Journal of Petroleum Science and Engineering 167 (2018) 300—315

Check
for
updates

Artificial intelligence techniques and their applications in drilling fluid

engineering: A review
Okorie E. Agwua'*, Julius U. Akpabioa, Sunday B. Alabia, Adewale Dosunmub
a
Department of Chemical and Petroleum Engineering, Faculty of Engineering, University of [Iyo, P.M.B. 1017, [Iyo, Nigeria
Department of Petroleum Engineering University of Port Harcourt, Rivers State, Nigeria

ARTICLE IN FO AB STRA CT
Keywords: For an oil well to be said to have been successfully and conclusively drilled, the drilling fluid lies at the heart
Drilling fluid of the solution. Therefore, the guarantee to solving issues in oil well drilling is to contrive an optimal drilling
Artificial neural network fluid. However, there is usually a complex interplay of factors involved during drilling fluid formulation,
Fuzzy logic
Case based reasoning property determination, its performance in the well and its relationship with other wellbore drilling
Hybrid intelligent systems parameters. This is so because drilling muds exhibit time dependent properties. This time dependency is the
Genetic algorithm direct product of the synergy among the various active additives that make up the mud and the characteristic
Particle swarm algorithm of each additive especially at downhole conditions where the effects of temperature and pressure are well
pronounced. These additives are more often than not diverse in size, chemical activity, density and surface
energy. Deriving knowledge from the data from these parameters in order to develop a functional relationship
between them is a challenging task requiring advanced modelling techniques as well as human intuition and
experience. The dependence on human intuition and on the experiential knowledge of professional mud
engineers lays bare the shortcomings of traditional mud design techniques. Artificial intelligence techniques
have been shown to alleviate this challenge. Exploiting the abundant literature on the various applications of
artificial intelligence in oil and gas operations, several works that show how and what artificial intelligence
techniques are used in the drilling fluid industry, and what have been achieved due to their use have been
selected. In this paper, a review of existing artificial intelligence techniques and their applications in drilling
fluid engineering is given. This paper also dug up and analyzed the strengths and pitfalls of each artificial
intelligence technique. The examination of the strengths and deficiencies was done using the following virtues
as the basic criteria: robustness against noise, self-organization, generalization ability, data volume
requirements and the convergence speed. The artificial intelligence techniques presented in this paper include:
artificial neural networks (ANNs), fuzzy logic, support vector machines (SVM), hybrid intelligent systems
(HIS), genetic algorithms (GA), case based reasoning (CBR) and particle swarm algorithm (PSA). An
overview of the applications of classical artificial intelligence in drilling fluid engineering is also presented.
From the review, it was gathered that the ANN technique is the most widely applied in drilling fluid
engineering accounting for over 54% of the papers reviewed; while lost circulation problem was the most
predicted well problem related to drilling fluids accounting for over 17% of the mud problems predicted. It
was also observed that a blend of Al techniques performed better than when each one of the Al techniques was
used singly. Finally, judging the Al techniques on the criteria mentioned above, ANN was found to meet all
the listed criteria except for its slow speed of convergence while ANN, GA, SVM and fuzzy logic were all
found to be robust against noise.

1. Introduction leaves the definition of Al as a knotty problem. Little

wonder Schatsky et al. (2014) opines that the field of Al
One of the most important advances in the scientific suffers from two extreme situations namely-too little
community that has appealed to virtually all fields of and too many definitions. To further buttress this point,
human endeavour is the concept of artificial intelligence Nilsson (2010), stated that Al lacks a universally
(Al). In defining the term artificial intelligence, Lea accepted definition. Whereas attempts to
(2015) is of the opinion that "artificial" is very easy to et al.
define. According to him, artificial can simply mean
having no occurrence in real life or not having the same
give an all-encompassing definition to the concept of Al
configuration in nature.
are hindered by explosive complexity, these
However, the definition of intelligence is wide and
complexities do not erode the fact that Al holds great
varied (Legg and Hutter, 2007). As a result, this then
O.E. Agwu
promise as tools for understanding the relationship error in the laboratory; with the initial guesses
between complex entities. However, this work would depending largely on the experience of the mud engineer
align itself with the definition of Al put forward by or the analyst in the laboratory. However, as Abdollahi
McCarthy (2007), wherein he defines Al as the science (2007) put it, the knowledge gained through experience
and engineering of developing computer programs could easily be lost when the laboratory analyst or the
which when fed into a machine makes the machine mud engineer who possesses the requisite experience
exhibit the intelligence of humans. Al is also referred to leave the company. Therefore, there is need for a system
as "soft computing", "virtual intelligence", and which would make use of all the data available to map
"computational intelligence" (Mohaghegh, 2000). out trends and ultimately offer better understanding of
According to Rabie (2017), the measurable values of Al drilling mud design (Shadravan et al., 2015). One area
include: the possibility of making machines solve that comes handy is artificial intelligence. The benefits
difficult problems at a faster speed than what the human of Al techniques in the present circumstance are
brain would take to solve the same problem; analysis of numerous. They include but are not limited to: ability to
big data in order to understand trends and make predict situations with complex relationships between
predictions of future scenarios in the best possible input and output
manner with the least waste of time and effort; and the Journal of Petroleum Science and Engineering 167 (2018) 300—315
consequence of which is huge savings in operating cost.
From the available literature, virtually all parts of the oil parameters, it allows finding the best values of model
and gas exploration and production chain have applied properties minimizing calculation errors in a given area
one artificial intelligence technique or another in the etc. Of course, there are phenomenal Al researches in
course of their operations. For instance, Field wellbore drilling as well as in drilling fluid engineering.
development: Doraisamy et al. (1998); Wellbore drilling As a result of this, many textbooks and scholarly papers
process: (Valipour et al., 2014; Skalle et al., 2000); have been published in both local and international
Drilling platform selection: (Wang et al., 2011); Well journals that address the subject of using artificial
placement: (Naderi and Khamehchi, 2017; Isebor et al., intelligence techniques in oilfield applications. Most of
2014); Reservoir properties determination: (Lim, 2005; these publications give qualitative and quantitative
Baarimah et al., 2015; Ho and Ehara, 2007); Sand studies on what has been done in using them and the
production prediction: (Khamehchi et al., 2014); Gas-lift successes recorded. However, due to the elaborate
performance prediction and long term gas-lift allocation nature of the studies that have been presented by diverse
optimization: (Rasouli et al., 2015); History matching: researchers, there is need to assess all the studies and the
(Firoozjaee and Khamehchi, 2014; Zaimy and Rasaei, research findings in one piece so as to isolate what has
2013); Two-phase flow in pipes: (Ternyik et al., 1995a, been achieved and identify possible areas of interest
1995b); Identification of well test interpretation models: where more studies should be directed. To achieve the
(Sung et al., 1995); Formation damage prediction: objective of digging deep into the application of Al in
(Nikravesh et al., 1996); Completion analysis: (Shelley mud engineering, this paper makes a chronicle of the
et al., 1998). The list still goes on and on. However, in available studies on the application of artificial
the oil and gas industry, drilling operations have been intelligence in drilling fluid engineering. The various
one of the most troublesome parts wherein various research findings as presented by diverse authors are
problems such as wellbore instability, lost circulation, highlighted and the future research needs on the subject
kicks and blowouts and stuck pipe may be encountered are presented. However, there is a limit to what can be
(Meschi et al., 2010). One major component of the oil accomplished in just one article, hence, considering the
well drilling operation which is often referred to as "the full list of all Al techniques and their applications in
blood of the drilling process" is the drilling fluid. The drilling fluid engineering in one review paper is in itself
drilling fluid is directly or indirectly related to the not wrong but it is deemed that their inclusion in this
problems or solutions of the problems enumerated work has an overloading effect on its contents.
above. For instance, increasing mud weight controls Therefore the artificial intelligence applications in
wellbore pressures and when it is too high, it fractures drilling fluid engineering was narrowed down to
the formation; and too low a mud weight leads to kick artificial neural networks (ANNs), fuzzy logic, support
and in extreme cases-blowouts. Little wonder Agwu et vector machines (SVM), hybrid intelligent systems
al. (2015) puts it that the drilling fluid is the "architect" (HIS), genetic algorithms, case based reasoning (CBR)
which can make the drilling operation either materialize and particle swarm algorithm (PSA). This paper is
or unrealizable. On the cost implication axis, Bloys et al. organized as follows: first, a highlight on the evolution
(1994) reports that drilling muds represent 50/0—150/0 of artificial intelligence in the exploration and
of drilling costs but may cause 100% of drilling production (E & P) industry is made, an expatiation on
problems. One way of preventing drilling problems is the applications of artificial intelligence in mud
optimal selection of drilling fluids, which is almost engineering follows, an overview of the strengths and
always based on the analysis of data obtained from weaknesses of artificial intelligence techniques follows
previously drilled wells in the areas. A typical drilling thereafter. Selected applications of classical artificial
mud programme provides information on the type of intelligence and molecular simulations for various
mud, mud constituents, mud rheology, mud density, drilling fluid related challenges are presented thereafter.
filtration and other mud properties as well as the mud This is closely followed by a discussion of the learnings
maintenance requirements (Schlumberger, 2017). from the review. This paper climaxes with some
However, in order to design drilling fluids to meet a concluding remarks and suggestions for further research
desired theology which would suit specific formation are made.
properties, combat wellbore challenges and prevent fluid
losses all require detailed analysis and decision making.
As Shadravan et al. (2015) put it, designing drilling
muds and other wellbore fluids are all done by trial and

