Measurement 145 (2019) 803–819

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Experimental investigations on machine vibration in blast-hole drills

and optimization of operating parameters
Rakesh Kumar a, L.A. Kumaraswamidhas a,⇑, V.M.S.R Murthy b, S.C Vettivel c
a
Department of Mining Machinery Engineering, Indian Institute of Technology (ISM), Dhanbad, India
b
Department of Mining Engineering, Indian Institute of Technology (ISM), Dhanbad, India
c
Department of Mechanical Engineering and Technology, Chandigarh College of Engineering & Technology, Chandigarh, India

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

Article history: The present study aims at minimizing vibration while drilling and to obtain optimal operating condition
Received 30 March 2018 to enhance drill performance. In this study, the level of vibration in blast-hole drilling machines in axial
Received in revised form 16 April 2019 and lateral directions was ascertained by placing accelerometers (B&K, Type 4508) at the mast. During
Accepted 21 May 2019
the investigation, Taguchi L27 orthogonal array method was applied to optimize the number of experi-
Available online 29 May 2019
ments for analysis of functional parameters. The signal to noise (S/N) ratio and analysis of variance
(ANOVA) were used to investigate the effect of various operational parameters, namely, rotational pres-
Keywords:
sure (Rp), air pressure (Ap), pulldown pressure (Pp) and rotational speed (Rs) during vibration at the time
Vibration
Blast-hole drill machine
of rock drilling. The multi-linear regression (MLR) and artificial neural network (ANN) techniques were
Operating parameters used as well to develop empirical models for predicting vibration for different operating parameters.
Rock-bit interaction ANN model was further optimized using genetic algorithm (GA) to obtain optimum value.
R.M.S Ó 2019 Elsevier Ltd. All rights reserved.
Taguchi method
ANOVA
MLR
ANN and GA

1. Introduction stick-slip, and whirling are the phenomenon occurring due to dif-
ferent modes of vibration. Also, a part of energy transmitted to
Open-pit mining activity involves a cyclic process that includes the bit gets dissipated due to unwanted vibration, enhancing the
drilling, blasting, ore handling, and hauling. Drilling as a primary drilling cost by about 10% [3]. Drilling of larger hole enables utiliza-
step plays a crucial role in the exploration and exploitation of min- tion of heavier drilling machine, capable of delivering higher tor-
erals in mining engineering. Mostly, rotary drilling is extensively que and pull down pressure. This increase in operating range
used in large surface mining projects for overburden excavation results in greater machine vibration up to 0.3 g [1,4]. Blast-hole
throughout the world [1]. Failure of drilling facilities, including drilling machine undergoes severe vibrations during bit-rock inter-
drill string/bit failure, can lead to huge cost escalation. Drill string action due to a sudden release of elastic energy at the time of rock
vibration is a complex phenomenon primarily occurring in axial, chip formation. This energy is stored as tension in mechanical parts
lateral, and torsional modes, each of which has different destruc- of the drilling machine. The severity of vibration varies depending
tive patterns. Drill string components’ failure, bit failure, inefficient upon rock formation, which, in turn, affects the service life of drill
borehole drilling, bit and stabilizer wear, low penetration rate and strings and down-hole equipment [5,6]. The drill string vibration
decreased efficiency are the major problems arising due to drill may lead to fatigue failures and wearing a bit, resulting in the
string vibration [2]. Bit-bounce, mass imbalance, eccentricity, irregular borehole wall, which eventually becomes responsible
for inefficient drilling operations. Thus, drilling companies are con-
tinuously engaged in developing technologies and methods to sup-
Abbreviations: Ax, acceleration along x-axis (lateral acceleration); Ay, accelera-
press unwanted vibration while drilling. Although the
tion along y-axis (axial acceleration); Ax RMS, root mean square acceleration along
x-axis; Ay RMS, root mean square acceleration along y-axis; MLR, multi-linear
Measurement While Drilling (MWD) tool provides down-hole data
regression; ANN, artificial neural network; GA, genetic algorithm; RMSE, root mean and lead towards adjusting real-time drilling parameter to avoid
square value; VAF, variance account for. severe vibration, their failure due to successive shocks in conjunc-
⇑ Corresponding author. tion with high cost has led drilling companies to develop drill
E-mail address: lakdhas1978@iitism.ac.in (L.A. Kumaraswamidhas).

https://doi.org/10.1016/j.measurement.2019.05.069
0263-2241/Ó 2019 Elsevier Ltd. All rights reserved.
804 R. Kumar et al. / Measurement 145 (2019) 803–819

string vibration models, and methods to optimize the drilling 86 datasets of Miduk copper mine. The study was also conducted
parameters. This paper focuses on the optimization of drilling related to environmental impact due to blasting operation. Arma-
parameters with minimum axial and lateral vibration in order to ghani et al. [19] used an adaptive neuro-fuzzy inference system
improve drilling performance. (ANFIS) for prediction of air-overpressure induced due to blasting.
The researchers have been studying drill string vibration to Mohamad et al. [20] used ANN and empirical model to predict and
improve drilling performance and lower the production cost. Var- compare the actual distance of flyrock. Armaghani et al. [21] exam-
ious drill string models have been developed based on physical ined the tunnel boring machine performance based on hybrid
approximation. Dareing et al. [7] describe longitudinal and angular model particle swarm optimization (PSO) – artificial neural net-
vibration model while drilling a vertical hole. Vandiver et al. [8] work (ANN) and imperial competitive algorithm (ICA)-ANN using
concluded that whirling and parametric instabilities lead to bend- 1286 datasets. Akhani et al. [22] predicted spectral acceleration
ing vibration of drill collar and linear coupling between weight on characteristics of strong ground motion based on a hybrid model,
bit. However, sporadic borehole contact was not mentioned in this genetic algorithm (GA)-neural network (NN)-regression analysis
context. Jansen [9] and Jansen et al. [10] made an in-depth study (RA) and found that the hybrid model showed a good accuracy as
on whirling. Their study also focused on the impact of vibration compared to other ground motion model. A summary of the previ-
on the borehole wall. The authors concluded that drill collar ous investigation related to the application of soft computing in the
motion could be periodic or chaotic in case of using the lumped field of ground vibration and machine performance are presented
parameter method. Yigit et al. [11] worked on the coupled tor- in Table 1.
sional and bending vibrations of the drill string. Weight on bit This paper focuses on the experimental investigation of machine
(WOB) was assumed to be a function of rotation of the bit. Their vibration based on its controllable parameters in given geological
study suggested that non-linear coupling significantly affects the condition. The controllable parameters are optimized through sta-
functioning of the drill machine. They also noticed instability at tistical technique, artificial intelligence technique and regression
critical rotary speed. However, these models lack the analytical analysis with minimum axial and lateral vibration to improve
solution desirable for real-time application. Challamel [12] made blast-hole drill machine performance. As far authors knowledge it
an analytical study to characterize drilling structure stability in is for the first time statistical technique and artificial intelligence
rock-bit interaction using the Liapounov and linearized method. is applied in the field of drill string vibration to minimize vibration
Hovda [13–15] developed dynamic-semi-analytical-lumped-mult and hence, to improve blast-hole machine performance. Earlier,
i-element models for axial and torsional vibration. A few research drill string models were developed and compared to suppress
findings are available on artificial intelligence technique and meth- vibration while drilling. The Previous study also lacks the applica-
ods for optimization of operating drilling parameters with mini- tion of artificial intelligence technologies such as ANN, GA, etc. to
mum vibration. predict drill machine vibration. Although this technique has been
Artificial intelligence technique such as ANN, PSO, GA, ACO, utilized to predict ground vibration induced by blasting operation.
GEP, etc. finds its application in the various field of mining research In this study, the axial and lateral vibration of the blast-hole drill
such as ground vibration prediction after blasting operation and machine was considered. The effect of machine operating parame-
drilling performance. Blasting operation is the step performed after ter i.e air pressure, rotational pressure, pull-down pressure, and
drilling of blast-hole. Faradonbeh et al. [16] developed a predictive rotational speed were investigated. During an investigation, based
model based on Gene expression programming (GEP) for estimat- on machine operating parameters Taguchi L27 orthogonal array
ing ground vibration produced by 102 blasting operations con- method was applied to optimize a number of an experiment for bet-
ducted in quarry mine. In another study of ground vibration, ter accuracy. Signal-to-noise (S/N) ratio and analysis of variance
Monjezi et al. [17] utilized artificial neural network (ANN), empir- (ANOVA) were utilized to study the effect of various controllable
ical and statistical model to predict peak particle velocity based on parameters. In the end, confirmation experiment was done to check
182 datasets. Hasanipanah et al. [18] developed classification and the developed model and was then compared with ANN and multi-
regression tree (CART), multiple regression and empirical models ple regression model (MLR). The experimental work, procedure,
to predict ground vibration due to a blasting operation based on and calculation are discussed further in detail in topics below.

