical appli- 2012. Predicting blasting propagation velocity and vibration frequency using
cability is still uncertain, since there is a need for repetition of such artificial neural network. International Journal of Rock Mechanics and Mining
Sciences 55, 108e116.
a model before it can be applied. This requires some level of Ambraseys, N.R., Hendron, A.J., 1968. Dynamic behavior of rock masses. In: Stagg, K.,
expertise which may be difficult to obtain among the targeted users Wiley, J. (Eds.), Rock mechanics in engineering practice. Wiley, London, UK,
of the model. pp. 203e207.
Amini, H., Gholami, R., Monjezi, M., Torabi, S.R., Zadhesh, J., 2012. Evaluation of
In rockburst prediction, shallow machine learning such as SVM flyrock phenomenon due to blasting operation by support vector machine.
which can only establish a relationship between the parameters Neural Computing and Applications 21 (8), 2077e2085.
that cause rockburst and the rockburst index has been used by Amiri, M., Amnieh, H.B., Hasanipanah, M., Khanli, L.M., 2016. A new combination of
artificial neural network and K-nearest neighbors models to predict blast-
many researchers. This method is incapable of capturing the induced ground vibration and air-overpressure. Engineering with Computers
inherent and spatial variability within the data. Therefore, there is a 32 (4), 631e644.
need to explore a deeper machine learning method for the accurate Arciszewski, T., Ziarko, W., 1992. Machine learning in knowledge acquisition. In:
Arciszewski, T., Rossman, L.A. (Eds.), Knowledge acquisition in civil engineering.
prediction of the rockburst phenomena in underground construc-
American Society of Civil Engineers (ASCE), pp. 50e68.
tion and mines. Armaghani, J.D., Hajihassani, M., Mohamad, E.T., Marto, A., Noorani, S.A., 2014a.
In summary, many studies have used AI to unravel various rock Blasting-induced flyrock and ground vibration prediction through an expert
artificial neural network based on particle swarm optimization. Arabian Journal
mechanics problems as discussed in this review, but there are still
of Geosciences 7, 5383e5396.
some problems in their applications ranging from insufficient data Armaghani, D.J., Hajihassani, M., Bejarbaneh, B.Y., Marto, A., Mohamad, E.T., 2014b.
to non-availability of a simple form of the models for the intending Indirect measure of shale shear strength parameters by means of rock index
users. Hence, future researches should be tailored towards the use tests through an optimized artificial neural network. Measurement 55, 487e
498.
of advanced numerical modeling methods like FEM and DEM to Armaghani, J.D., Mohamad, E.T., Hajihassani, M., Abad, S., Marto, A.,
simulate the field situations and generate enough data for the input Moghaddam, M.R., 2015a. Evaluation and prediction of flyrock resulting from
into the AI. Bayesian approach coupled with Monte Carlo sampling blasting operations using empirical and computational methods. Engineering
with Computers 32, 109e121.
techniques (Wang and Aladejare, 2016; Aladejare and Akeju, 2020) Armaghani, J.D., Hajihassani, M., Monjezi, M., Mohamad, E.T., Marto, A.,
can as well be used to generate reliable input datasets for the AI Moghaddam, M.R., 2015b. Application of two intelligent systems in predicting
model. This can enhance generalization of the AI models and environmental impacts of quarry blasting. Arabian Journal of Geosciences 8,
9647e9665.
eliminate the problem of non-availability datasets. Monte Carlo Armaghani, J.D., Momeni, E., Khalil, A.S., Khandelwal, M., 2015c. Feasibility of ANFIS
sampling application in AI was demonstrated by Ai and Zsaki model for prediction of ground vibrations resulting from quarry blasting.
(2017). This highly coupled versatile model will be very useful in Environmental Earth Sciences 74, 2845e2860.
Armaghani, D.J., Tonnizam Mohamad, E., Momeni, E., Monjezi, M.,
the future, most importantly in various aspects of mining engi- Narayanasamy, M.S., 2016. Prediction of the strength and elasticity modulus of
neering that are very difficult to acquire data due to threat to life granite through an expert artificial neural network. Arabian Journal of Geo-
and cost of equipment. sciences 9 (48), 1e16.
Asl, P.F., Monjezi, M., Hamidi, J.K., Armaghani, D.J., 2018. Optimization of flyrock and
rock fragmentation in the Tajareh limestone mine using metaheuristics method
Declaration of competing interest of firefly algorithm. Engineering with Computers 34 (2), 241e251.
Badal, K.K., 2010. Blast vibration studies in surface mines. BS Thesis. Department of
The authors declare that they have no known competing Mining Engineering, National Institute of Technology Rourkela, Rourkela, India.
Bahrami, A., Monjezi, M., Goshtasbi, K., Ghazvinian, A., 2011. Prediction of rock
financial interests or personal relationships that could have fragmentation due to blasting using artificial neural network. Engineering with
appeared to influence the work reported in this paper. Computers 27, 177e181.
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 263

Bai, Y., Wang, D., 2006. Fundamentals of fuzzy logic control-fuzzy sets, fuzzy rules Ghasemi, E., Kalhori, H., Bagherpour, R., 2016. A new hybrid ANFISePSO model for
and defuzzification. In: Advanced Fuzzy logic technologies in industrial appli- prediction of peak particle velocity due to bench blasting. Engineering with
cations. Springer, pp. 17e36. Computers 32, 607e614.
Beiki, M., Majdi, A., Givshad, A.D., 2013. Application of genetic programming to Gheibie, S., Aghababaei, H., Hoseinie, S.H., Pourrahimian, Y., 2009. Modified Kuz-
predict the uniaxial compressive strength and elastic modulus of carbonate Ram fragmentation model and its use at the Sungun copper mine. Interna-
rocks. International Journal of Rock Mechanics and Mining Sciences 63, 159e tional Journal of Rock Mechanics and Mining Sciences 46 (6), 967e973.