301
2. Evolution of artificial intelligence in the complex phenomenon. For instance, Sheremetov et al.
exploration & production (E & P) industry (2008) reports that a hybrid of expert systems, fuzzy
logic and web services, called SmartDrill, is useful in
Before delving into the history of Al applications in solving the problem of lost circulation. They further
the E & P industry, it is pertinent to look at how the added that the software is at field testing stage in a
concept of Al came to being. iScoop (2016) reports that Mexican oil company. Recently, an Al strategy based on
Al was conceived in the 1950s. According to Alegre supervised learning (MLP neural network with K*) as
(1991 the term artificial intelligence was coined as a reported by Cristofaro et al. (2017) was developed to
result of a confluence of ideas generated during a generalize lost circulation data collected over a period of
conference held on the campus of Dartmouth College, three years from five hundred events pertaining to lost
USA by a group of scientists in the summer of 1956. In circulation. The Al strategy developed with a user
general, the timelines in the development of the Al friendly interface was applied to two wells experiencing
techniques and their developers arranged lost circulation in the Santos Basin in Brazil deep water
chronologically is shown in Table 1. From the table, it is pre salt. This application was essentially able to predict
observed that artificial neural networks were the earliest and recommend treatment procedures for the lost
Al techniques to be developed while hybrid intelligent circulation with the resultant effect of saving six days
systems are the latest. wellbore drilling time.
Narrowing down the history of Al to its application
in the E & P industry reveals that research papers on Al 3. Applications of artificial intelligence in drilling
application in the industry as early as in the 1970s exist mud engineering
(Bravo et al., 2012). However, during the
In this section, a survey of the state-of-the-art
Table 1 researches which involves the applications of Al
Evolutionary trend of various artificial intelligence techniques. techniques in drilling mud engineering is presented.
Al technique Year of development Developer Firstly, the fundamentals of these artificial intelligence
techniques are highlighted; secondly, their applications
Artificial neural network 1943 McCulloch andinPitts
mud engineering are discussed.
Fuzzy logic 1965 Lofti A. Zadeh
Genetic algorithm 1970 John Holland
3.1. Artificial neural network (ANN)
Schank and Abelson
Case based reasoning 1977
Support vector machines 1995 Vapnik, V
Particle swarm algorithm 1995 Artificial neural networks (ANNs)
Eberhart and Kennedy are a form of
Hybrid intelligent systems artificial intelligence that has proved to provide a high
O.E. Agwu et al. level of competency in solving many complex
engineering problems that are beyond the computational
capability of classical mathematics and traditional
early stages of Al development in the industry, Expert procedures (Shahin et al., 1999). For the fundamental
systems were the main Al technique used. Their usage details about ANNs, a good read is Petrus et al. (1995).
was mainly in management of drilling operations and
the treatment of wellbore drilling problems (Courteille et 3.1.1. Cases where ANN were applied in drilling fluid
al., 1986).
engineering
Popa and Cassidy (2012), however reports that Al
techniques have been used in the oil and gas industry Table 2 summarizes the research efforts made on the
since the early 1990s. This view is in sync with the work application of ANNs to mud engineering. Though
of Bello et al. (2015) wherein he asserts that Al tracing the researches from the earliest was difficult, the
techniques were first applied in the E &P industry in summary chronicled available research papers from
1989. 2003 up until 2017. The areas in mud engineering where
In order to consolidate on the gains of using Al and ANN have been applied from the review include: (a) lost
to encourage researches in this area, the Artificial circulation prediction (b) mud density prediction (c)
Intelligence Applications Institute (AIAI) was created in mud rheological properties prediction (d) filtration loss
1983. According to the Institute, AIAI is a technology and mud cake permeability predictions (e) mud flow
transfer organisation that encourages the applications of pattern prediction in wellbore annulus (D hole cleaning
Al research to the advantage of individuals, corporate and
Journal of Petroleum Science and Engineering 167 (2018) 300—315
bodies, industries and the government (Artificial
Intelligence Applications Institute, 2014). This move
triggered the development of several Al based software cuttings transport efficiency of mud (g) settling velocity
packages aimed at solving real life oilfield problems. of cuttings in mud prediction and (h) frictional pressure
Hence, in the drilling aspect of the oil industry, loss prediction. The table highlights the type of study
softwares such as MUDMAN and DrillEdge were conducted by each researcher, the ANN method used,
developed (Barton and Sviokla, 1988; Boman, 2014). the ANN architecture, the input and output parameters
On the one hand, the MUDMAN developed by Baroid as well as the performance evaluation of the network
Fluids Company is built to analyse drilling fluids. On developed in each case (this is represented in terms of
the other hand, the DrillEdge — a case based reasoning the correlation coefficient (R 2), root mean square error
software developed by Verdande Technologies is built (RMSE), average absolute percent relative error
to identify potential drilling problems involving drilling (AAPE), etc). From the summary, the following are
equipment as well as problems relating to the drilled observed:
well. In recent times, E & P companies are actively
integrating various Al techniques to build hybrid
intelligent systems that have the capability of handling
O.E. Agwu
1. Most of the researchers (about 90%) carried out their neural network toolbox in the MATLAB software for
works with multiple inputs and single output except their work. Examples are found in the works of
for the authors Jeirani and Mohebbi (2005); Moazzeni et al. (2010); Rooki et al. (2012, 2014); Al-
Shadravan et al. (2015) and Elkatatny et al., (2016) Baiyat and Heinze (2012); Jahanbakhshi et al. (2014);
whose works involved 2, 7 and 4 outputs, Adesina et al. (2015); Elkatatny (2017) etc.
respectively. This shows that ANN can be used to Additionally, many of these researchers used the back
model parameters with multiple outputs. propagation (BP) algorithm to train the networks.
However, just a handful used the neuro solutions
2. The most widely used performance evaluation criteria software in their work. This is seen in the cases of Miri
is the R2 value. The average R2 value across the et al. (2007) and Siruvuri et al. (2006). From Table 2,
research works reviewed was approximately 0.939 since only two neural network softwares were found to
showing a good model performance across board. be used by the researchers in their studies, it is
3. The aspects of mud engineering that ANN was somewhat inadequate to draw a conclusion on the basis
mostly applied are prediction of stuck pipe and lost of just those two softwares as to which of them is
circulation. These accounted for about 18% and 15% recommended for any of the wellbore challenge listed in
of the researches, respectively. Since lost circulation, this work. However, viewing the matter from the
according to Drilling Specialties Company (Online), standpoint of which of them is the most widely used by
is the most costly mud related problem and stuck pipe majority of the authors in this work, the MATLAB
problems according to Siruvuri et al. (2006) could software tops the recommendation chart. Though the
gulp over 40% of total well costs in certain areas; this authors never gave any reason for choosing the software
could probably be the main driver for numerous they used, the choice by most authors to use the
prediction studies on them using ANN. MATLAB software may not be unconnected with the
expert view of Schultz, a researcher at Halliburton who
4. Most of the authors used data sets extracted from reports that the real results with using the software
open literature to create their network models. include: facilitating ease of simulation, provision of an
accurate, productionstandard algorithm and huge time
Beyond the four areas highlighted above, one area savings (Mathworks, online). However, when using the
worth mentioning is the Al software used by each author neural network tool in MATLAB software, it must be
for estimating the drilling fluid related challenge they appreciated that several learning algorithms exist and the
studied. It must be said that there are numerous ANN choice of the algorithm by the user may likely affect the
softwares for developing applications, and some of them outcome of developed models. For instance, in
are designed for industrial use. The neural network predicting stuck pipe, Miri et al. (2007) and Siruvuri et
toolbox in the MATLAB software is one of these tools al. (2006) used the same software-neuro solutions but
(Bravo et al., 2012). Other softwares include neuro got different results. In this case, Miri et al. (2007) used
solutions, pythia, neuroph, neural designer etc. It is seen MLP and RBF while Siruvuri et al. (2006) used GFNN.
that majority of the researchers in Table 2 used the
302