Table 1
Summary of previous related papers.

Reference Technique applied Input Parameter Output Parameter No. of dataset

Akhani et al. (2019) GA-NN-RA FRV, T, Mw,Rrup,Vs30 Sa 1561
Faradonbeh et al (2016) GEP, NLMR BS, HD, ST, PF, MC, DI PPV 102
Monjezi et al. (2011) ANN MC, DI, ST, HD PPV 182
Hasanipanah et al (2017) CART MC, DI PPV 86
Armaghani et al. (2015) ANFIS, ANN, MR PF, MC, ST, BS, DI AOP 128
Mohamad et al. (2012) ANN PF, ST, CL, HB, HD, HA, HD, EH FRD 16
Armaghani et al. (2017) ICA-ANN, PSO-ANN, ANN UCS, BTS, RQD, RMR, WZ, TF, RPM PR 1286
Hasanipanah et al. (2016) [23] ANN, PSO-ANN HVSR, Coh, Y MSS 143
Amiri et al. (2016) [24] ANN-KNN, ANN MC, DI AOP, PPV 75
Monjezi et al. (2013) [25] ANN TC, MC, DI PPV 20
Taheri et al. (2017) [26] ABC-ANN, ANN MC, DI PPV 89
Armaghani et al. (2016) [27] ANN, ICA-ANN MC, DI AOP 70

FRV – Fault type; T – period; Mw – Earthquake magnitude; Rrup – closest distance co-seismic rupture; Vs30 – shear wave velocity averaged to 30 (m) depth; BS – Burden-to-
spacing ratio; HD – Hole depth; ST – Stemming length; PF – Powder factor; MC – Maximum charge per delay; DI – Distance from the blast face; PPV – Peak particle velocity;
CL – charge length; HB-hole diameter; HA – hole angle; EH – Explosive per hole; FRD – flyrock distance; Sa – spectral acceleration; PPV – peak particle velocity; AOP – air
overpressure; UCS – Uniaxial Compressive Strength (UCS), BTS – Brazilian Tensile Strength (BTS); RQD – Rock Quality Designation (RQD); RMR – Rock Mass Rating (RMR); WZ
– Weathering Zone; TF – Thrust Force; RPM – Revolution per minute; HVSR – Horizontal to Vertical Stress Ratio, Coh – Cohesion; Y – Young’s modulus; MSS – Maximum
Surface Settlement; GA – Genetic algorithm; NN – Neural network; GEP – Gene expression programming; NMLR – Nonlinear multiple regression; ANN – Artificial neural
network; CART – Classification and regression tree; ANFIS – Adaptive neuro-fuzzy inference system; ICA – Imperial competitive algorithm; PSO – Particle swarm
optimization.
 R. Kumar et al. / Measurement 145 (2019) 803–819 805

2. Field investigations and details of experiment air pressure is one of the important parameters in drilling
operation.
2.1. In-situ rock hardness test A view of a rotary drill with tri-cone bit, used in a well-known
northern coal mine of India and the rock-bit interaction are
The Schmidt rebound hardness test [1,28,29] was conducted depicted in Fig. 2.
using NR-type Schmidt Hammer before vibration measurement. The machine’s higher operating range is prone to failure due to
The NR-type hammer, having a range of 10–70 N/mm2 compres- vibration generated during rock-bit interaction. In general, vibra-
sive strength, was used to determine Schmidt rebound hardness tion of 0.1 g is considered unpleasant and vibration of 0.5 g is intol-
in the field. Schmidt rebound values were determined by applying erable [30].
the ISRM suggested methods. An area with 1 m
1 m was marked
at the sites and divided into 16 square blocks of 25 cm
25 cm.
2.3. Drill machine vibration measurement
The Schmidt hammer was held downwards as shown in Fig. 1 in
vertical position and 20 sets of values were recorded from a single
The role of controlling factors on drill machine vibration are yet
impact. These values were then averaged to get final rebound value
to be fully explored. The present paper highlights the experimental
[28,29].
work carried out in an open pit coal mine on a rotary drill machine
Six experimental drilling holes were studied for determining
using a tri-cone roller bit. The paper investigates the severity of
machine vibration based and the Schmidt Hammer tests were con-
vibration in rock-bit interaction during drilling as the drill string
ducted at the referred drilling site to characterize the surface hard-
rotates at different rotational speeds (80, 85 and 90 rpm). The
ness. The configuration of the rotary drill rig and hardness of the
objective is to use the derived data to alleviate vibration, hence
rock under which it was operated are listed in Table 2.
to improve machine health as well as to maximize drilling rates.
The concept of vibration has been analysed to find out a definite
approach to minimise it by selecting a proper parametric combina-
2.2. Rock drilling mechanism and rock disintegration tion. This research work is concentrated on drill dynamics, fol-
lowed by vibration analysis along lateral (side-to-side) and axial
During the rock-bit interaction in drilling process, mechanical (longitudinal motion along drill string) directions. The vibration
energy is dissipated through the bit to the rock. The rock is sub- measurement along the axis of drill string was considered as
jected to failure by a combination of pull down pressure on the Y-axis (i.e., axial direction) and the axis perpendicular to it was
bit and its rotation pressure. The pull-down pressure and the rota- considered as X-axis (i.e., lateral).
tion pressure are varied by operator of the drilling machine, DM-H, Ingersol Rand and Atlas copco brands of drill machines
depending upon the rock strata. The force exerted on the bit causes were covered in the study. The machines are powered by electric
penetration of rock. As the bit penetrates into rock, resistance is motor, and self-propelled. The drill machine is specially designed
offered by the rock surface, which results into vibration of drilling for rotary blast-hole drilling, using tri-cone roller bit. DM-H is an
components. Compressed air is used in sufficient volume to ensure electric-hydraulic drill in which electric motor having power rating
a minimum velocity of upward airflow for flushing drill cuttings of 373 kW is the prime mover. The machine has pull-down capac-
and keeping the hole clean. In case of its failure, air pressure starts ity range of 0–90,000 lbs. The rotation mechanism is operated by
increasing, leading to inefficient rock cutting and vibration. Hence, two stage spur gears with two hydraulic motors providing torques

Fig. 1. Schmidt Hammer Rebound Hardness test: A) Mine site 1. B) Mine site 2.

Table 2
Configuration of the rotary drill rig and the rock hardness.

Mine Drill Machine Rock hardness Bit Diameter Rotation pressure Pull Down pressure Air-line pressure Rotary speed
Location Model N/mm2 (m) (MPa) (MPa) (MPa) (RPM)
Site 1 DM-H Ingersoll Rand 16.38–24.64 0.311 0–39.23 0–27.5 0–2.94 0–200
Site 2 DM-H Atlas Copco 16.33–21.28 0.311 0–39.23 0–27.5 0–2.94 0–200
806 R. Kumar et al. / Measurement 145 (2019) 803–819

Fig. 2. Rotary Drill Machine with Tri-cone roller bit and Rock bit Interaction mechanism during drilling.

up to 10,000 ft-lb. Bit air is supplied by rotary screw compressor The vibration data recorded was on real time processing project
rated at 1050–2400 cfm. of B&K RT Pro Photon+.
During experimentation, vibration of drilling machines while in
operation was measured using a B&K RT Pro Photon+ instrument
by placing accelerometers (B&K, Type 4508) in lateral and axial 3. Design of optimization technique
directions on the mast of blast-hole drill. These accelerometers
are electronic sensors which are used to detect vibration signal. 3.1. Taguchi method
These sensors record acceleration attained through variation of
operating parameters, which are then sent to the main system The Taguchi method is an efficient design of experiment (DOE)
through data acquisition process as depicted in Fig. 3. technique, used for statistical analysis [32]. Design of experiment
The recording of acceleration was based on variation of rota- using Taguchi method provides the systematic approach to deter-
tional pressure (kg/cm2), air pressure (kg/cm2), pull-down pressure mine optimal condition. The method is based on an orthogonal
(kg/cm2), and rotational speed (rpm) [30,31]. Such an experiment array for analysis of the process variables spread out in a lesser
at an operational mine site is extremely challenging and time con- digit of the experiment [33]. The advantage of this method is that
suming. For this experiment, the maximum operating limit of it reduces the need for trials and time for an experiment. It is sim-
90 rpm was set by the operator from the safety point of view. ple, precise, and also used for determining significant factors in the

Fig. 3. Schematic Representation of On-field Vibration Measurement.