169. Ghosh, A., Daemen, J.K., 1983. A simple new blast vibration predictor. In: Pro-
Bishop, C.M., 1995. Neural network for pattern recognition, 1st ed. Oxford University ceedings of the 24th U.S. Symposium of rock mechanics. American Rock Me-
Press. chanics Association (ARMA), pp. 151e161.
Blake, W., Hedley, D.G.F., 2003. Rockbursts, case studies from North American hard- Giorleo, G., Memola Capece Minutolo, F., Sergi, V., 1997. Fuzzy logic modeling and
rock mines. Society for Mining, Metallurgy, and Exploration, New York, USA. control of steel rod quenching after hot rolling. Journal of Materials Engineering
Boser, B.E., Guyon, I.M., Vapnik, V.N., 1992. A training algorithm for optimal margin and Performance 6 (5), 599e604.
classifiers. In: Haussler, D. (Ed.), Proceedings of the 5th Annual ACM Workshop Gorgulu, K., Arpaz, E., Demirci, A., Kocaslan, A., Dilmac, M.K., Yuksek, A.G., 2013.
on Computational learning theory (COLT). ACM Press, pp. 144e152. Investigation of blast-induced ground vibrations in the Tulu boron open pit
Carbonell, J., 2003. Artificial intelligence 15-381 today: introduction to AI and mine. Bulletin of Engineering Geology and the Environment 72, 555e564.
search methods. http://www.cs.cmu.edu/w15381/Lectures/intro-search.ppt. Gupta, P., Kulkarni, N., 2013. An introduction of soft computing approach over hard
Cawsey, A., Aylett, R., 2009. Artificial intelligence introduction. http://www.macs. computing. International Journal of Latest Trends in Engineering and Technol-
hw.ac.uk/walison/ai3/ppt/l1.ppt. ogy 3 (1), 254e258.
Ceryan, N., Okkan, U., Kesimal, A., 2012. Prediction of unconfined compressive Hajihassani, M., Armaghani, D.J., Monjezi, M., Mohamad, E.T., Marto, A., 2015a.
strength of carbonate rocks using artificial neural networks. Environmental Blast-induced air and ground vibration prediction: a particle swarm
Earth Sciences 68 (3), 807e819. optimization-based artificial neural network approach. Environmental Earth
Cheng, Y.M., Lansivaara, T., Wei, W.B., 2007. Two-dimensional slope stability anal- Sciences 74, 2799e2817.
ysis by limit equilibrium and strength reduction methods. Computer and Hajihassani, M., Armaghani, D.J., Marto, A., Mohamad, E.T., 2015b. Ground vibration
Geotechnics 34 (3), 137e150. prediction in quarry blasting through an artificial neural network optimized by
Chou, S.Y., Chang, Y.H., 2008. A decision support system for supplier selection based imperialist competitive algorithm. Bulletin of Engineering Geology and the
on a strategy-aligned fuzzy SMART approach. Expert Systems with Applications Environment 74 (3), 873e886.
34, 2241e2253. Hasanipanah, M., Monjezi, M., Shahnazar, A., Armaghani, D.J., Farazmand, A., 2015a.
Coley, D.A., 1999. An introduction to genetic algorithms for scientists and engineers. Feasibility of indirect determination of blast induced ground vibration based on
World Scientific Publishing Co. Pte. Ltd. support vector machine. Measurement 75, 289e297.
Cook, N.G.W., Hoek, E., Pretorius, J.P.G., Ortlepp, W.D., Salamon, M.D.G., 1966. Rock Hasanipanah, M., Armaghani, D.J., Khamesi, H., Bakhshandeh Amnieh, H.,
mechanics applied to the study of rockbursts. Journal of South Africa Institute of Ghoraba, S., 2015b. Several non-linear models in estimating air-overpressure
Mining and Metallurgy 66, 436e528. resulting from mine blasting. Engineering with Computers 32 (3), 441e455.
Cunningham, C.V.B., 1987. Fragmentation estimations and the Kuz-Ram model - Hasanipanah, M., Armaghani, D.J., Amnieh, H.B., Abd Majid, M.Z., Tahir, M.M.D.,
four years on. In: Proceedings of the 2nd international symposium on rock 2016. Application of PSO to develop a powerful equation for prediction of
fragmentation by blasting, pp. 475e487. flyrock due to blasting. Neural Computing and Applications 28, 1043e1050.
Das, B.M., 2008. Fundamentals of geotechnical engineering. Thomson, Singapore. Hasanipanah, M., Naderi, R., Kashir, J., Noorani, S.A., Qaleh, A.Z.A., 2017a. Prediction
Dehghan, S., Sattari, G., Chelgani, S.C., Aliabadi, M., 2010. Prediction of uniaxial of blast-produced ground vibration using particle swarm optimization. Engi-
compressive strength and modulus of elasticity for Travertine samples using neering with Computers 33 (2), 173e179.
regression and artificial neural networks. Mining Science and Technology 20, Hasanipanah, M., Shahnazar, A., Bakhshandeh Amnieh, H., Armaghani, D.J., 2017b.
41e46. Prediction of air-overpressure caused by mine blasting using a new hybrid
De Jong, K.A., 1993. Genetic algorithms are NOT function optimisers. Foundations of PSOeSVR model. Engineering with Computers 33 (1), 23e31.
Genetic Algorithms 2, 5e17. Haykin, S., 1994. Neural networks: a comprehensive foundation. MacMillan College
Dhekne, P., Pradhan, M., Jade, R.K., 2016. Assessment of the effect of blast hole Publishing Co., New York, USA.
diameter on the number of oversize boulders using ANN model. Journal of the He, M., Sousa, L.R., Miranda, T., Zhu, G., 2015. Rockburst laboratory tests database
Institution of Engineers (India): Series D 97 (1), 21e31. -Application of data mining techniques. Engineering Geology 185, 116e130.