303
 andEngineering167(2018) 300-375

CIA'Ad 01pues0t'OZ!S01011 adld

Il!lS 8u9 VHH P!IOS'SSOIP!IIIJ
一

96878
名Ⅱ

pntu d 01011'a.mssa.ld
NIdHtId a•IOJaqad!d10 '1B It-Ist-ptequeqer
pupsan.tado.ldpntu
~6181
BU!lD!Pa•Id
SP!t11J
P!t11J
S/W880•0
JournalofPetroleumScience

'Ku
suapP!t11J 'Kl!
suap t—
C1—9
NIdHtId10
PIOOH ad
pntuCIO1121100111.
1P'ad!d 1JO (tto
wanedMOIt:1 OCI'KI!SOJSIA "0!suapP!IIIJ MOIt:I t~9 9 Ndqt:ltl!uunpeIO SKep0M1Il!110!aeIt1D.I!OJO1S0 JOJunotue

一
一
巴

一
'SSOIP!IIIJpntu'pntuJO
P
!
I
O
S
pnttl'axnssa.lddumd

巴
“
'
d
u
m
d

SKEP

Ill
snorna.ld
OMI

304
aulnIOA
J
O
98e19Ae I!CI110MJO
'811110.1011 'aoeJ.tnspunoaBuro dOlUO!IEUI.IOJ!.letuse't10!loas
at-I!II!.IP Id pall!.1P

名
AdVV

ASIAI
H
 300-375

AdVV

O(Idd)
dOMIJ
IIIJaseq
ASIAIH

JournalofPetroleumScience andEngineering167(2018) 300-375

9Z966
986dJOKJ!JOPA

NdHt:It:1 10
d
e
n
a
♂

•
0
W90

((
W90

eqdo•ner)

psa!(l)

a9866d

(A/T)
66d

uoge.nuaouoo

I Kl!l!qeqo.ldad!dNdqt:lt:l(010I-PPEZ!PEHS
>Pt1JS

110!a!pt10Jad!d

MOUJOad.'O'
IPanssa.ld

IP/dP'do-

(600

,/uep.10f

(lJ/!sd)

SSOI

SSOI

a.lnssa.ld

一
0

:
'sap!.101t-p

'aul!l'ssol

.10dtu

P!IIIJ

'anssa.ld

IHdH

'KJ!I!CIE

305
 s
tilnuup•a- IOCIIPM 8a06dMOUP!IIIJ
AdVV

ASIAI
H
300-375

9966d 88Z800•0
lepgaadnsP!nb!l'd()žl'KJIDOIOA SEE)
!IV
666666d
-
(9to
6Z8 0 ISOIJO MOUpntu Idap'anssa.lddumd Nldqtlt:1
「

!U!
OSSOHPUB'IEJpnoutpq ulËOJ P!IIIJ JO (9to
CC6'0 96Z'9
NEIfIH pupPIOOH
'.'O!suap uB!
sappnn(910 MOUp!nb!l'ssans
Pli11J aq H
一

666d 96e8
t~019
SSOIa-Inssa.ld(910 PIOOH
(WHOPIOt1ED)
91m213dE9.L
NIdHtId pntu910tIUMOPBU!lD!Pa•Id
eti!SôPV
snlnpotuBunOA't10!lEJua!.1091m
u
a
s
一

8
u
9
s
O
A!
ss
a.I
d
u
1
0
D
sa
SI
EI
LI
O
Z!
.I
O
H
ut
n
ull
tll
ul
S
O

一
pnu•l
0t
1S0 n

一
巴
(910
OLL8'0
'Kl!llqeatu.tadUO!aeUI- IOJ'Kl!so.10d d 010H

名
1S0
a.Imoe.IJ
SODS
IOS

一
1S0
10 '1B It-Ist-ptequeqer
sprnIJ
06d~98d
96
LIO!SO.IO

307
01
ap!l.led
EL6'() 60a0•0
P!t11J t—CI—L
ITEM JO
Ct76'() 9180•0
"0!suap "0!suap
Ndqt:lt:l
P!IOS SODS A

一
ASIAIH
AdVV
996d 'anssa.lddumdpnu•l'dumdpntuJO91mMOU
K

UOtIEUI.IOJ91e1
S ODS

'dnUOIIEUI.I
OJJOtpdap
illdap
'(111-1011pup

1S0
(þto
ulËOJ
t'16'() 9686'9
t~01~9
10 '1B PIOOH
luao.lad
06d 969>

《
pue91m213dE9J-
969'9 (8t AdVI'N
96eI'8
dlL'Nt:ItI
一

AdVV

ASIAI
H
 300-375

AdVV

ASIAIH

JournalofPetroleumScience andE

luao.lad

866d

86d AdVI'N 96 LASIN

d86 96 9

IICId)

ssa.ns

.1eaqs

'Kusuaa

110!lEtuusa

KIISODS!A

309
 910

O.E. Agwu
(
8tIEOH

et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

Table 3
Summary of researches on mud engineering to which fuzzy logic was applied.