 R. Kumar et al. / Measurement 145 (2019) 803–819 807

shortest time. Taguchi’s method can be divided into subsequent Table 3

procedures, starting with selecting the process parameters, quality Process parameters and corresponding levels.

characteristics and the orthogonal array to design and conduct an Process parameters Levels
experiment. Then, analyze the experimental result using the 1 2 3
signal-to-noise (S/N) ratio and the analysis of variance (ANOVA).
Rotational pressure (Rp) kg/cm2 1200 1400 1600
Finally, depending on the result of the analysis, obtain the optimal Air pressure (Ap) kg/cm2 51.33 53.17 55
process parameters. Flow chart for Taguchi method is shown as in Pull-down pressure (Pp) kg/cm2 45 52 59
Fig. 4. Rotational speed (Rs) rpm 80 85 90

3.2. Selection of process parameters Table 4

Taguchi L27 orthogonal array design.
For the current experiment, controllable operating parameters i.
Experiment No. Input Parameters
e rotational pressure (kg/cm2), air pressure (kg/cm2), pull-down
pressure (kg/cm2) and rotational speed (rpm) were selected as pro- Rp kg/cm2 Ap kg/cm2 Pp kg/cm2 Rs rpm

cess variables affecting the vibration of blast-hole drill machines. 1 1200 51.33 45 80
Other factor affecting the vibration of blast-hole drill includes the 2 1200 51.33 45 85
3 1200 51.33 45 90
geological factor such as rock strength. Geological factors are
4 1200 53.16 52 80
uncontrollable whereas machine operating parameters are con- 5 1200 53.16 52 85
trollable and should be optimized. The process parameters and cor- 6 1200 53.16 52 90
responding levels are presented in Table 3. 7 1200 55.00 59 80
8 1200 55.00 59 85
9 1200 55.00 59 90
3.3. Selection of orthogonal array design 10 1400 51.33 52 80
11 1400 51.33 52 85
Orthogonal arrays are the special standard experimental model 12 1400 51.33 52 85
13 1400 53.16 59 80
proposed by Taguchi. Only a small number of experimental trials 14 1400 53.16 59 85
are needed to determine the main factors affecting production. In 15 1400 53.16 59 90
this study for greater accuracy, an orthogonal matrix of Taguchi 16 1400 55.00 45 80
L27 was used for the experimental test as it provides a basis for 17 1400 55.00 45 85
18 1400 55.00 45 90
Taguchi’s method analysis. The Taguchi orthogonal table indicates
19 1600 51.33 59 80
that all the controlled variables are taken into account equally in 20 1600 51.33 59 85
the experiments. The minimum number of trials must be 21 1600 51.33 59 90
greater than or equal to the total degree of freedom. Each line of 22 1600 53.16 45 80
the orthogonal table represents a different combination of 23 1600 53.16 45 85
24 1600 53.16 45 90
level parameters. The orthogonal Taguchi L27 array is shown in 25 1600 55.00 52 80
Table 4. 26 1600 55.00 52 85
27 1600 55.00 52 90

Fig. 4. Flow chart for Taguchi method.

808 R. Kumar et al. / Measurement 145 (2019) 803–819

4. Results and discussion taken in lateral and axial directions from the mast. Operating
parameters were varied in order to find out characteristics of vibra-
4.1. Drill machine vibration tion under different rock conditions. Vibration analysis was done at
a real-time processing project. Signal in time domain was analysed
The blast hole drilling machine, used in open-pit mining opera- as function of time.
tion, undergoes severe vibration during interaction with rocks at The vibration versus time plot in terms of acceleration, as
different layers of rock formation. In the drilling process involving shown in Figs. 5 and 6, provides information that help to character-
rock-bit interaction, functional parameters play an independent ize behaviour of the drilling operation on the basis of rock charac-
role [6]. An increase in rotational speed leads to an increase in pen- teristics, by measuring the maximum vibration value (or peak)
etration rate. At the same time, it causes vibration due to variation level. During the present experiment, the peak acceleration
in rock strata. The experiment carried out in the mines had med- obtained at 80 rpm at the ‘mine site 1’ was 0.9 m/s2 along Y-axis
ium hard rock characteristics. The vibration measurements were and 0.05 m/s2 along X-axis. The peak RMS values were noted to

Fig. 5. Vibration signal (acceleration and RMS value) at varying rotational speed: (a) 80 rpm, (b) 85 rpm, and (c) 90 rpm in Mine site 1.
 R. Kumar et al. / Measurement 145 (2019) 803–819 809

Fig. 6. Vibration signal (acceleration and RMS value) at varying rotational speed: (a) 80 rpm, (b) 85 rpm, and (c) 90 rpm in Mine site 2.

be 0.07 m/s2 along the Y-axis and 0.048 along X-axis. The drilling in disintegration of rocks. As the drilling progressed downwards, it
operation, conducted at 85 rpm, shows the peak acceleration to encountered rocks with varying degree of hardness. Moreover,
be 0.2 m/s2 along Y-axis and 0.09 m/s2 along X-axis. The peak changes in functional operations to overcome resistive force of
RMS values, on the other hand, were recorded as 0.13 m/s2 along rocks also caused fluctuation in vibration. High vibration along
Y-axis and 0.08 m/s2 along X-axis. The peak acceleration values the axial direction can damage drilling components and cause bit
at 90 rpm were 0.4 m/s2 along Y-axis and the 0.2 m/s2 along the wear.
X-axis. The peak RMS values obtained were 0.19 m/s2 along Y- Similar test, conducted in ‘Mine Site 2’, showed the peak accel-
axis and 0.10 m/s2 along X-axis. The test, conducted in ‘Mine Site eration and peak RMS values with varying RPM, as presented in
1’, showed the peak acceleration and peak RMS values with varying Fig. 6.
RPM, as presented in Fig. 5. The peak accelerations attained were 0.11 m/s2 along Y-axis
The data as recorded suggests that vibration along Y-axis is and 0.06 m/s2 along X-axis at 80 rpm. The peak RMS values
greater as compared to x-axis. The vibration generated during attained were 0.06 m/s2 along Y-axis and 0.045 m/s2 along X-axis
the initial drilling process was due to drill string instability as well at 80 rpm. At 85 rpm, peak acceleration values attained were
as high pull-down pressure and rotational pressure, which resulted 0.15 m/s2 along Y-axis and 0.07 m/s2 along X-axis. The peak RMS
810 R. Kumar et al. / Measurement 145 (2019) 803–819

attained were 0.10 m/s2 along Y-axis and 0.04 m/s2 along X-axis at The FFT graph, as shown in Fig. 8 (Mine 2), on the other hand,
85 rpm. The peak acceleration values attained at 90 rpm were shows peak acceleration of 0.035 m/s2 at 10 Hz along Y-axis and
0.2 m/s2 along Y-axis and 0.09 m/s2 along X-axis. The peak RMS 0.018 m/s2 at 250 Hz along X-axis at 80 rpm. At 85 rpm, peak
values recorded were 0.11 m/s2 along Y-axis and 0.07 m/s2 at acceleration attained through FFT was 0.05 m/s2 at 10 Hz along
90 rpm. Y-axis and 0.02 m/s2 at 20 Hz along X-axis. At 90 rpm, peak accel-
Table 5 represents the peak acceleration and the peak RMS eration attained was 0.030 m/s2 along Y-axis at 10 Hz and 0.012 m/
value with percentage variation at different rpm measured on s2 at 9 Hz along X-axis.
blast-hole drill machines. In both cases, drilling operation generated greater vibration in
As part of the analysis of vibration signal at various frequencies, axial direction (along Y-axis) as compared to the lateral direction
Fast Fourier Transform (FFT) graphs in Fig. 7 (Mine 1) show peak (along X-axis). This was mainly due to variation in operating
amplitude at different rpms as 0.022 m/s2 at 50 Hz along Y-axis parameters, i.e., rotational pressure, air pressure, pull-down pres-
at 80 rpm, 0.06 m/s2 at 10 Hz along Y-axis at 85 rpm and sure and rotational speed with rock characteristic at different lay-
0.081 m/s2 at 10 Hz along Y-axis at 90 rpm. ers. The operator of the drilling machine feeds the value depending

Table 5
Peak acceleration and the peak RMS value with percentage variation at different rpm.