Dowding, C.H., Andersson, C.A., 1986. Potential for rock bursting and slabbing in Hoek, E., Brown, E.T., 1980. Underground Excavation in Rock. Institute of Mining and
deep caverns. Engineering Geology 22, 265e279. Metallurgy, London, UK.
Duvall, W.I., Petkof, B., 1959. Spherical propagation of explosion generated strain Holland, J.H., 1975. Adaptation in natural and artificial systems. University of
pulses in rock. U.S. Department of the Interior, Bureau of Mines. Michigan Press, Ann Arbor, USA.
Ebrahimi, E., Monjezi, M., Khalesi, M.R., Armaghani, D.J., 2015. Prediction and Holling, C.S., 1978. Adaptive environmental assessment and management. Wiley.
optimization of back-break and rock fragmentation using an artificial neural Hustrulid, W., 1999. Blasting principles for open pit mining. 1. General design
network and a bee colony algorithm. Bulletin of Engineering Geology and concepts. A.A. Balkema.
Environment 75, 27e36. Hussain, A., Surendar, A., Clementking, A., Kanagarajan, S., Ilyashenko, L.K., 2019.
Enayatollahi, I., Bazzazi, A.A., Asadi, A., 2014. Comparison between neural networks Rock brittleness prediction through two optimization algorithms namely par-
and multiple regression analysis to predict rock fragmentation in open-pit ticle swarm optimization and imperialism competitive algorithm. Engineering
mines. Rock Mechanics and Rock Engineering 47, 799e807. with Computers 35 (3), 1027e1035.
Erzin, Y., Cetin, T., 2011. The use of neural networks for the prediction of the critical Idris, M.A., Basarir, H., Nordlund, E., Wettainen, T., 2013. The probabilistic estimation
factor of safety of an artificial slope subjected to earthquake forces. Scientia of rock masses properties in Malmberget mine, Sweden. Electronic Journal of
Iranica A 19 (2), 188e194. Geotechnical Engineering 18B, 269e287.
Esmaeili, M., Salimi, A., Drebenstedt, C., Abbaszadeh, M., Bazzazi, A.A., 2014. Idris, M.A., Saiang, D., Nordlund, E., 2015. Stochastic assessment of pillar stability at
Application of PCA, SVR, and ANFIS for modeling of rock fragmentation. Arabian Laisvall mine using artificial neural network. Tunnelling and Underground
Journal of Geosciences 8, 6881e6893. Space Technology 49, 307e319.
Elevli, B., Arpaz, E., 2010. Evaluation of parameters affected on the blast induced Idris, M.A., Nordlund, E., Saiang, D., 2016. Comparison of different probabilistic
ground vibration (BIGV) by using relation diagram method (RDM). Acta Mon- methods for analyzing stability of underground rock excavations. Electronic
tanistica Slovaca 15 (4), 261e268. Journal of Geotechnical Engineering 21, 6555e6585.
Farid, M., Hossein Abadi, M.M., Yazdani-Chamzini, A., Haji Yakhchali, S., Basiri, M.H., Iphar, M., Yavuz, M., Ak, H., 2008. Prediction of ground vibrations resulting from the
2013. Developing a new model based on neuro-fuzzy system for predicting roof blasting operations in an open-pit mine by adaptive neuro-fuzzy inference
fall in coal mines. Neural Computing and Application 23, 129e137. system. Environmental Geology 56 (1), 97e107.
Feng, X.T., Katsuyama, K., Wang, Y.J., Lin, Y.M., 1997. A new direction - intelligent IS 6922, 1973. Criteria for safety and design of structures subject to underground
rock mechanics and rock engineering. International Journal of Rock Mechanics blast. Bureau of Indian Standards (BIS), New Delhi, India.
and Mining Sciences 34 (1), 135e141. ISRM, 1992. Suggested method for blast vibration monitoring. International Journal
Fisne, A., Kuzu, C., Hudaverdi, T., 2011. Prediction of environmental impacts of of Rock Mechanics and Mining Sciences and Geomechanical Abstract 29 (2),
quarry blasting operation using fuzzy logic. Environmental Monitoring and 145e146.
Assessment 174 (1e4), 461e470. Jain, A.K., Mao, J., Mohiuddin, K.M., 1996. Artificial neural networks - a tutorial.
Flood, I., Kartam, N., 1994. Neural networks in civil engineering. I: principles and Computer 29 (3), 31e44.
understanding. Journal of Computing in Civil Engineering 8 (2), 131e148. Jenkins, W.M., 1997a. An introduction to neural computing for the structural en-
Gao, F., Zhang, K., Xie, B., Wang, X.Q., 2002. Using intelligent approach to predict gineer. The Structural Engineer 75 (3), 38e41.
blast-induced ground vibration. Journal of Convergence Information Technol- Jenkins, W.M., 1997b. Approximate analysis of structural grillages using a neural
ogy 7 (14), 278e285. network. Proceedings of the Institution of Civil Engineers: Structures and
Ghasemi, E., Araei, M., Hashemolhosseini, H., 2012. Development of a fuzzy model Buildings 122, 355e363.
for predicting ground vibration caused by rock blasting in surface mining. Jenkins Jr., S.S., Floyd, J., 2000. Stemming enhancement tests. In: Proceedings of
Journal of Vibration and Control 19 (5), 755e770. the 26th annual conference on explosives and Blasting Technology, pp. 191e
Ghasemi, E., Amini, H., Ataei, M., Khalokakaei, R., 2014. Application of artificial in- 204.
telligence techniques for predicting the flyrock distance caused by blasting Jing, L., Hudson, J.A., 2002. Numerical methods in rock mechanics. International
operation. Arabian Journal of Geosciences 7, 193e202. Journal of Rock Mechanics and Mining Sciences 39, 409e427.