Authors Type of study conducted Input Parameters Output parameters Errors (Performance evaluation criteria)
MSE

Lost circulation problem

Sheremetov et al. (2008) Prediction of pipe
Murillo et al. (2009) sticking mechanism Stuck pipe
Measured depth, TVD, bit flow rate, 10 s and 10 min gels, mud
weight, PV, YP, calcium filtrate, chloride filtrate, torque, circulating condition
Drilling fluid leak pressure, WOB, drag, bit size, BHA, ROP, RPM
Yunhu et al. (2011)
zone identification Formation pressure, well depth, mud density, mud viscosity, mud
Ataallahi and Blowout flow rate
Shadizadeh (2015) Impact, duration and flow rate Consequence
Ahmadi et al. (2016) Mud density prediction
Chhantyal et al.(2017) Mud flow rate Initial mud density, pressure, temperature Density 0.7237 69.0907
prediction Ultrasonic level measurements (LT-I, LT-2 and LT-3) Mud flow rate 0.98 1.74% (MAPE)

3.2. Fuzzy logic around a smooth sphere. They examined the performance of the
models by comparing it with other models. Using the sum of squared
A very good read for the fundamentals and applications of fuzzy of logarithmic deviations as a criterion; they report that the model
logic is found in the work ofRoss (2010). gave approximately 16% and 69% better results than the existing
models.
3.2.1. Cases where fuzzy logic were applied in drilling fluid
engineering 3.5. Support vector machines (SVMs)
Research papers reviewed on the use of fuzzy logic in drilling
fluid engineering indicates that aspects of mud engineering such as The work by Steinwart and Christmann (2008) is a good read on
lost circulation, stuck pipe prediction, mud density prediction and the the subject of support vector machines.
consequences of kick were in the spotlight as seen in Table 3.
3.5.1. Cases where support vector machines were applied in drilling
3.3. Hybrid intelligent systems fluid engineering

3.3.1. Cases where hybrid systems were applied in drilling

fluid engineering For a good overview of this technique, the work
of Medsker (1995) is useful. From Table 4, the application of hybrid
intelligent systems to solving problems in mud engineering is
highlighted. Practical problems tackled with the use of hybrid
systems include: kicks, lost circulation and effects of high
temperature high pressure (HTHP) conditions on mud densities. It is
also noteworthy that the combination of Al techniques to make a
hybrid is such that two techniques with unique characteristics are
chosen. For instance, since the training of ANN have slow speed of
convergence but training of fuzzy intelligent systems (FIS) have high
speed of convergence, their combination to form ANFIS by Deregeh
et al. (2013) would prove to be far more reaching in its performance
than when just one of them is used. In this case, the effects of slow
convergence by ANN would be neutralized by the high speed of
convergence of FIS. Moreover, since PSO and GA are optimization
algorithms, hybrids of PSO and ANN to form PSO-ANN as well as
GA and FIS to form GA-FIS would essentially make their
performance better off than when used singly.

3.4. Genetic algorithms (GA)

3.4.1. Cases where genetic algorithms were applied in drilling fluid

engineering
A very good read on the fundamentals of genetic algorithms can
be found in the review paper of Ab Wahab et al. (2015). A summary
of research papers on the use of genetic algorithms in drilling fluid
engineering indicates that aspects of mud engineering such as
rheological properties determination, particle settling velocity in
drilling mud etc. were in the spotlight as seen in Table 5.
In addition to the above, Barati et al. (2014) used multi gene
genetic programming, a branch of genetic algorithm to develop
empirical relations that can predict the drag coefficient of flow

310
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315
Table 6 highlights the research endeavours of various researchers Case-based reasoning (CBR) as a subset of artificial intelligence
on the application SVMs to solving problems in mud engineering. (Al) solves new problems by reusing the experience of a similar
Practical problems tackled with the use of SVMs include: fluid flow problem that had been solved in the past (Aamodt and Plaza, 1994;
pattern
Table 4
Summary of researches on mud engineering to which hybrid intelligent systems was applied.

Authors Type of study Type of HYBRID Number of Input Parameters Output parameters Errors (Performance evaluation criteria) conducted
data points
MSE

Ghaffari Analysis of mud Self-Organizing 392 Mud Weight, Funnel viscosity, Solid Mud loss pattern

et al. loss patterns Neuro-fuzzy content, Plastic Viscosity, Yield

(2009) Inference system Point, RPM loss (bbl): the rate of
(SONFIS) Mud loss & Pump Pressure
Deregeh Kick detection ANFIS 412 Depth d-exponent 15.9753%
et al.
(2013)
Ahmadi Prediction of mud PSO-ANN 664 Initial mud density, pressure, Density 0.9964 0.0001374
et al. density GA-FIS temperature 0.9397 0.091
(2016)
Zhou et al. HTHP drilling PSO-BP Neural Initial mud density, pressure, Density at input

(2016) fluid density of network difference temperature difference, temperature &

WBM & OBM water volume fraction, oil volume pressure
fraction

Table 5 Summary of researches on mud engineering to which genetic Raja et al., 2011). An overview of the CBR technique can be found in
algorithm was applied. the work of Aamodt and Plaza (1994). The following are the cases in
which CBR were applied in drilling mud engineering:

Authors Type of study Parameters Errors (Performance conducted

evaluation criteria) (a) Lost circulation: Skalle et al. (2000) used a CBR tool to solve the
problem of lost circulation by creating 50 base cases based on
SSE
information obtained from fields in the North Sea. The cases
Rooki et al. Rheological 0.9972a 1.8305a were narrowed down to two best fitting cases from which the
Yield stress, fluid
(2012) parameter consistency and new case was solved. In addition, Rui et al. (2013) used CBR to
determination for flow index predict the risks associated with lost circulation. The result of
Herschel Bulkley their study shows that using the technique of CBR can predict
drilling fluids
Goldstein Prediction of 0.97 0.26 drilling risks such as lost circulation in a simple manner which
Nominal diameter of
and particle settling the settling particle, (RMSE) can practically be applied in the field.
Coco velocity kinematic viscosity (b) Hole cleaning: Shokouhi and Skalle (2009) used CBR to show
(2014) of the fluid, and
submerged specific
how the root cause of poor hole cleaning could be effectively
gravity of the determined. Their results were satisfactory though they
particle recommended that some improvements could still be done.
Average values. (c) Wellbore leakage: Abdollahi et al. (2008) used CBR to create 18
cases, 12 of which were solved and 6 were unsolved cases. They
categorized the cases into five groups according to the main
prediction, stuck pipe prediction, frictional pressure loss prediction,
cause of leakage. To effectively determine the main cause of the
density and viscosity predictions, lost circulation and mud theology
leakages, they assert that predefined rules can be used.
estimations. It is necessary to state here that the most common
kernel function used by most researchers is the radial basis function. 4. Strengths and weaknesses of artificial intelligence
techniques
3.6. Particle swarm algorithm
Of the seven Al techniques highlighted in this work, it is pertinent
A review of the particle swarm algorithm technique can be found to ask if there is any one of them that can be said to be 100%
in the work by Eberhart and Shi (2001) flawless and suitable for use in all circumstances. In what seems to
be an answer to the above question, Luchian et al. (2015) asserts
3.6.1. Cases where particle swarm algorithm were applied in drilling fluid that it is more beneficial focusing on solving the problem rather than
engineering wasting time to find the best method. However, a theorem
As at the time of this review, the only recorded case of using propounded by Wolpert and Macready (1997) called the No Free
particle swarm algorithm in drilling fluid engineering was the work Lunch Theorem for Optimization (NFLTO) states as follows: Given
done by Toreifi et al. (2014) where they used the method in that all problems are assumed to be equal in strength and
conjunction with modular neural network to improve the prediction irrespective of the criteria used for judging its performance, all
of the lost circulation problem. methods used for solving the problem have the same performance.
In support of the NFLTO, Anifowose et al. (2016) opines that there is
3.7. Case based reasoning not a single all-encompassing Al approach that will effectively
address all challenges in all data and computing conditions since