RPM (rpm) 80 85 90
Machine 1 (DM-H Ingersoll Rand) Peak acceleration value (m/s2) X-axis 0.05 0.09 0.2
Y-axis 0.9 0.2 0.4
% variation X-axis – 80 300
Y-axis 350 – 100
Peak RMS value (m/s2) X-axis 0.05 0.08 0.1
Y-axis 0.07 0.13 0.19
% variation X-axis – 60 100
Y-axis – 85.71 171.42
Machine 2 (DM-H Atlas Copco) Peak acceleration value (m/s2) X-axis 0.06 0.07 0.09
Y-axis 0.11 0.15 0.2
% variation X-axis – 16.66 50
Y-axis – 36.36 33.33
Peak RMS value (m/s2) X-axis 0.05 0.04 0.07
Y-axis 0.06 0.1 0.11
% variation X-axis 25 – 75
Y-axis – 66.66 83.33

In this table bold values signifies the minimum peak acceleration vibration value and minimum peak RMS value obtained at given RPM. Based on minimum vibration value its
percentage variation was obtained.

Fig. 7. FFT for DM-H Blast-hole drill machine at Mine Site 1, for varying rotational speed (a) 80 rpm, (b) 85 rpm and (c) 90 rpm.
 R. Kumar et al. / Measurement 145 (2019) 803–819 811

Fig. 8. FFT for DM-H Blast-hole drill machine at Mine Site 2, for varying rotational speed (a) 80 rpm, (b) 85 rpm and (c) 90 rpm.

!
on rock hardness encountered during the drilling process. The pull- 1X n

down pressure acting on the bit enables it to penetrate into rock. lSTB ¼ 10log y2 ð2Þ
n i¼1 i
During the process of rock-bit interaction, excess pull-down pres-
sure is generated resulting in vibration along the axial direction. At
!
the same time, rock chips, produced due to drilling, need to be 1X n
1
cleared out efficiently for unhindered penetration into the rock. lLTB ¼ 10log ð3Þ
n i¼1 y2 i
During this period, appropriate air-pressure is also needed for
removing rock chips from borehole. Increase in air pressure and lNTB ,lSTB ,lLTB = S/N ratios are for nominal-the-best, smaller-the-
rotational pressure results in vibration along the axial direction. better and larger-the- better, respectively. y0 is the average value of
The vibration generated along the y-axis is greater during the data recorded, sy2 is the variation of y, n is the number of tests
removal of rock chips. If removal of rock chip is slow as compared and yi is the value of test results.
to the penetration rate of bit into rock, it can generate significant As lower value of vibration is desirable during rock drilling,
vibration leading to stick slip condition, which again results in smaller-the-better S/N ratio was chosen for this study. As per the
axial vibration. Herein lies the importance of determining suitable Taguchi method, for attaining the optimum condition, the S/N ratio
combination of operating parameters in order to prevent genera- has to be the maximum. Table 6 shows how the S/N ratio for differ-
tion of drill string vibration. ent combinations of operating parameters generates different
degrees of vibration. The average value for Ax RMS and Ay RMS
4.2. Signal-to noise (S/N) ratio analysis was calculated as 0.062 m/s2 and 0.113 m/s2, respectively. Simi-
larly, the average S/N ratio calculated for Ax RMS and Ay RMS
In the current experiment, the Taguchi method was applied to was found to be 24.558 (dB) and 19.492 (dB), respectively.
find out optimum operating parameters for reduced vibration. Tables 7 and 8 depict the S/N response in relation to the vibra-
The RMS acceleration along axial and lateral directions was acti- tion generated with different operational parameters. As per the
vated as response factor for the vibration analysis. Taguchi pro- Taguchi method, at the optimum level, each contributing factor
poses S/N ratio to serve as objective function for optimization, in should have the maximum value. In the tables, the maximum value
order to help in data analysis and to predict optimum result. For of operating factors contributing to vibration has been marked in
the present study, S/N ratio was used to analyze vibration genera- bold and ranked accordingly.
tion and drilling parameters. As the smaller-the-better condition was selected for this study,
The Taguchi method offers three modes for calculation of S/N the desired vibration condition obtained with the maximum S/N
ratio: smaller-the-better, larger-the-better and nominal-the-best ratio for each parameter under this condition is presented in the
[34]. These calculations are presented by the following equations: S/N response graph as in Figs. 9 and 10.
!
y0 2 From Fig. 9, the highest S/N ratio for mean vibration was
lNTB ¼ 10log ð1Þ obtained at Rp of 1200 kg/cm2 (level 1), Ap of 51.33 kg/cm2 (level
sy2
1), Pp of 52 kg/cm2 (level 2) and Rs of 90 rpm (level 3). The
812 R. Kumar et al. / Measurement 145 (2019) 803–819

Table 6
Experimental results and S/N ratio for L27 orthogonal layout.

Experiment No. Input Parameters output parameter

Rp Ap Pp Rs AxRMS S/N Axrms AyRMS S/N AyRMS
1 1200 51.33 45 80 0.045 26.936 0.060 24.437
2 1200 51.33 45 85 0.041 27.744 0.062 24.152
3 1200 51.33 45 90 0.040 27.959 0.059 24.582
4 1200 53.16 52 80 0.050 26.021 0.065 23.741
5 1200 53.16 52 85 0.044 27.131 0.067 23.478
6 1200 53.16 52 90 0.047 26.558 0.069 23.223
7 1200 55.00 59 80 0.045 26.936 0.070 23.098
8 1200 55.00 59 85 0.049 26.196 0.071 22.975
9 1200 55.00 59 90 0.052 25.680 0.074 22.615
10 1400 51.33 52 80 0.040 27.959 0.110 19.172
11 1400 51.33 52 85 0.042 27.535 0.112 19.016
12 1400 51.33 52 85 0.041 27.744 0.109 19.251
13 1400 53.16 59 80 0.063 24.013 0.120 18.416
14 1400 53.16 59 85 0.060 24.437 0.124 18.132
15 1400 53.16 59 90 0.057 24.883 0.126 17.993
16 1400 55.00 45 80 0.080 21.938 0.13 17.721
17 1400 55.00 45 85 0.075 22.499 0.131 17.655
18 1400 55.00 45 90 0.079 22.047 0.133 17.523
19 1600 51.33 59 80 0.070 23.098 0.110 19.172
20 1600 51.33 59 85 0.072 22.853 0.113 18.938
21 1600 51.33 59 90 0.069 23.223 0.115 18.786
22 1600 53.16 45 80 0.085 21.412 0.150 16.478
23 1600 53.16 45 85 0.088 21.110 0.151 16.420
24 1600 53.16 45 90 0.086 21.310 0.154 16.249
25 1600 55.00 52 80 0.100 20.000 0.190 14.425
26 1600 55.00 52 85 0.056 25.036 0.192 14.334
27 1600 55.00 52 90 0.091 20.819 0.193 14.289
Mean 0.062 24.558 0.113 19.492
*Corresponding unit kg/cm2 m/s2 dB m/s2 dB

Table 7
S/N response table for vibration along lateral direction due to functional parameters.

Level Rp Ap Pp Rs
1 26.77 26.11 23.65 24.47
2 24.78 24.08 25.25 23.92
3 21.93 23.29 24.58 25.09
Delta 4.84 2.82 1.60 1.17

In this table bold values signify the maximum value of operating factors con-
tributing to vibration as per the Taguchi method.

Table 8
S/N response table for vibration along axial direction due to functional parameters.

Level Rp Ap Pp Rs
1 23.58 20.83 19.47 18.97
2 18.32 19.34 18.99 20.02
3 16.56 18.29 20.01 19.47
Delta 7.02 2.54 1.02 1.05

In this table bold values signify the maximum value of operating factors con-
tributing to vibration as per the Taguchi method.
Fig. 10. Mean S/N ratio graph for vibration along axial direction due to functional
parameters.

parametric combination for minimizing vibration along the lateral

direction was found to be Rp1Ap1Pp2Rs3. Similarly, from Fig. 10, the
highest S/N ratio for mean vibration was obtained at Rp of 1200 kg/
cm2 (level 1), Ap of 51.33 kg/cm2 (level 1), Pp of 59 kg/cm2 (level 3)
and Rs of 85 rpm (level 2). The parametric combination for mini-
mizing vibration along the axial direction was found to be
Rp1Ap1Pp3Rs2.