264 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

Jing, L., 2003. A review of techniques, advances and outstanding issues in numerical Li, X.B., Zhou, J., Wang, S.F., Liu, B., 2017b. Review and practice of deep mining for
modelling for rock mechanics and rock engineering. International Journal of solid mineral resources. Chinese Journal of Nonferrous Metals 27 (6), 1236e
Rock Mechanics and Mining Sciences 40, 283e353. 1262 (in Chinese).
Jia, Y., Lu, Q., Shang, Y., 2013. Rockburst prediction using particle swarm optimi- Li, X., Huang, J.H., Luo, Y., Dong, Q., Li, Y.H., Wan, Y., Liu, T.T., 2017. Numerical
zation algorithm and general regression neural network. Chinese Journal of simulation of blast vibration and crack forming effect of rock-anchored beam
Rock Mechanics and Engineering 32 (2), 343e348 (in Chinese). excavation in deep underground caverns. Shock and Vibration 2017, 1e13.
Kabwe, E., 2018. Velocity of detonation measurement and fragmentation analysis to Liu, Y., Hou, S., 2019. Rockburst prediction based on particle swarm optimization
evaluate blasting efficacy. Journal of Rock Mechanics and Geotechnical Engi- and machine learning algorithm. In: Proceedings of the 3rd international
neering 10, 523e533. conference. ICITG, pp. 290e303.
Kainthola, A., Singh, P.K., Verma, D., Singh, R., Sarkar, K., Singh, T.N., 2015. Prediction Lu, W., Yang, J., Chen, M., Zhou, C., 2011. An equivalent method for blasting vibration
of strength parameters of Himalayan rocks: a statistical and ANFIS approach. simulation. Simulation Modelling Practice and Theory 19, 2050e2062.
Geotechnical and Geological Engineering 33 (5), 1255e1278. Lundborg, N., 1974. The hazards of flyrock in rock blasting. Technical Reports DS 12.
Kaiser, P.K., Tannant, D.D., McCreath, D.R., 1996. Canadian rockburst support Swedish Detonic Research Foundation, Stockholm, the Netherlands.
handbook. Geomechanics Research Centre, Laurentian University, Sudbury, Lundborg, N., Persson, P.A., Ladegaard-Pedersen, A., Holmberg, R., 1975. Keeping the
Ontario, Canada. lid on flyrock in open-pit blasting. Engineering Mining Journal 176 (5), 95e100.
Kanchibotla, S.S., Valery, W., Morrell, S., 1999. Modelling fines in blast fragmentation Maerz, N.H., Zhou, W., 2000. Calibration of optical digital fragmentation measuring
and its impact on crushing and grinding. In: Proceedings of Explo’99 - a con- systems. Fragblast-International Journal for Blasting and Fragmentation 4 (2),
ference on rock breaking. The Australasian Institute of Mining and Metallurgy, 126e138.
Carlton, Australia, pp. 137e144. Majdi, A., Beiki, M., 2010. Evolving neural network using a genetic algorithm for
Katkuri, S., Deb, D., Reddy, B.V., Kumar, H., 2019. Neural network assisted analysis predicting the deformation modulus of rock masses. International Journal of
for longwall gate road stability using measured roof convergence data. Rock Mechanics and Mining Sciences 47 (2), 246e253.
Geotechnical and Geological Engineering 37 (5), 3843e3860. Mahdevari, S., Torabi, S.R., Monjezi, M., 2012. Application of artificial intelligence
Kaunda, R.B., Brian, Asbury, 2016. Prediction of rock brittleness using non- algorithms in predicting tunnel convergence to avoid TBM jamming phenom-
destructive methods for hard rock tunneling. Journal of Rock Mechanics and enon. International Journal of Rock Mechanics and Mining Sciences 55, 33e44.
Geotechnical Engineering 8 (4), 533e540. Mahdevari, S., Shahriar, K., Sharifzadeh, M., Tannant, D.D., 2017. Stability prediction
Kennedy, J., Eberhart, R., 1995. Particle swarm optimization. In: International con- of gate roadways in longwall mining using artificial neural networks. Neural
ference on neural networks. Institute of Electrical and Electronics Engineers Computing and Applications 28 (11), 3537e3555.
(IEEE), pp. 42e48. Majid, G., Nematollah, A., Dindarloo, S.R., Hamed, S., 2016. Prediction of blast
Khandelwal, M., Roy, M.P., Singh, P.K., 2004. Application of artificial neural network boulders in open pit mines via multiple regression and artificial neural net-
in mining industry. Indian Mining Engineering Journal 43, 19e23. works. International Journal of Mining Science and Technology 26 (2), 183e186.
Khandelwal, M., Singh, T.N., 2009. Prediction of blast-induced ground vibration Manouchehrian, A., Sharifzadeh, M., Moghadam, R.H., 2012. Application of artificial
using artificial neural network. International Journal of Rock Mechanics and neural networks and multivariate statistics to estimate UCS using textural
Mining Sciences 46 (7), 1214e1222. characteristics. International Journal of Mining Science and Technology 22,
Khandelwal, M., 2010. Evaluation and prediction of blast-induced ground vibration 229e236.
using support vector machine. International Journal of Rock Mechanics and Marto, A., Hajihassani, M., Armaghani, D.J., Mohamad, E.T., Makhtar, A.M., 2014.
Mining Sciences 47, 509e516. A novel approach for blast induced flyrock prediction based on imperialist
Khandelwal, M., Kankar, P.K., Harsha, S., 2010. Evaluation and prediction of blast competitive algorithm and artificial neural network. Scientific World Journal 5,
induced ground vibration using support vector machine. Mining Science and 643e715.
Technology 20, 64e70. Masters, T., 1993. Practical neural network recipes in Cþþ. Academic Press, San
Khandelwal, M., Kankar, P.K., 2011. Prediction of blast-induced air overpressure Diego, California, USA.
using support vector machine. Arabian Journal of Geosciences 4, 427e433. Mehrdanesh, A., Monjezi, M., Sayadi, A.R., 2018. Evaluation of effect of rock mass
Khandelwal, M., Kumar, D.L., Yellishetty, M., 2011. Application of soft computing to properties on fragmentation using robust techniques. Engineering with Com-
predict blast-induced ground vibration. Engineering with Computers 27 (2), puters 34 (2), 253e260.