311
O.E. Agwu
each of the Al techniques is locked up in its unique strengths and
inevitable defects. Table 7 summarizes the strengths and
weaknesses of some of the Al techniques. The strengths and
weaknesses of four techniques namely ANN, fuzzy logic, SVM and
GA were benchmarked on six criteria of robustness against noise,
convergence speed, susceptibility to overfitting, data volume
requirements, ability to self-organize and generalization ability. It is
observed that ANN, fuzzy logic, SVM and GA are all robust against
noise, while fuzzy logic has a better speed of convergence compared
with ANN, SVM and GA. The duo of ANN and SVM has the ability to
generalize while ANN can self-organize and requires huge data to
predict complex phenomena well while SVM requires small data
volumes.
To show the effectiveness or otherwise of these Al techniques, a
few researchers have attempted to compare one or two of the Al
techniques. Table 8 shows a few of these researches. It is noticed
that the researchers mostly compared ANN against SVM. It is
observed from the performance criteria of better performance with
lower RMSE, MAPE, MSE and MAE values and high R 2 values, that
SVM tends to perform better in all of the cases cited. A unique
comparison between fuzzy logic and two hybrid systems was done
by Ahmadi et al. (2016). In their work, it is observed that the hybrid
performed better that the fuzzy logic used alone.

5. Selected applications of classical artificial intelligence in

drilling fluid engineering

Also known as symbolic artificial intelligence, classical artificial

intelligence refers to the branch of artificial intelligence that deals
with representing the implicit nature of human knowledge in an
explicit form by using facts and rules (Reingold and Nightingale,
1999). Classical artificial intelligence often deals with numerical
simulation. Numerical simulation is becoming one of the leading
techniques used to study several engineering problems that are
complex in nature (Ablat and Qattawi, 2017). Carrying out numerical
and molecular simulations on drilling muds parameters and
challenges related to them are very important. For instance,
wellbore challenges such as stuck pipes and blowout where
controlled experiments are difficult or impossible to perform can
easily be studied using numerical simulations. Several drilling fluid
related challenges have been studied through numerical
simulations. Hence, in this section, applications of numerical
simulation in drilling fluid engineering are discussed. The following
are some of the selected cases:
Feng and Gray (2017) developed a numerical model on the basis
of the finite-element method for simulating the problem of lost
circulation. The model combines dynamic circulation of mud in the
borehole and induced-fracture propagation into the formation. The
model provides an approximate value for the wellbore pressure,
rate of fluid loss, and fracture profile during wellbore drilling. The
results from their work show that the viscous pressure losses in the
annulus of the wellbore resulting from the dynamic circulation of
mud can lead to remarkable increases in wellbore pressure and fluid
loss.

312
 300-375
JournalofPetroleumScience andEngineering167(2018)

313
KJISODS!A ssa.ns.1eaqspueKusuaa (910
snlnpotuBunox 0 1u9 091nt2

日
一
aAIssa.IduIOO saS12111041011UIIIUIIU!I„N'SSOI
P!t11JIdV'KJ!SODS!A ptiLAI P!IOS 0VS0 CIA SLIO!IJLIIIJ
'Ad 91u Il! '(BAE) dumd 'dOèI IE!UIOUÁIOd (910

一
1986d 1S0 JO 'a.mssa.ld 'Kl!l!qeatu.taduonetu-IOJ'Kl!SO.10d'qadap010H pue UP!SSIIP•Ð JSOIJO 9JOUO!lJ!Pa.td pue

一
SP!IIIJ
•IMON LIP!LIOVAON-UOLI
'IMON-UOU9686·SAdVV 96r9>AdVV SSOIa.lnssa.ld JO110 13 pa.tdSOSSOIa.lnssa.td 10 unffios 91e1MOU LIO!leuIIlSa (910

一
66d adld one auadJO SSOIa.lnssa.ld OOICIE•IV !PLIEHS .1110!AeqaqMOU'xaptl! aptued
《

'Kl!suapP!lllt:l EJ1112M (ÞIOC)'1B YI MOU'dAAd'SIOB 0t 'aanssa.ldU0!aetU-10J'0101.1 uao.ladP!IOS 01112 11031100

一
一
一
忐

日
'INdH'dOžI'snlnuueJO
O.E. Agwu
300-375

'SSOIP!IIIJ
UO!lDUt1JS!SP•flIP•!PP•H ad!d (
-RI AdVL'N
ASI,NH
9OI(IEI
Butll!.tppupsanaado.ldp
ntu

314
1011.19>1
O.E. Agwu
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

Table 7 volume simulation and analytical calculations to develop a

Summary of the strengths and weaknesses of various Al techniques. comprehensive model of Herschel-Bulldey fluid flow through porous
media. The result of the model shows that yield stress has a
Benchmark FUZZY SVM GA
remarkable effect on the flow of the fluid through the porous media.
High High Song et al. (2017) reports the numerical simulation findings of the
Robustness against High High
stability of a wellbore in the presence of nanoparticle based mud. The
noise
Speed of convergence Slow Fast Slow findings from their work show that the filtrate from water-based
Prone to overfitting? Yes but depends on No muds had a high likelihood to invade the shale matrix and increase
how the training is the instability in shales. They however posited that the transmission
done of pore fluid pressure could be reduced with the addition of silica
Data Requirements Huge data required Small data
required nanoparticles to the water based mud.
Self-organization Yes No Wu et al. (2004) developed a numerical algorithm to simulate the
Ability to generalize Yes Yes physics behind the invasion of mud-filtrate in vertical and directional
Huang et al. (1998) carried out numerical simulation on the effect wellbores. The sensitivity studies from the work showed that the
of water based mud interaction with shale on the formation pressure process of mud filtrate invasion is influenced by the geometry of the
and swelling of shale. The simulation results indicate that drilling permeable beds gravity segregation, effective porosity, capillary
mud with potassium chloride salt has the least impact on shale pressure, relative permeability, hydraulic communication between
formation whereas the mud with calcium chloride salt has the most adjacent layers and permeability anisotropy. They concluded that
impact on shale formation. standard piston-like representations of drilling fluid filtrate invasion
Lichun et al. (2014) carried out numerical simulation on the effect frequently used in interpretation of well logs can lead to erroneous
of fracture roughness on lost circulation dynamics of a drilling mud interpretations of measurements from wireline tools.
exhibiting pseudoplastic behaviour. The results show that the
pseudoplastic characteristic of power-law drilling fluid can lead to 6. Selected applications of molecular simulation in drilling
high fluid loss rate at the inception of the mud loss event. The result fluid engineering
of their work indicates that as the tortuosity and stiffness of the
fracture increases, the lower the mud loss would be. They also added The following are areas where molecular simulations were
that increases in the length, opening, dip of the fracture, the higher applied in drilling fluid engineering.
the rate of mud loss.
Meng et al. (2015) investigated the pressure variation in wellbore 6.1. Fluid chemistry
caused by gas kick/injection. The model was numerically solved by
the finite volume method and advection upstream splitting method V Wang et al. (2015) used molecular dynamics (MD) modelling to
(AUSMV) scheme. The simulation results indicate that the gas and study the interlayer microstructures of montmorillonite (bentonite)
liquid mass flow rates and well depth have a strong influence on the fused with 1-hexadecyl-3-methylimidazolium chloride monohydrate
variation of the bottom pressure caused by gas kick/injection. In (C12mimCl). The results of their work showed that the insertion of
comparison with experimental data, the simulation results showed C12minCl into montmorillonite (bentonite) interlayer space
good agreement. influenced the theology of the system and improved various
Morais et al. (2009) worked on numerically simulating the flow properties. They further added that the organic clay composite
behaviour of non- Newtonian fluids in porous media. From their formed was found to be environmentally friendly and could be used
work, they concluded that the structure of the pore, the inertia and the in drilling fluid systems.
yield stress would generate a condition in which the fluid flow in the
media is increased. 6.2. Chemical reaction
Ouyang et al. (2013) used a combination of three dimensional finite
Jin et al. (2017) carried out molecular simulations to assist in the
Table 8 comprehension of the hidden mechanisms of clay damage. From their
Comparative studies of various Al techniques done by previous researchers.
Researcher Study conducted No. of data Al techniques Performance criteria
points compared