4.3. ANOVA method

ANOVA is a statistical method to identify the effect of operating

parameters on vibration [32–34]. In the present study, the relevant
Fig. 9. S/N ratio graph for vibration along lateral direction due to functional independent operational parameters underwent ANOVA, providing
parameters. 95% of confidence level. This method calculates the percentage
 R. Kumar et al. / Measurement 145 (2019) 803–819 813

Table 9
The result attained from ANOVA: Ax RMS.

Source DF Adj.SS Adj.MS F-value P Percentage contribution

Rp 2 0.005192 0.002596 39.58 0.00 67.37
Ap 2 0.001648 0.000824 12.36 0.00 21.04
Pp 2 0.000706 0.000353 5.38 0.00 9.16
Rs 2 0.000188 0.000094 1.43 0.00 2.43
Error 18 0.001181 0.000066 0.01
Total 26 0.008915 100

Table 10
The result attained from ANOVA: Ay RMS.

Source DF Adj.SS Adj.MS F-value P Percentage contribution

Rp 2 0.033962 0.016981 4057.41 0.00 76.84
Ap 2 0.006204 0.003102 741.13 0.00 14.04
Pp 2 0.001898 0.000949 226.70 0.00 4.29
Rs 2 0.002130 0.001065 254.42 0.00 4.81
Error 18 0.000075 0.000004 0.02
Total 26 0.044268 100

contribution of each process variable, which has a significant

impact on the vibration of drill machine. The result of ANOVA for
Ax RMS and Ay RMS are presented in Tables 9 and 10 respectively.
The tables show that the P-value for each parameter is less than
the significance level of 0.05, which indicates that contribution of
the factor is significant. The percentage contribution of parameters
is obtained by F-value.

4.3.1. Effect of operating parameters on responses

Tables 11 and 12 show the operating parameters’ mean
response. The effect of operational parameters on generation of
vibration in drill machine is depicted in Figs. 11 and 12. The opti-
mized values of the factors contributing to generation of vibration
are presented as Rp1Ap1Pp2Rs3 along the lateral direction and
Rp1Ap1Pp3Rs2 along the axial direction.
As shown in Figs. 11 and 12, vibration increases with gradual
increase of rotational pressure (Rp) and air pressure (Ap). In the
Fig. 11. Mean of means for vibration along lateral direction at different operating
graph, vibration increases at higher values of the operating param-
parameters.
eters due to instability or variation of rock hardness as drilling pro-
gress downwards.

4.3.2. Estimation of optimum vibration along axial and lateral

direction
For validation of optimized condition obtained through the
Taguchi technique, an experiment was carried out [35]. The esti-
mation of optimum vibration was done with Eqs. (4) and (5)

Table 11
Response table for mean for vibration along lateral direction.

Level Rp Ap Pp Rs
1 0.04589 0.05111 0.06878 0.06144
2 0.05967 0.06444 0.05678 0.06511
3 0.07967 0.06967 0.05967 0.05867
Delta 0.03378 0.01856 0.01200 0.00644

In this table bold value signifies the optimized value of factors contributing to
generation of vibration along lateral and axial direction.
Fig. 12. Mean of means for vibration along axial direction at different operating
parameters.
Table 12
Response table for mean for vibration along axial direction.

Level Rp Ap Pp Rs      
Ax RMSopt ¼ Rp1  ax rms þ Ap1  ax rms þ Pp2  ax rms
1 0.06633 0.09444 0.11444 0.12511
2 0.12167 0.11400 0.12300 0.10367 þ ðRs3  ax rmsÞ þ ax rms ð4Þ
3 0.15200 0.13156 0.10256 0.11122
Delta 0.08567 0.03711 0.02044 0.02144      
Ay RMSopt ¼ Rp1  ay rms þ Ap1  ay rms þ Pp3  ay rms
In this table bold value signifies the optimized value of factors contributing to  
generation of vibration along lateral and axial direction. þ Rs2  ay rms þ ay rms ð5Þ
814 R. Kumar et al. / Measurement 145 (2019) 803–819

In the above equations, (Rp1Ap1Pp2Rs3) and (Rp1Ap1Pp3Rs2) b0 b1. . ..bn signify regression coefficients, e symbolises the
represent the average value of vibration at the optimum level along error. In MLR technique, multi-collinearity can become an impor-
the lateral and axial directions, respectively. In the experiment, ax tant aspect when dealing with correlated independent variables.
rms and ay rms denoted the average of all vibration values As standard error of the concerned coefficient is likely to be high
obtained along the lateral and axial directions. Thus, Ax in case of multi-collinearity, the MLR analysis may turn out to be
RMSopt = 0.0274 m/s2 and Ay RMSopt = 0.0280 m/s2. undependable. In order to determine the extent of multi-
The 95% confidence interval for the vibration response along the collinearity, generally closely related statistics of Variance inflation
lateral and axial directions and the confirmation experiment are factor (VIF) and tolerance are used. As a matter of fact, the recipro-
expressed as: cal of the VIF is called the tolerance. When the value of VIF or tol-
   erance is equal to 1, one can conclude that the independent
  1 1
CIAx RMS ; Ay RMS ¼ Fa;1;dofðeÞ :Va þ ð6Þ variables do not have any linear correlation. On the other hand, if
N R
the value of VIF is greater than 10 or the value of tolerance is smal-
n ler than 0.1, the MLR model can be held to have significant multi-
N ¼ ð7Þ collinearity in the regression model [36–38].
1 þ tdof
In the present investigation, MLR technique was applied to
Here, Fa,1,dof(e) = F ratio is at 95% confidence, a = significance determine relationship between dependent variables, Ax RMS and
level, dof(e) = degree of freedom error, Va = variance error, Ay RMS as well as independent variables Rp, Ap, Pp, and Rs. For
N = effective number of replication, R = validation experiment the models, i.e., Ax RMS and Ay RMS the following equations were
replication number, n = total experiment number and tdof = total formulated using SPSS software.
main degree of freedom.
By substituting Fa,1,dof(e) = 4.41 (from standard F table), Ax RMS ¼ 0:268 þ ð8:444E  0:005ÞRp þ 0:005Ap  0:001Pp
Vax = 0.000066 and Vax = 0.000004 (from Tables 9 and 10), R = 3,  0:001Rs ð9Þ
n = 27 and tdof = 8, we get N = 3, CI AxRMS = ±0.0133 and CI
AyRMS = ±0.0034. As per Eqs. (9) and (10), the predicted optimal
vibration along the lateral and axial directions was calculated to Ay RMS ¼ 0:562 þ 0:001Rp þ 0:010Ap  0:001Pp  0:001Rs ð10Þ
be 0.0274 ± 0.000194 m/s2 and 0.0280 ± 0.000012 m/s2, respec-
where, Rp, Ap, Pp, and Rs denote operational parameters of drill
tively, at confidence level of 95%.
machine causing vibration during rock drilling.
In the referred two models, VIF values of independent variables
4.4. Model development
can be noted to be equal to 1. As a result, one can conclude that
these models are free from multi-collinearity. Fig. 13 depicts coef-
Multi-linear regression (MLR) and artificial neural network
ficients of determination (R2).
(ANN) are statistical methods, mostly used for prediction and fore-
The ANOVA of the regression models for dependent factors
casting in engineering applications. In the present study, the MLR
gives p-value of 0.00, which means these models are statistically
and ANN have been used for development of an empirical model.
significant.

4.4.1. Multi-linear regressions (MLR)

MLR is a statistical process by applying which one can study 4.4.2. Artificial neural network (ANN)
relationship between two or more dependent and independent The ANN is computational algorithm resembling the biological
variables. For the purpose, one has to fit a linear equation to the neural network, which is used to establish theoretical relation
data, obtained from the study. The MLR technique is also termed between independent and dependent variables, based on given
as the least-squares model because with this process a line is fitted experimental data. Nowadays, ANN is widely used in both indus-
to data points by means of minimum sum of squares of difference trial and research labs. In recent years, it has gained wide popular-
between the value determined during the investigation and the ity for vibration related investigations as well. Vibration and noise
value already predicted. By means of equation, it is expressed as: control, impact of independent parameters on vibration, etc., are
some of the aspects which are studied with the help of ANN. The
Y ¼ b0 þ b1X1 þ :::::bnXn þ e ð8Þ
application of ANN in vibration related issues helps to identify
where, Y denotes dependent variable, X1; X2;. . .; Xn represent inde- output-based controlling parameters, measured in time domain
pendent variables, or frequency domain [39].