117e125. Mesec, J., Kovac, I., Soldo, B., 2010. Estimation of particle velocity based on blast
Khandelwal, M., Monjezi, M., 2013a. Prediction of flyrock in open pit blasting event measurements at different rock units. Soil Dynamics and Earthquake
operation using machine learning method. International Journal of Mining Engineering 30 (10), 1004e1009.
Science and Technology 23, 313e316. McKenzie, C.K., 2009. Flyrock range and fragment size prediction. In: http://docs.
Khandelwal, M., Monjezi, M., 2013b. Prediction of backbreak in open-pit blasting isee.org/ISEE/Support/Proceed/General/09GENV2/09v206g.pdf.
operations using the machine learning method. Rock Mechanics and Rock En- Michalski, R.S., Michalski, I., Bratko, I., Kubat, M., 1998. Machine learning and data
gineering 46 (2), 389e396. mining: methods and applications. Wiley.
Koopialipoor, M., Armaghani, D.J., Haghighi, M., Ghaleini, E.N., 2019. A neuro-ge- Minns, A.W., Hall, M.J., 1996. Artificial neural networks as rainfall-runoff models.
netic predictive model to approximate overbreak induced by drilling and Hydrological Sciences Journal 41 (3), 399e417.
blasting operation in tunnels. Bulletin of Engineering Geology and the Envi- Minsky, M., Papert, S., 1969. Perceptrons: an introduction to computational geom-
ronment 78 (2), 981e990. etry. MIT Press, Cambridge, USA.
Kricak, L., Kecojevic, V., Negovanovic, M., Jankovic, I., Zekovic, D., 2012. Environ- Mishra, D.A., Basu, A., 2013. Estimation of uniaxial compressive strength of rock
mental and safety accidents related to blasting operation. American Journal of materials by index tests using regression analysis and fuzzy inference system.
Environmental Sciences 8 (4), 360e365. Engineering Geology 160, 54e68.
Kulatilake, P.H.S.W., Wu, Q., Hudaverdi, T., Kuzu, C., 2010. Mean particle size pre- Mishra, D.A., Srigyan, M., Basu, A., Rokade, P.J., 2015. Soft computing methods for
diction in rock blast fragmentation using neural networks. Engineering Geology estimating the uniaxial compressive strength of intact rock from index tests.
114, 298e311. International Journal of Rock Mechanics and Mining Sciences 80, 418e424.
Kuznetsov, V.M., 1973. The mean diameter of the fragments formed by blasting Mohamad, E.T., Hajihassani, M., Armaghani, D.J., Marto, A., 2012. Simulation of
rock. Soviet Mining Sciences 9, 144e148. blasting-induced air overpressure by means of artificial neural networks. In-
Kumar, R., Choudhury, D., Bhargava, K., 2016. Determination of blast-induced ternational Review on Modelling and Simulations 5 (6), 2501e2506.
ground vibration equations for rocks using mechanical and geological proper- Mohamad, E.T., Armaghani, D.J., Momeni, E., Yazdavar, A.H., Ebrahimi, M., 2018.
ties. Journal of Rock Mechanics and Geotechnical Engineering 8 (3), 341e349. Rock strength estimation: a PSO-based BP approach. Neural Computing and
Kumar, S., Mishra, A.K., Choudhary, B.S., Sinha, R.K., Deepak, D., Agrawal, H., 2020. Applications 30, 1635e1646.
Prediction of ground vibration induced due to single hole blast using explicit Mohamed, L., 2011a. Reservoir history matching optimisation and uncertainty
dynamics. Mining, Metallurgy and Exploration 37, 733e741. quantification in flow prediction. PhD Thesis. Heriot-Watt University, Edin-
Kuzu, C., 2008. The importance of site-specific characters in prediction models for burgh, Scotland, UK.
blast induced ground vibrations. Soil dynamics and Earthquake Engineering 28 Mohamed, M.T., 2011b. Performance of fuzzy logic and artificial neural network in
(5), 405e414. prediction of ground and air vibrations. International Journal of Rock Mechanics
Langefors, U., Kihlstrom, B., 1963. The modern technique of rock blasting. John and Mining Sciences 48 (5), 845e851.
Wiley and Sons, New York, USA. Mohammadnejad, M., Gholami, R., Ramazanzadeh, A., Jalali, M.E., 2012. Prediction
Lawal, A.I., Idris, M.A., 2020. An artificial neural network-based mathematical model of blast-induced vibrations in limestone quarries using support vector machine.
for the prediction of blast-induced ground vibrations. International Journal of Journal of Vibration and Control 18 (9), 1322e1329.
Environmental Studies 77 (2). https://doi.org/10.1080/00207233.2019.1662186. Momeni, E., Armaghani, D.J., Hajihassani, M., Amin, M.F.M., 2015a. Prediction of
Leu, S.-S., Chen, C.-N., Chang, S.-L., 2001. Data mining for tunnel support stability: uniaxial compressive strength of rock samples using hybrid particle swarm
neural network approach. Automation in Construction 10, 429e441. optimization-based artificial neural networks. Measurement 60, 50e63.
Leu, S.-S., Lo, H.-C., 2004. Neural-network-based regression model of ground surface Monjezi, M., Rezaei, M., Varjani, A.Y., 2009. Prediction of rock fragmentation due to
settlement induced by deep excavation. Automation in Construction 10, 429e blasting in Gol-E-Gohar iron mine using fuzzy logic. International Journal of
441. Rock Mechanics and Mining Sciences 46, 1273e1280.
Li, X.B., Gong, F.G., Tao, M., Dong, L.J., Du, K., Ma, C.D., Yin, T.B., 2017a. Failure Monjezi, M., Bahrami, A., Yazdani Varjani, A., 2010a. Simultaneous prediction of
mechanism and coupled static-dynamic loading theory in deep hard rock fragmentation and flyrock in blasting operation using artificial neural
mining: a review. Journal of Rock Mechanics and Geotechnical Engineering 9 network. International Journal of Rock Mechanics and Mining Sciences 47,
(4), 767e782. 476e480.