Oladunni and Trafalis (2011) Predicting fluid flow patterns in wellbore 92 ANN vs. SVM SVM ANN SVM
annulus 0.117 0.9889 0.9994
0.9528
ANN vs. SVM 0.9028 RMSE
wang et al. (2012) Predicting the drilling fluid density of 339
MAPE
WBM, OBM and SBM
0.872 ANN
MNN vs. MLP 4.343 0.575
Toreifi et al. (2014) Lost circulation prediction 1630 MSE MAE
766.98 20.66 0.9851
0.964
ANN vs. SVM 0.745
Jahanbakhshi and Keshavarzi Lost circulation prediction 260
(2015) MSE 65.28
FIS vs. PSO-ANN vs. GA- 5815.38 0.8770
Ahmadi et al. (2016) Prediction of the density of mud 664 FIS PSO-ANN
0.9964 GA-FIS
ANN vs. SVM 0.7237 0.9397
Chhantyal et al. (2016) Viscosity estimation
2.74%
DANN vs.Static ANN 23%
Chhantyal et al. (2016) Mud flow rate estimation 1800 vs. MAPE Static
DANN ANN SVR
5.6% 8.5% 7.7%

315
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

studies, they proposed that clay damage occurs through two ways characteristic of the flow within ± 0.7% accuracy for the rate of flow
namely-swelling and instability of the clay structure. They further and ± 1% accuracy for the free surface level.
added that clays swell when they come in contact with water and the Aliouane et al. (2015) used two artificial intelligence techniques
swelling occurs in a series of distinct stages, forming one or more namely: artificial neural network (multilayer perceptron) and fuzzy
layers. Their work subsequently revealed that the simulation results logic to predict the pore pressure in two horizontal wells that were
were in good agreement with the experimental data for the sodium drilled in the lower Barnett shale formation. The data from one of the
montmorillonite but not for potassium montmorillonite. wells was used for the training of the fuzzy set and the neural
network while the data of another horizontal well was used for the
6.3. Drilling fluid additives generalization. They used four input parameters namely: the neutron
porosity, natural gamma ray, the slowness of the compression and
Anderson et al. (2010) used molecular dynamics to undertake shear wave. The output parameter was the estimated pore pressure
simulations of clay swelling inhibitor systems. From their work, the obtained when Eaton's model was used. From the results they
structure and interaction of three different swelling inhibitor obtained, they reported that the fuzzy logic was better than the neural
polymers namely: poly (propylene oxide) —diamine, poly (ethylene network in predicting the pore pressure of the shale gas reservoirs.
glycol) and poly (ethylene oxide)-diacrylate inhibitor molecules with Ba et al. (2009) used marginalized particle filters to detect faults
montmorillonite clay was studied. Their results allowed them an in real time. The process model was developed in order to get a part
interpretation of bulk experimental data of the mode of interaction of which represents the faulty case and another that represents the safety
swelling inhibitors with clay minerals at an atomistic and molecular one. The distribution of these two process states was updated through
level. a Kalman filter and the concerned state was involved by using a
particle filter. By using simulated data and real field data to carry out
7. IN-SITU monitoring of drilling mud flow track route using a performance analysis indicate that the approach is efficient in terms
artificial intelligence for fault detection and isolation tasks.
Bektas et al. (2015) studied how to design and apply a steady-
Variations in the downhole temperature and pressure of state Kalman filter to optimally predict in real-time the pore pressure
formations, rock heterogeneities, inputs into and out of the drilled of a formation. This they did by combining outputs from both Eaton's
well and a host of other factors contribute to making the model and from logging while drilling (LWD). Based on their results,
measurement of drilling and drilling fluid parameters a challenging they concluded that the Kalman filter can be a very effective tool for
one. Therefore, to ensure success in drilling, it is incumbent on all detecting over-pressured zones before penetrating such zones. They
drilling personnel to ensure that the drilling process is monitored also reported that the developed algorithm can be used as a design
constantly, accurately and in real — time (Ba et al., 2009; Sui et al., tool in both onshore and offshore drilling operations along with real-
2011). One aspect of the drilling process that requires monitoring is time LWD data to optimally estimate the pore pressure of the next
the behaviour of the drilling fluid downhole, its parameters and its interval ahead of the bit during the drilling operation.
flow track route (Thorsrud et al., 2000). To achieve this, a real-time Gu and Oliver (2007) in predicting multiphase flow through a
model of the drilling fluid flow route especially in fractures or medium containing voids, used an ensemble Kalman filter. From
conductive pores is necessary. Hardware measurement devices such their work, the iterative filter — ensemble randomized maximum
as measurement while drilling (MWD) devices may provide real time likelihood filter (EnRML) gave accurate results for problems that
measurements but may fail under harsh downhole conditions (Sui et were small, linear and static in nature, and showed good agreement
al., 2011). Soft sensors are useful in overcoming this challenge. Soft with the ensemble Kalman filter (EnKF) for a problem that is linear
sensors are essentially data driven tools that require a model that and dynamic in characteristic. Focusing on the problem of two-phase
details the relation between the measured data and the quantities to be flow in a porous medium, their work reveals that the trio of ensemble
estimated (Leskens et al., 2008). This makes it possible to estimate randomized maximum likelihood filter (EnRML), ensemble Kalman
parameters which were hitherto difficult to be estimated and then filter (EnKF) and conforming ensemble Kalman filter gave results
serve as the virtual eyes of the rig crew that enables them make that were similar for the permeability and porosity (model variables)
informed decisions (Verikas, 2009). In the literature, several soft while the prediction of saturation (the state variable) was more
sensing tools abound (Leskens et al., 2008). One soft sensing tool satisfactory from EnRMI„
commonly used is the Kalman filter (KF) (Kalman, 1960). The Keshavarzi and Jahanbakhshi (2013) used an artificial neural
Kalman filter by definition is an efficient recursive data fusion network (feed-forward neural network with back-propagation
algorithm that estimates the future state of a variable of interest algorithmFFBP and a generalized regression neural network -GRNN)
(which hitherto was difficult to predict) by using a series of historical to estimate pore pressure gradient in a formation in Iran. The input
data measurements that contains noise and several other errors parameters they used include: porosity, permeability, rock density
(Aanonsen et al., 2009). The following are the selected cases of using and true vertical depth. The output parameter was the pore pressure
the Kalman filter to estimate drilling fluid flow in wellbores. gradient. The models developed were compared with Eaton's method
Agu (2014) used the numerical method of the I-D saint venant and were found to be better than the calculated values by Eaton's
equation to predict the rate of flow of fluid in both the steady and method. The pore pressure gradient values predicted by developed
dynamic states using a rectangular open channel venturi. The results FFBP when compared with GRNN show the best similarity and
from the work indicate that it is only possible to determine the rate of agreement with the measured values of pore pressure gradient for the
flow in the steady state condition. Using an ensemble Kalman filter in test data.
combination with the I-D saint venant equation, the work shows that Ljones (2013) used unscented Kalman filters to estimate two
it is possible to predict in real-time the flow rate within an accuracy drilling fluid properties namely: drilling mud density and viscosity.
level of 0.5%. He further combined the filter results for density and viscosity with
Agu and Lie (2014) in predicting the flow rate of fluids in real an unscented Kalman filter estimating the mud pump pressure, choke
time using a venturi channel applied the ensemble Kalman filter. In pressure, flow through the bit, and a geometry parameter. In each
the work, they report that this filter can be applied to a non- case, test data was generated by Simulink and WeMod simulation. It
Newtonian fluid with high flow velocity upstream within the throat was observed that while the unscented Kalman filter estimating
where the flow has slower velocities. The results indicates that the density and plastic viscosity had an acceptable performance for
algorithms for realtime predictions are able to predict the wavy Simulink model generated testdata, the unscented Kalman filters