Fig. 13. Relationship between the predicted and the measured value of vibration.
 R. Kumar et al. / Measurement 145 (2019) 803–819 815

A neural network involves three layers; input layer, output validation and testing purposes also increases. With this incre-
layer and hidden layer connected to one another [40–42]. The four ment, an improved regression value can be obtained. For this
neurons, i.e., rotational pressure (Rp) parameter, Air pressure (Ap) study, the regression coefficient is 0.99 for a dataset of 70%, 15%,
parameter, Pull-down pressure (Pp), and rotational speed (Rs) rep- and 15%, respectively. There is therefore no further modification
resent input layer while vibration along axial and lateral direction of the percentage of increased data set. Previous researchers [48–
represents output layer. 54] have proposed equations to determine the number of hidden
nodes, as shown below:
4.4.2.1. Artificial neural network (ANN): training, testing and valida-  2
Ni þ 1 ð11Þ
tion. Numerous ANN algorithms, for instance, radial basis function
[43], Elman recurrent [44] and Hopfield neural networks [45] have
ðNi þ No Þ
been proposed. However, back-propagation algorithm [46,47] is ð12Þ
2
the most popular because it is simple, accurate, and faster. In the
present work also, back-propagation algorithm has been applied.
In this method, the available data are arbitrarily distributed among 2 þ ðNo
Ni Þ þ ð0:5No
ðN2o þ Ni Þ  3
ð13Þ
three groups for training, testing and validation of the model. ðNi þ No Þ
The back propagation algorithm comprises error output, i.e.,
actual node output errors. It is shared between input layer and hid- 2N i
ð14Þ
den neurons. This method enables quick convergence in network 3
on the basis of satisfactory local minima for errors. In order to
pffiffiffiffiffiffiffiffiffi
improve functioning of this network, the number of hidden layers N i
N o ð15Þ
is enhanced for conduct of training. One is allowed a maximum
number of 1000 epochs (iteration) for carrying out training. In 2N i ð16Þ
order to resolve the problem of over-fitting, early stopping is pos-
sible. In fact, early stopping along with a higher number of training where, N i = Number of input parameters; No = Number of output
data with cross validation ensure a higher degree of accuracy. parameter. Based on the proposed equations the upper range for
Three subsets of data, i.e., 70%, 15%, 15% were created respec- the hidden nodes is 9. During the study, the number of hidden
tively for the purpose of training, validation and testing [47]. If nodes in hidden layers at the initial stage was kept at one and
the percentage of dataset chosen increases, the dataset used for increased continuously up to nine node. ANN model with hidden

Fig. 14. Overall ANN Predictive performances for RMS Ax and RMS Ay.
816 R. Kumar et al. / Measurement 145 (2019) 803–819

node (Nh) = 6 showed the best performance among all ANN models presented in the table below, is less than 20%. Hence, they can be
with hidden node (Nh) ranging between 1 and 9 [55]. considered to be within the acceptable range.
Hence, an architecture of (4
6
2) was selected and intro-
duced for solving the drill vibration problem by using ANN. Here, 4.4.4. Comparison of the MLR and ANN models
4 = number of Input, 6 = number of hidden neuron, 2 = number of The measured verses predicted values of vibration in terms RMS
output [56–61]. The selected ANN model consisted of vibration acceleration along the lateral and axial directions were obtained by
as an output in terms of RMS acceleration, measured at lateral using the ANN model. The determination coefficients (R2) for ANN
(X-axis) and axial (Y-axis) directions, respectively. Effect of input model 0.99 for RMS Ax and RMS Ay,. Thus, the predicted values are
parameters, i.e., Rp, Ap, Pp, and Rs were measured. These opera- consistent with the measured values of RMS acceleration in the lat-
tional parameters were found to affect the overall vibration of eral and axial directions. The MLR values, have R2 as 0.79 for RMS
the drill string. As the bit gradually penetrated into rock, character- Ax and 0.92 for RMS Ay. The R2, obtained by the MLR method,
istics of vibration changed according to rock hardness and operat- shows greater deviation from the experimental measurements.
ing parameters of the drill machine. The overall predictive The ANN R2 value has higher correlation in the prediction of vibra-
performance of ANN is presented in Fig. 14 with coefficient of tion as compared to the MLR prediction model. These results indi-
determination 0.991. The root mean square value (RMSE) and vari- cate that the ANN method as compared to the MLR method provide
ance account for (VAF) [62–66], for the ANN model was computed more sensible and more precise vibration response predictions by
using Eqs. (17) and (18), and were obtained as RMSE = 0.004 and using the operating parameters of drilling, as shown in Fig. 15.
VAF = 99.137.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4.4.4.1. Optimization of ANN model using genetic algorithm (GA). The
u
u1 X N genetic algorithm is a technique of artificial intelligence used to
RMSE ¼ t ðy  y1 Þ2 ð17Þ solve optimization tasks according to the theory of ‘‘survival of
N i¼1 o
the strongest” and ‘‘natural selection” proposed by Charles Darwin.
The genetic algorithm repeatedly changes the population of indi-
varðyo  y1 Þ vidual solutions. The genetic algorithm randomly selects individu-
VAF ¼ ½1  
100 ð18Þ
varðyo Þ als from the current population and uses them to produce next-
generation children. Throughout generations the population
where yo = measured value; y1 = predicted value; N = total number ‘‘evolves” to an optimal solution. The genetic algorithm uses three
of data basic types of rules at each stage to create the next generation of
the current population [63–65]:
4.4.3. Confirmation test
Table 13 shows comparison of the experimental values and the  The selection rules select the individuals, called parents, which
predicted values, obtained by means of the Taguchi method, MLR contribute to the population of the next generation.
and ANN. The error value for the reliable statistical analysis is less  The crossover rules involve two parents to train children for the
than 20% [62]. The error in the experimental and predicted values, next generation.

Table 13
Experimental and predicted values comparison.

Taguchi prediction MLR prediction ANN prediction

experimental predicted error% experimental predicted error% experimental predicted error%
AxRMS (m/s2)
0.062 0.027 0.556 0.062 0.058 0.055 0.062 0.061 0.011
AyRMS (m/s2)
0.113 0.028 0.752 0.113 0.112 0.011 0.113 0.112 0.007

In this table the bold value signifies the percentage error value between experimental and predicted value along X-axis and Y axis for all three methods.

Fig. 15. Comparison of Acceleration values obtained from experimental, MLR and ANN.
 R. Kumar et al. / Measurement 145 (2019) 803–819 817

Fig. 16. A flow chart diagram for integrated ANN-GA approach.

 Mutation rules apply random changes to individual parents to comprises of: population size = 18, crossover = scattered crossover,
train their children. mutation rate = 0.1, probability of crossover = 0.8 and the number
of iteration over which GA evolved = 116. The GA process to obtain
A genetic algorithm is widely being used to optimize complex the optimal solution was restricted within the bounds of experi-
engineering problems where traditional methods have limited mental range. Fig. 17 show the Pareto front graph between the
scope. In recent trend many integrated artificial intelligence model two objective i.e Ax RMS and Ay RMS. The optimum value obtained
have been developed for solving optimization problem [66–69]. was 0.057 m/s2 for Ax RMS and 0.026 m/s2 for Ay RMS respectively
ANN despite all advantages has a demerit of slow convergence at the operating condition of Rp of 1223.76 kg/cm2, Ap of 51.84 kg/
and of being stuck in local minima. In this study the GA is cm2, Pp of 58.96 kg/cm2, Rs of 83 rpm.
employed to the input parameters of the ANN model to minimize
the machine vibration. Fig. 16 shows the flow chart diagram for the 5. Conclusion
integrated ANN-GA approach for this study. The GA parameters
A systematic investigation was conducted on-field to analyze
the effects of drill operating parameters, Rp, Ap, Pp, and Rs, on vibra-
tion of drilling components. The study also focused on measure-
ment of vibration. Statistical methods were used to ascertain the
optimum level of vibration and to develop empirical relations for
optimizing vibration.
The highlights, drawn from the study, are as follows:

 The vibration experienced in the axial direction during experi-

mentation was 0.113 m/s2, and in lateral direction, it was
0.062 m/s2. The axial vibration was found to be greater com-
pared to the lateral vibration. Presumably, vibration in the axial
direction was greater due to initial instability of the drill string
component, rock bit interaction and inefficient control of the
operating parameters while drilling.
 The following are the underlining points from the Taguchi L27
orthogonal array, applied to carry out tests:
1. The optimum value of parameters for drilling operation for
reduction of vibration was obtained by means of mean S/N
ratio. The resulting combination of parameters for genera-
tion of minimum vibration during rock drilling was found
Fig. 17. Pareto front plot of GA using ANN function. at Rp of 1200 kg/cm2 (level 1), Ap of 51.33 kg/cm2 (level 1),
818 R. Kumar et al. / Measurement 145 (2019) 803–819

Pp of 52 kg/cm2 (level 2) and Rs of 90 rpm (level 3) along [16] R. Shirani Faradonbeh et al., Prediction of ground vibration due to quarry
blasting based on gene expression programming: a new model for peak
the lateral direction. The levels along the axial direction
particle velocity prediction, Int. J. Environ. Sci. Technol. 13 (6) (2016) 1453–
were at Rp of 1200 kg/cm2 (level 1), Ap of 51.33 kg/cm2 1464.
(level 1), Pp of 59 kg/cm2 (level 3) and Rs of 85 rpm (level [17] M. Monjezi, M. Ghafurikalajahi, A. Bahrami, Prediction of blast-induced ground
2). vibration using artificial neural networks, Tunnelling Underground Space
Technol. 26 (1) (2011) 46–50.
2. The statistical analysis done with ANOVA revealed that Rp, [18] Mahdi Hasanipanah et al., Forecasting blast-induced ground vibration
Ap, Pp, and Rs were the major parameters having signifi- developing a CART model, Eng. Comput. 33 (2) (2017) 307–316.
cant role in generation of vibration in blast hole drilling [19] Danial Jahed Armaghani et al., Neuro-fuzzy technique to predict air-
overpressure induced by blasting, Arabian J. Geosci. 8 (12) (2015) 10937–
machine. Rp was the most influential factor, having 10950.
67.37% (lateral direction) and 76.84% (axial direction), fol- [20] Edy Tonnizam Mohamad et al., Prediction of flyrock in boulder blasting using
lowed by Ap as 21.04% (lateral direction) and 14.04% (axial artificial neural network, Electron. J. Geotech. Eng. 17 (2012) 2585–2595.
[21] Danial Jahed Armaghani et al., Development of hybrid intelligent models for
direction). predicting TBM penetration rate in hard rock condition, Tunn. Undergr. Space
 The Taguchi method, MLR and ANN were applied to predict the Technol. 63 (2017) 29–43.
optimum value. The MLR and ANN were found to be more effec- [22] Mohsen Akhani et al., A hybrid computational intelligence approach to predict
spectral acceleration, Measurement 138 (2019) 578–589.
tive for predicting vibration response with the minimum error, [23] Mahdi Hasanipanah et al., Feasibility of PSO-ANN model for predicting surface
thereby establishing significant relationship between operating settlement caused by tunneling, Eng. Comput. 32 (4) (2016) 705–715.
parameters. [24] Maryam Amiri et al., A new combination of artificial neural network and K-
nearest neighbors models to predict blast-induced ground vibration and air-
 Comparison of empirical models were done for predicting vibra-
overpressure, Eng. Comput. 32 (4) (2016) 631–644.
tion in terms of RMS acceleration, with the MLR and the ANN [25] Masoud Monjezi, Mahdi Hasanipanah, Manoj Khandelwal, Evaluation
techniques. The ANN empirical model offered more precise and prediction of blast-induced ground vibration at Shur River Dam, Iran, by
and less non-conformity as compared to the MLR model. artificial neural network, Neural Comput. Appl. 22 (7-8) (2013)
1637–1643.
 ANN model was further optimized using GA to obtain the optimal [26] Khalil Taheri et al., A hybrid artificial bee colony algorithm-artificial neural
value of 0.057 m/s2 for Ax RMS and 0.026 m/s2 for Ay RMS respec- network for forecasting the blast-produced ground vibration, Eng. Comput. 33
tively, at the operating condition of Rp of 1223.76 kg/cm2, Ap of (3) (2017) 689–700.
[27] Danial Jahed Armaghani, Mahdi Hasanipanah, Edy Tonnizam Mohamad, A
51.84 kg/cm2, Pp of 58.96 kg/cm2, Rs of 83 rpm. combination of the ICA-ANN model to predict air-overpressure resulting from
blasting, Eng. Comput. 32 (1) (2016) 155–171.
[28] A. Prakash, V.M.S.R. Murthy, K.B. Singh, Production evaluation of surface miner
from rock/rock mass properties: some investigations, J. Mines Metals Fuels
Acknowledgements (2012) 244–248.
[29] A. Prakash, V.M.S.R. Murthy, K.B. Singh, Performance simulation of surface
The field studies were conducted in Nigahi and Dudichuwa miners with varied machine parameters and rock conditions: some
investigations, J. Geol. Min. Res. 5 (1) (2013) 12–22.
opencast projects of NCL, Singrauli. Experimental investigations [30] Claude E. Aboujoude, Thesis on Feedback Control of Vibration in Surface Rotary
were carried out in Advanced Research Lab and Rock Excavation Blasthole Drilling, April 1997.
Lab of IIT(ISM) Dhanbad. Authors acknowledge the management [31] V. Wowk, Machinery Vibration: Measurement and Analysis, McGraw-Hill, Inc.,
United States of America, 1991.
of NCL for needful permissions and support. [32] E.M. Anawa, A.G. Olabi, Optimization of tensile strength of ferritic/austenitic
laser-welded components, Opt. Lasers Eng. 46 (2008) 571–577.
[33] S. Pandiaraja, S. SenthilKumaran, L.A. Kumaraswamidhas, R. Saravanan,
References
Interfacial microstructure and optimization of friction welding by Taguchi
and ANOVA method on SA 213 tube to SA 387 tube plate without backing
[1] Omid Saeidi, Seyed Rahman Torabi, Moham Adataei, Jamal Rostami, A block using an external tool, J. Alloy. Compd. 654 (2016) 534–545.
stochastic penetration rate model for rotary drilling in surface mines, Int. J. [34] S. Kannan, S. SenthilKumaran, L.A. Kumaraswamidhas, An Investigation on
Rock Mech. Min. Sci. 68 (2014) 55–65. mechanical property of commercial copper tube to aluminium 2025 tube plate
[2] J.R. Bailey, E. Biedigner, S. Sundararaman, A.D. Carson, W.C. Elks, F.E. Dupriest, by FWTPET process, J. Alloy. Compd. 672 (2016) 674–688.
Development and application of BHA vibrations model, Proceedings of the [35] Ilhan Asiltürk, Süleyman Nesßeli, Multi response optimisation of CNC turning
International Petroleum Technology Conference, 3-5 December 2008. Kuala parameters via Taguchi method-based response surface analysis,
Lumpur, Malaysia, 2008. Measurement 45 (4) (2012) 785–794.
[3] S. Jardine, D. Malone, M. Sheppard, Putting dampers on drilling’s bad [36] A. Giacomino, O. Abollino, M. Malandrino, E. Mentasti, The role of
vibrations, Oilfield Rev. 6 (1) (1994) 15–20. chemometrics in single and sequential extraction assays: a review. Part II.
[4] A.P. Christoforou, A.S. Yigit, Active control of stick-slip vibrations: the role of Cluster analysis, multiple linear regressions, mixture resolution, experimental
fully coupled dynamics, Proceedings of the SPE Middle East Oil Show, 17-20 design and other techniques, Anal. Chim. Acta 688 (2011) 122–139.
March 2001, SPE no. 68093. Manama, Bahrain, 2001. [37] N.G. Yilmaz, M. Yurdakul, R.M. Goktan, Prediction of radial bit cutting force in
[5] Edward A. Branscombe, Thesis on ‘‘Investigation of Vibration Related Signals high-strength rocks using multiple linear regression analysis, Int. J. Rock Mech.
for Monitoring of Large Open-Pit Electric Blasthole Drills” 2010. Min. Sci.. 44 (2007) 962–970.
[6] H. Liu, Karen Yin, Analysis and interpretation of monitored rotary blasthole [38] H. Jang, E. Topal, Optimizing overbreak prediction based on geological
drill data, Int. J. Surface Min. Reclam. Environ. 15 (3) (2001) 177–203. parameters comparing multiple regression analysis and artificial neural
[7] D.W. Dareing, Billy Joe Livesay, Longitudinal and angular drill-string vibrations network, Tunn. Undergr. Space Technol. 38 (2013) 161–169.
with damping, J. Eng. Industry 90 (4) (1968) 671–679. [39] Md Sazzad Hossain, Ong Zhi Chao, Zubaidah Ismail, Siamak Noroozi, Shin Yee
[8] J.K. Vandiver, J.W. Nichoison, Rong-Juin Shyu, Case studies of the bending Khoo, Artificial neural networks for vibration based inverse parametric
vibration and whirling motion of drill collars, Soc. Pet. Eng. 5 (4) (1990) 282– identifications: a review, Appl. Soft Comput. 52 (2017) 203–219.
290. [40] M. Conforti, S. Pascal, G. Robustelli, F. Sdao, Evaluation of prediction capability
[9] J.D. Jansen, Non-linear rotor dynamic as applied to oil well drill string of the artificial neural networks for mapping landslide susceptibility in the
vibrations, J. Sound Vib. 147 (1991) 115–135. Turbolo River catchment (northern Calabria, Italy), Catena 113 (2014) 236–
[10] J.D. Jansen, L. Vanden, Steen, Erik Zachariasen, Active damping of torsional 250.
drill string vibrations with a hydraulic top drive, SPE Drill. Complet. (1995) [41] Wenyan Wu, Graeme C. Dandy, Holger R. Maier, Protocol for developing ANN
250–254. models and its application to the assessment of the quality of the ANN model
[11] A.S. Yigit, A.P. Christoforou, Coupled torsional and bending vibration of drill development process in drinking water quality modelling, Environ. Modell.
strings subject to impact with friction, J. Sound Vib. 215 (1) (1998) 167–181. Software 54 (2014) 108–127.
[12] N. Challamel, Rock destruction effect on the stability of a drilling structure, J. [42] J. Parinet, M. Julien, P. Nun, Richard J. Robins, G. Remaud, P. Hohener,
Sound Vib. 233 (2) (2000) 235–254. Predicting equilibrium vapor pressure isotope effects by using artificial neural
[13] Sigve Hovda, Semi-analytical model of the axial movements of an oil-well networks or multi-linear regression-A quantitative structure property
drillstring in vertical wellbores, J. Sound Vib. 417 (2018) 227–244. relationship approach, Chemosphere 134 (2015) 521–527.
[14] Sigve Hovda, Semi-analytical model of the axial movements of an oil-well [43] S.R. Shadizadeh, F. Karimi, M. Zoveidavianpoor, Drilling stuck pipe prediction
drillstring in deviated wellbores, J. Sound Vib. 433 (2018) 287–298. in iranian oil fields: an artificial neural network approach, Iranian J. Chem. Eng.
[15] Sigve Hovda, Automatic detection of abnormal torque while reaming, J. Petrol. 7 (4) (2010), IAChE.
Sci. Eng. 166 (2018) 13–24.
 R. Kumar et al. / Measurement 145 (2019) 803–819 819