A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266 265

Monjezi, M., Ahmadi, M., Sheikhan, M., Bahrami, A., Salimi, A.R., 2010b. Predicting Rogers, J.L., 1994. Simulating structural analysis with neural networks. Journal of
blast-induced ground vibration using various types of neural networks. Soil Computing in Civil Engineering 8 (2), 252e265.
Dynamics and Earthquake Engineering 30 (11), 1233e1236. Rosin, P., Rammler, E., 1933. The laws governing the fineness of powdered coal.
Monjezi, M., Rezaei, M., Yazdian, A., 2010c. Prediction of backbreak in open-pit Journal of the Institute of Fuel 7, 29e36.
blasting using fuzzy set theory. Expert Systems with Applications 37, 2637e Roy, D.H., Singh, T.N., 2019. Predicting deformational properties of Indian coal: soft
2647. computing and regression analysis approach. Measurement 149, 106975.
Monjezi, M., Bahrami, A., Varjani, A.Y., Sayadi, A.R., 2011a. Prediction and controlling https://doi.org/10.1016/j.measurement.2019.106975.
of flyrock in blasting operation using artificial neural network. Arabian Journal Saadat, M., Khandelwal, M., Monjezi, M., 2014. An ANN based approach to predict
of Geosciences 4, 421e425. blast-induced ground vibration of Gol-E-Gohar iron ore mine, Iran. Journal of
Monjezi, M., Ghafurikalajahi, M., Bahrami, A., 2011b. Prediction of blast-induced Rock Mechanics and Geotechnical Engineering 6 (1), 67e76.
ground vibration using artificial neural networks. Tunnelling and Under- Saghatforoush, A., Monjezi, M., Shirani Faradonbeh, R., Armaghani, D.J., 2016.
ground Space Technology 26 (1), 46e50. Combination of neural network and ant colony optimization algorithms for
Monjezi, M., Mohamadi, H.A., Barati, B., Khandelwal, M., 2012a. Application of soft prediction and optimization of flyrock and back-break induced by blasting.
computing in predicting rock fragmentation to reduce environmental blasting Engineering with Computers 32 (2), 255e266.
side effects. Arabian Journal of Geosciences 7 (2), 505e511. Sakellariou, M.G., Ferentinou, M.D., 2005. A study of slope stability prediction using
Monjezi, M., Khoshalan, H.A., Varjani, A.Y., 2012b. Prediction of flyrock and back- neural networks. Geotechnical and Geological Engineering 23, 419e445.
break in open pit blasting operation: a neurogenetic approach. Arabian Journal Salimi, A.R., Esmaeili, M., Drebenstedt, C., Dehghani, M.H., 2012. A neurofuzzy
of Geosciences 5 (3), 441e448. approach for prediction of rock fragmentation in open pit mines. In: Pro-
Monjezi, M., Khoshalan, H.A., Razifard, M., 2012c. A neuro-genetic network for ceedings of the 21th international symposium on Mine Planning and Equip-
predicting uniaxial compressive strength of rocks. Geotechnical and Geological ment Selection (MPES), pp. 656e666.
Engineering 30 (4), 1053e1062. Sarkar, K., Tiwary, A., Singh, T.N., 2010. Estimation of strength parameters of rock
Monjezi, M., Mehrdanesh, A., Malek, A., Khandelwal, M., 2013a. Evaluation of effect using artificial neural networks. Bulletin of Engineering Geology and the
of blast design parameters on flyrock using artificial neural networks. Neural Environment 69, 599e606.
Computing and Applications 23, 349e356. Sayadi, A., Monjezi, M., Talebi, N., Khandelwal, M., 2013. A comparative study on the
Monjezi, M., Hasanipanah, M., Khandelwal, M., 2013b. Evaluation and prediction of application of various artificial neural networks to simultaneous prediction of
blast-induced ground vibration at Shur River Dam, Iran, by artificial neural rock fragmentation and backbreak. Journal of Rock Mechanics and Geotechnical
network. Neural Computing and Applications 22 (7e8), 1637e1643. Engineering 5 (4), 318e324.
Murlidhar, B.R., Armaghani, D.J., Mohamad, E.T., Changthan, S., 2018. Rock frag- Setnes, M., Babuska, R., Verbruggen, H.B., 1998. Rule-based modeling: precision and
mentation prediction through a new hybrid model based on imperial transparency. IEEE Transactions on Systems Man and Cybernetics, Part C 28,
competitive algorithm and neural network. Smart Construction Research 2 (1), 165e169.
1e12. Scholkopf, B., Smola, A., 2002. Learning with kernels. MIT Press, Cambridge, MA,
Nateghi, R., 2011. Prediction of ground vibration level induced by blasting at USA.
different rock units. International Journal of Rock Mechanics and Mining Sci- Shahin, M.A., Jaksa, M.B., Maier, H.R., 2009. Recent advances and future challenges
ences 4 (6), 899e908. for artificial neural systems in geotechnical engineering applications. Advances
Nguyen, H., 2019. Support vector regression approach with different kernel func- in Artificial Neural Systems. https://doi.org/10.1155/2009/308239.
tions for predicting blast-induced ground vibration: a case study in an open-pit Shahin, M.A., Jaksa, M.B., Maier, H.A., 2008. State of the art of artificial neural
coal mine of Vietnam. SN Applied Sciences 1, 283. https://doi.org/10.1007/ networks in geotechnical engineering. Electronic Journal of Geotechnical En-
s42452-019-0295-9. gineering. http://www.ejge.com/Bouquet08/Shahin/Shahin_ppr.pdf.