316
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

estimating density and plastic viscosity did not perform as desired in

the case of the WeMod simulation.
• While reviewing the various limitations of the existing Al
Magalhães et al. (2016) designed a device that was able to techniques, it was found that hybrid systems were proposed by
continuously acquire experimental data of the electrical stability of some researchers as robust ways to handle and overcome some of
oil based mud and then compared the result against that obtained the limitations of the other Al techniques.
from bench tests according to API. The device was installed and
tested in an automated drilling fluid production plant. In this plant, 9. Conclusions
not only electrical stability was measured on line, but also pressure,
flow rate, temperature, viscosity, density, solids content and size and This paper reviewed the artificial intelligence technologies in
water-in-oil ratio. From their work, they found out that the on line drilling fluid engineering with recent research results and applications
and bench test data showed good agreement, as well as with other presented. It started with the evolution of Al in the exploration and
authors results. production industry before delving into the many applications of Al
Magalhaes et al. (2014) evaluated the ability of a commercial and in mud engineering. On the basis of the above findings, the following
built in property sensor to predict the theological parameters of mud, conclusions are drawn:
mud weight, water-oil content, emulsion electrical stability, fluid
conductivity and particle size distribution. They used the multi-layer 1. Mud engineering like other aspects of wellbore drilling
perceptron type of neural networks to connect the on-line equipment engineering presents an area occasioned by complex optimization
result to increase the reliability of the mud properties determination. and decisionmaking.
Their findings indicate that the online measurements were in
agreement with conventional measurements. 2. Presently, Al techniques are at different stages of application in
Mahdianfar et al. (2013) used joint unscented Kalman filter to mud engineering with artificial neural networks being the most
estimate some unknown model parameters and the bottom-hole used while case based reasoning and particle swarm algorithm are
pressure (BHP) in managed pressure drilling. The results obtained the least applied. This however does not preclude the fact that
from the study show that reliable estimation of the BHP and these two techniques may be effectively applied in other areas of
uncertain parameters can be made, even in circumstances when there drilling engineering.
are no measurements for the BHP especially when connections are to 3. The discussion presented in this paper shows that no single Al
be made. technique can solve all problems; hence, the results achieved by
means of one tool can be better in certain areas than those obtained
8. Learnings from the review using other methods. This was shown in the strength and
weaknesses section of this work.
Among the findings in relation to artificial intelligence
applications in the wellbore drilling fluid programme are: 4. Al techniques when carefully chosen and blended can be
complementary in their function resulting in predictive capabilities
that are sometimes better than when they are used singly and the
• The numbers of research papers are indicative of the relevance results added together. For instance, merging the computational
and growth of artificial intelligence in mud engineering. Focusing power of ANNs with the symbolic level of processing associated
on the contents of the available literature with information about with fuzzy logic gives rise to a hybrid system that improves the
various applications of Al in mud engineering, we can roughly speed of convergence speed, improved ability to deal with faults
estimate that about 54% papers are related to ANN, 22% are and general adaptability to varying conditions.
connected to support vector machines, about 6% — to hybrid
intelligent systems, a little over 8% were dedicated to fuzzy logic, 10. Recommendations for future research
3% to genetic algorithms and about 6% to CBR while about 1%
was dedicated to particle swarm algorithm.
1. From the review, multi-gene genetic programming-a variant form
• Lost circulation and downhole mud density predictions were the of genetic programming was found to be unique in that it produced
most studied aspects of mud engineering by the researchers whose an explicit model unlike other Al techniques. Hence, for
works were reviewed; showing that these are critical areas to the comparative purposes, it is recommended that its predictive
wellbore drilling. capability be tested vis a vis other Al techniques by using it to
model other complex phenomena relating to wellbore drilling.
• Artificial intelligence techniques are unique in the ways in which
they obtain knowledge, depict, keep, apply the knowledge gained, 2. A number of artificial intelligence techniques have been identified
adapt to new problems, deal with variability, etc. Hence, it is that are most likely suitable for predicting and/or optimizing
always vital to consider what technique would be best for drilling mud parameters and mud related problems with acceptable
predicting different phenomena of the wellbore drilling fluid levels of uncertainty. One of them is case based reasoning. To get
program. the most out of this technique requires a database of previous
scenarios and how they were solved. Thus, in order to benefit the
• The most commonly used artificial intelligence technique in mud industry, the industry operators should make available historical
engineering is artificial neural network with multiple inputs and data to CBR researchers to enable them create a comprehensive
single outputs; however, sometimes the problem under database that can diagnose more than one wellbore challenge at a
consideration may involve multiple inputs and multiple outputs. time.
• With a plethora of data, there is need to choose carefully the most 3. It is hoped that the review this paper presents would provide for
important parameters in the data in order to decrease the size of the reader a broad view of what artificial intelligence has
the datasets which in turn increases the predictive performance of accomplished in drilling fluid engineering without necessarily
the models. Additionally, the commanding importance of data limiting the reader's imaginative thoughts about the existence of
preprocessing steps such as data normalization cannot be future research opportunities which are, literally speaking,
overemphasized given the boosting effect they impart to the limitless.
performance of Al models.
Acknowledgement