[44] R. Rakkiyappan, P. Balasubramaniam, Delay-dependent asymptotic stability [57] Roohollah Shirani Faradonbeh et al., Prediction and minimization of blast-
for stochastic delayed recurrent neural networks with time varying delays, induced flyrock using gene expression programming and firefly algorithm,
Appl. Math. Comput. 198 (2) (2008) 526–533. Neural Comput. Appl. 29 (6) (2018) 269–281.
[45] M.Q. Nguyen, P.M. Atkinson, H.G. Lewis, Super-resolution mapping using [58] Danial Jahed Armaghani et al., Prediction of the strength and elasticity
Hopfield neural network with fused images, IEEE Trans. Geosci. Remote Sens. modulus of granite through an expert artificial neural network, Arabian J.
44 (3) (2006) 736–749. Geosci. 9 (1) (2016) 48.
[46] K. Werea, D.T. Buic, Qystein B. Dick, Bal Ram Singh, A comparative assessment [59] E. Tonnizam Mohamad et al., Rock strength estimation: a PSO-based BP
of support vector regression, artificial neural networks, and random forests for approach, Neural Comput. Appl. 30 (5) (2018) 1635–1646.
predicting and mapping soil organic carbon stocks across an Afromontane [60] Mohammadreza Koopialipoor et al., The use of new intelligent techniques in
landscape, Ecol. Indic. 52 (2015) 394–403. designing retaining walls, Eng. Comput. (2019) 1–12.
[47] E. Ali, Wu Guang, A. Ibrahim, Empirical relations between compressive [61] Danial Jahed Armaghani et al., Application of two intelligent systems in
strength and micro-fabric properties of amphibolites using multivariate predicting environmental impacts of quarry blasting, Arabian J. Geosci. 8 (11)
regression, fuzzy inference and neural networks: a comparative study, Eng. (2015) 9647–9665.
Geol. 183 (2014) 230–240. [62] M.H. Cetin, B. Ozcelik, E. Kuram, E. Demirabas, Evaluation of vegetable based
[48] Hecht-Nielsen Robert, Kolmogorov’s mapping neural network existence cutting fluids with extreme pressure and cutting parameters in turning of AISI
theorem, Proceedings of the IEEE International Conference on Neural 304L by Taguchi method, J. Cleaner Prod. 19 (17–18) (2011) 2049–2056.
Networks III, IEEE Press, 1987. [63] Edy Tonnizam Mohamad et al., An optimized ANN model based on genetic
[49] Brian D. Ripley, Statistical aspects of neural networks. Networks and chaos— algorithm for predicting ripping production, Neural Comput. Appl. 28 (1)
statistical and probabilistic aspects, 50 (1993), 40–123. (2017) 393–406.
[50] J.D. Paola, Neural Network Classification of Multispectral Imagery Diss. MSc [64] Mohammadreza Koopialipoor et al., Three hybrid intelligent models in
thesis, The University of Arizona, USA, 1994. estimating flyrock distance resulting from blasting, Eng. Comput. 35 (1)
[51] Changfeng Wang, A theory of generalization in learning machines with neural (2019) 243–256.
network applications‘‘, 1994. [65] Hui Chen et al., Assessing dynamic conditions of the retaining wall: developing
[52] Timothy Masters, Practical Neural Network Recipes in C++, Morgan Kaufmann, two hybrid intelligent models, Appl. Sci. 9 (6) (2019) 1042.
1993. [66] D. Jahed Armaghani et al., Blasting-induced flyrock and ground vibration
[53] Iebeling Kaastra, Milton Boyd, Designing a neural network for forecasting prediction through an expert artificial neural network based on particle swarm
financial and economic time series, Neurocomputing 10 (3) (1996) 215–236. optimization, Arabian J. Geosci. 7 (12) (2014) 5383–5396.
[54] Ioannis Kanellopoulos, Graeme G. Wilkinson, Strategies and best practice for [67] Mohsen Hajihassani et al., Ground vibration prediction in quarry blasting
neural network image classification, Int. J. Remote Sens. 18 (4) (1997) 711– through an artificial neural network optimized by imperialist competitive
725. algorithm, Bull. Eng. Geol. Environ. 74 (3) (2015) 873–886.
[55] H. Sonmez et al., Estimation of rock modulus: for intact rocks with an artificial [68] Mohsen Hajihassani et al., Blast-induced air and ground vibration prediction: a
neural network and for rock masses with a new empirical equation, Int. J. Rock particle swarm optimization-based artificial neural network approach,
Mech. Min. Sci. 43 (2) (2006) 224–235. Environ. Earth Sci. 74 (4) (2015) 2799–2817.
[56] Hossein Moayedi, Danial Jahed Armaghani, Optimizing an ANN model with [69] E. Momeni et al., Prediction of pile bearing capacity using a hybrid genetic
ICA for estimating bearing capacity of driven pile in cohesionless soil, Eng. algorithm-based ANN, Measurement 57 (2014) 122–131.
Comput. 34 (2) (2018) 347–356.