Nguyen, H., Buil, X.-N., Tran, Q.-H., Le, T.-Q., Do, N.-H., Hoa, L.T.H., 2019. Evaluating Shams, S., Monjezi, M., Majd, V.J., Armaghani, D.J., 2015. Application of fuzzy
and predicting blast-induced ground vibration in open-cast mine using ANN: a inference system for prediction of rock fragmentation induced by blasting.
case study in Vietnam. SN Applied Sciences 1, 125. https://doi.org/10.1007/ Arabian Journal of Geosciences 8 (12), 10819e10832.
s42452-018-0136-2. Shang, Y., Nguyen, H., Bui, X.N., Tran, Q.-H., Moayedi, H., 2020. A novel artificial
Nicholls, H.R., Charles, F.J., Duvall, W.I., 1971. Blasting vibrations and their effects on intelligence approach to predict blast-induced ground vibration in open-pit
structures. U.S. Department of the Interior, Bureau of Mines. mines based on the firefly algorithm and artificial neural network. Natural
Noble, W.S., 2006. What is a support vector machine? Nature Biotechnology 24 (12), Resources Research 29, 723e737.
1565e1567. Sharma, L.K., Visha, V., Singh, T.N., 2017. Developing novel models using neural
Onifade, M., Lawal, A.I., Aladejare, E.A., Bada, S., Idris, M.A., 2019. Prediction of gross networks and fuzzy systems for the prediction of strength of rocks from key
calorific value of solid fuels from their proximate analysis using soft computing geomechanical properties. Measurement 102, 158e169.
and regression analysis. International Journal of Coal Preparation and Utiliza- Shawe-Taylor, J., Cristianini, N., 2004. Kernel methods for pattern analysis. Cam-
tion. https://doi.org/10.1080/19392699.2019.1695605. bridge University Press, Cambridge, MA, USA.
Ouchterlony, F., 2005. The SwebrecÓ function: linking fragmentation by blasting Shepard, R.B., 2005. Quantifying environmental impact assessments using fuzzy
and crushing. Mining Technology - Transactions of the Institutions of Mining logic. Springer.
and Metallurgy: Section A 114 (1), 29e44. Shi, X.Z., Zhou, J., Wu, B.B., Huang, D., Wei, W., 2012. Support vector machines
Ouchterlony, F., 2016. The case for the median fragment size as a better fragment approach to mean particle size of rock fragmentation due to bench blasting
size descriptor than the mean. Rock Mechanics and Rock Engineering 49 (1), prediction. Transactions of Nonferrous Metals Society of China 22 (2), 432e441.
143e164. Singh, S.P., Narendrula, R., Duffy, D., 2005. Influence of blasted muck on the pro-
Ozkahraman, H.T., 2006. Fragmentation assessment and design of blast pattern at ductivity of the loading equipment. In: Proceedings of the 3rd EFEE conference
Goltas Limestone Quarry, Turkey. International Journal of Rock Mechanics and on explosives and blasting. European Federation of Explosives Engineers (EFEE),
Mining Sciences 43 (4), 628e633. pp. 347e353.
Parthiban, A., Ravikumar, R., Zubar, A., Duraiselvam, M., 2014. Experimental inves- Singh, T.N., Dontha, L.K., Bhardwaj, V., 2008. Study into blast vibration and fre-
tigation of CO2 laser cutting on AISI 316L sheet. Journal of Scientific and In- quency using ANFIS and MVRA. Transactions of the Institution of Mining and
dustrial Research 73, 387e393. Metallurgy, Section A: Mining Technology 117 (3), 116e121.
Pu, Y., Apel, D.B., Lingga, B., 2018. Rockburst prediction in kimberlite using decision Singh, S.P., Narendrula, R., 2009. Causes, implications and control of oversize during
tree with incomplete data. Journal of Sustainable Mining 17 (3), 158e165. blasting. In: Proceedings of the 9th international symposium on rock frag-
Pu, Y., Apel, D.B., Liu, V., Mitri, H., 2019. Machine learning methods for rockburst mentation by blasting. CRC Press, pp. 311e317.
prediction-state-of-the-art review. International Journal of Mining Science and Singh, R., Kainthola, A., Singh, T.N., 2012. Estimation of elastic constant of rocks
Technology 29, 565e570. using an ANFIS approach. Applied Soft Computing 12 (1), 40e45.
Rabbani, E., Sharif, F., Koolivand Salooki, M., Moradzadeh, A., 2012. Application of Singh, R., Umrao, R.K., Ahmad, M., Ansari, M.K., Sharma, L.K., Singh, T.N., 2017.
neural network technique for prediction of uniaxial compressive strength using Prediction of geomechanical parameters using soft computing and multiple
reservoir formation properties. International Journal of Rock Mechanics and regression approach. Measurement 99, 108e119.
Mining Sciences 56, 100e111. Starfield, A.M., Cundall, P.A., 1988. Towards a methodology for rock mechanics
Rafiq, M.Y., Bugmann, G., Easterbrook, D.J., 2001. Neural network design for engi- modeling. International Journal of Rock Mechanics and Mining Sciences &
neering applications. Computers and Structures 79, 1541e1552. Geomechanics Abstracts 25, 99e106.
Raina, A.K., Murthy, V.M.S.R., Soni, A.K., 2014. Flyrock in bench blasting: a Sterling, R.L., Nelson, S., 1982. Planning the development of underground space:
comprehensive review. Bulletin of Engineering Geology and the Environment Practical guidelines and a case study of Minneapolis. Undergr. Space 7, 86e103.
73 (4), 1199e1209. Stojadinovic, S., Pantovic, R., Zikic, M., 2011. Prediction of flyrock trajectories for
Raina, K.A., Murthy, V.M.S.R., Soni, A.K., 2015. Flyrock in surface mine blasting: forensic applications using ballistic flight equations. International Journal of
understanding the basics to develop a predictive regime. Current Science 108 Rock Mechanics and Mining Sciences 48, 1086e1094.
(4), 660e665. Swingler, K., 1996. Applying neural networks a practical guide. Academic Press, New
Rezaei, M., Monjezi, M., Varjani, A.Y., 2011. Development of a fuzzy model to predict York, USA.
flyrock in surface mining. Safety Science 49 (2), 298e305. Tokar, S.A., Johnson, P.A., 1999. Rainfall-runoff modeling using artificial neural
Rezaei, M., Majdi, A., Monjezi, M., 2014. An intelligent approach to predict uncon- networks. Journal of Hydrologic Engineering 4 (3), 232e239.
fined compressive strength of rock surrounding access tunnels in longwall coal Torabi-Kaveh, M., Naseri, F., Saneie, S., Sarshari, B., 2015. Application of artificial
mining. Neural Computing and Applications 24 (1), 233e241. neural networks and multivariate statistics to predict UCS and E using physical
266 A.I. Lawal, S. Kwon / Journal of Rock Mechanics and Geotechnical Engineering 13 (2021) 248e266

properties of Asmari limestones. Arabian Journal of Geosciences 8 (5), 2889e Xue, X., Xiao, M., 2017. Deformation evaluation on surrounding rocks of under-
2897. ground caverns based on PSO-LSSVM. Tunnelling and Underground Space
Torano, J., Rodriguez, R., 2003. Simulation of the vibration produced during the rock Technology 69, 171e181.
excavation by different methods. In: Proceedings of the 11th international Yagiz, S., Gokceoglu, C., 2010. Application of fuzzy inference system and nonlinear
conference on computational methods and experimental measurements. WIT regression models for predicting rock brittleness. Expert Systems with Appli-
Press, pp. 343e349. cations 37 (3), 2265e2272.
Torano, J., Rodriguez, R., Diego, I., Rivas, J.M., Casal, M.D., 2006. FEM models Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8 (3), 338e353.
including randomness and its application to the blasting vibrations prediction. Zhao, Z., Gross, L., 2017. Using supervised machine learning to distinguish micro-
Computers and Geotechnics 33 (2), 121e131. seismic from noise events. In: SEG international exposition and annual meeting.
Tripathy, A., Singh, T.N., Kundu, J., 2015. Prediction of abrasiveness index of some Society of Exploration Geophysicists (SEG).
Indian rocks using soft computing methods. Measurement 68, 302e309. Zhou, J., Li, X., Shi, X., 2012. Long-term prediction model of rockburst in under-
Trivedi, R., Singh, T.N., Raina, A.K., 2014. Prediction of blast induced flyrock in Indian ground openings using heuristic algorithms and support vector machines.
limestone mines using neural networks. Journal of Rock Mechanics and Safety Science 50 (4), 629e644.
Geotechnical Engineering 6, 447e454. Zhou, J., Li, X., Mitri, H.S., 2016. Classification of rockburst in underground projects:
Trivedi, R., Singh, T.N., Gupta, N., 2015. Prediction of blast-induced flyrock in comparison of ten supervised learning methods. Journal of Computing in Civil En-
opencast mines using ANN and ANFIS. Geotechnical and Geological Engineering gineering 30 (5), 04016003. https://doi.org/10.1061/(ASCE)CP.1943-5487.0000553.
33 (4), 875e891. Zhou, J., Li, Z., Mitri, H.S., 2018. Evaluation method of rockburst: state-of-the-art
Verkis, H., 2011. Flyrock: a continuing blast safety threat. In: http://docs.isee.org/ literature review. Tunnelling and Underground Space Technology 8, 632e
ISEE/Support/Proceed/General/11GENV1/11v161g.pdf. 659.
Verkis, H., Lobb, T., 2007. Flyrock revisited: an ever-present danger in mine blasting.
In: http://docs.isee.org/ISEE/Support/Proceed/General/07GENV1/07v109g.pdf.
Verma, A.K., Kaushal, K., Chatterjee, S., 2016. Prediction model of longwall powered
support capacity using field monitored data of a longwall panel and
uncertainty-based neural network. Geotechnical and Geological Engineering Sangki Kwon obtained his BSc and MSc degrees in
34, 2033e2052. Department of Mineral Resources and Petroleum Engi-
Wallace, M.I., Ng, K.C., 2016. Development and application of underground space neering from Seoul National University, Korea, in 1987 and
use in Hong Kong. Tunnelling and Underground Space Technology 55, 257e279. 1989, respectively, and his PhD in Mining Engineering
Wang, H., Zheng, B., Yoon, S.W., Ko, H.S., 2018. A support vector machine-based from University of MissourieRolla, USA, in 1996. He had
ensemble algorithm for breast cancer diagnosis. European Journal of Opera- worked for Korea Atomic Energy Research Institute
tional Research 267, 687e699. (KAERI) from 2000 to 2011. Currently, he is a Professor
Wang, Y., Aladejare, A.E., 2016. Bayesian characterization of correlation between in the Department of Energy Resources Engineering at
uniaxial compressive strength and Young’s modulus of rock. International Inha University, Incheon, Korea. His research interests
Journal of Rock Mechanics and Mining Sciences 85, 10e19. are rock mechanics for underground excavation, thermo-
Wu, Y.K., Hao, H., Zhou, Y.X., Chong, K., 1998. Propagation characteristics of blast hydro-mechanical (THM) coupling analysis for high-level
induced shock waves in a jointed rock mass. Soil Dynamics and Earthquake radioactive waste disposal, analysis of time-dependent
Engineering 17 (6), 407e412. deformation of rock, characterization of excavation
Wu, C., Lu, Y., Hao, H., 2004. Numerical prediction of blast-induced stress wave from damaged zone around an underground opening, investi-
large-scale underground explosion. International Journal of Numerical and gation of blasting impact, and application of artificial neural network in rock
Analytical Methods in Geomechanics 28, 93e109. mechanics.

You might also like