317
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

The authors would like to appreciate the management of the One Dimensional
University of Uyo for providing an enabling environment to carry out 2-D Two Dimensional
this research.
AAPE Average Absolute Percentage Error
Abbreviations Al Artificial Intelligence
ANFIS Artificial Neuro Fuzzy Intelligent
System
ANN Artificial Neural Network
AUSMV Advection Upstream Splitting Method
V
BHP Bottom Hole Pressure; BP Back
Propagation
BPNN Back Propagation Neural Network
CBR Case Based Reasoning
DANN Dynamic Artificial Neural Network
Extreme Learning Machine
EnKF Ensemble Kalman Filter
EnRML Ensemble Randomized Maximum
Likelihood filter Exploration and
Production
FFBPN Feed Forward Back Propagation Network
FFMLP Feed Forward Multilayer Perceptron
FFNN Feed Forward Neural Network
FIS Fuzzy Intelligent Systems
GA Genetic Algorithm
GA - FIS Genetic Algorithm Fuzzy Intelligent
System
GA - FIS Genetic Algorithm Fuzzy Intelligent
System
GA-RBF Genetic Algorithm Radial Basis Function
GA-RBF Genetic Algorithm Radial Basis Function
GFNN Generalized Feed Forward Neural Network
HIS Hybrid Intelligent Systems
HTHP High Temperature High Pressure
Kalman Filter
LWD Logging While Drilling
MSE Mean Square Error
MAE Maximum Absolute Error
MAPE Maximum Absolute Percent Relative Error
MLP Multilayer Perceptron
MNN Modular Neural Network
MWD Measurement While Drilling
NFLTO No Free Lunch Theorem for Optimization
NMSE Normalized Mean Squared Error
NPT Non-Productive Time
OBM Oil Based Mud
PSA Particle Swarm Algorithm
PSO Particle Swarm Optimization
PSO-ANN Particle Swarm Optimization Artificial Neural Network
PSO-ANN Particle Swarm Optimization Artificial Neural Network
PSO-BP Particle Swarm Optimization Back
Propagation
PSO-BP Particle Swarm Optimization Back
Propagation

318
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

R2 Correlation Coefficient challenges ahead. J. Petroleum Explor. Prod. Technol. 7 (Issue 1), 251—263.
Springer link Publishers. https://doi.org/10.1007/s13202-016-0257-3.
RBF Radial Basis Function Artificial Intelligence Applications Institute, 2014. Centre for Intelligent Systems and
RBNN Radial Basis Neural Network Their Applications. School of Informatics. The University of Edinburgh. http://www.
RMSE Root Mean Square Error aiai.ed.ac.uk/, Accessed date: 5 June 2017.
Ataallahi, E. , Shadizadeh, S.R., 2015. Fuzzy consequence modelling of blowouts in
RNN Recurrent Neural Network Iranian drilling operations; HSE consideration. Saf. Sci. 77, 152—159. Elsevier
SONFIS Self -Organizing Neuro Fuzzy Inference Publishers. https://doi.org/10.1016/j.ssci.2015.03.014.
Ba, A. , Hbaieb, S. , Mechbal, N. , Vergé, M. , 2009. Faults detection by marginalized
Systems particle filters: application to a drilling process. In: Proceedings of the 7th IFAC
SPE Society of Petroleum Engineers Symposium on Fault Detection, Supervision and Safety of Technical Processes
SVM Support Vector Machines; Barcelona, Spain.
Baarimah, S.O., Gawish, A.A., BinMerdhah, A. B. , 2015. Artificial intelligence
Support Vector Regression techniques for predicting the reservoir fluid properties of crude oil systems. Int. Res.
SBM Synthetic Based Mud J. Eng. Technol. 2 (Issue 7).
WBM Water Based Mud Barati, R., Neyshabouri, S.A.A.S., Ahmadi, G. , 2014. Development of empirical models
with high accuracy for estimation of drag coefficient of flow around a smooth
Funding sphere: an evolutionary approach. Powder Technol. 257, 11—19. Elsevier
Publishers. https:// doi.org/10.1016/j.powtec.2014.02.045.
No funding was provided for this research by any individual or Barton, D.L., Sviokla, J. , 1988. Putting Expert Systems to Work. Havard Business
Review. https://hbr.org/1988/03/putting-expert-systems-to-work, Accessed date: 7
corporate organisation. June 2017.
Behnoud far, P. , Hosseini, P., 2016. Estimation of lost circulation amount occurs during
Conflicts of interest under balanced drilling using drilling data and neural network. Egypt. J. Petroleum
Elsevier Publishers. https://doi.org/10.1016/j.ejpe.2016.09.004.
Bektas, E., Miska, S.Z., Ozbayoglu, E.M., Yu, M., Takach, N., Velazquez-Cruz, D,
No conflicts of interest exist. Shahri, M.P., 2015. Application of Kalman filter to predictions of pore pressure
while drilling. In: Paper SPE -174908-MS Presented at the SPE Annual Technical
Conference and Exhibition, Houston, Texas, USA, . https://doi.org/10.2118/174908-
References MS.
Bello, O. , Holzmann, J. , Yaqoob, T. , Teodoriu, C. , 2015. Application of Artificial
Aamodt, A. , Plaza, E. , 1994. Case-based Reasoning: Foundational Issues, Intelligence Methods in Drilling System Design and Operations: a Review of the
Methodological Variations, and System Approaches. Al Communications, vol. 7. State of the Art, vol. 5. De Gruyter International, pp. 121—139 Number 2.
IOS Press, pp. 39—59 Number 1. Bispo, V.D.d. S. , Scheid, C.M., Calçada, LA., da Cruz Meleiro, L.A., 2017.
Abdollahi, J. , 2007. Analysing Complex Oil Well Problems through Case-based Development of an ANN-based soft-sensor to estimate the apparent viscosity of
Reasoning. Doctoral Thesis Submitted to the Department of Petroleum Engineering water-based drilling fluids. J. Petroleum Sci. Eng. 150, 69—73.
and Applied Geophysics. Norwegian University of Science and Technology, https://doi.org/10.1016/j.petrol.2016.11. 030.
Trondheim. Bloys, B. , Davis, N. , Smolen, B. , Bailey, L, Fraser, L, Hodder, M. , 1994. Designing
Abdollahi, J., Carlsen, I.M., Randhol, P., Tenold, E., Haga, H.B., Jakobsen, T., 2008. A and Managing Drilling Fluid. Schlumberger Resources. Oilfield Review. pp. 33—43.
case based approach to understand the complexity of causal connections related to Boman, K., 2014. Artificial Intelligence Software Aids Decision-making in Onshore
Drilling. Rigzone Incoporated. http://www.rigzone.com/news/oil_gas/a/133973/
Artificial_Intelligence_Software_Aids_DecisionMaking_in_Onshore_Drilling/?all
Drilling Conference, Orlando, Florida, USA. = HG2, Accessed date: 7 June 2017.
Ablat, M.A., Qattawi, A. , 2017. Numerical simulation of sheet metal forming: a review. Bravo, C., Saputelli, L, Rivas, E, Pérez, A.G., Nikolaou, M., Zangl, G. , de Guzman,
Neil, Mohaghegh, S. , Nunez, G., 2012. State of the art of artificial intelligence and
Int. J. Adv. Manuf. Technolology 89 (Issue 1 - 4), 1235-1250. https://doi.org/10.
predictive analytics in the E&P industry: a technology survey. In: Paper SPE 150314
1007,600170-016-9103-5.
Presented at the SPE Western Regional Meeting Held in Bakersfield, California,
Ab Wahab, M.N., Nefti-Meziani, S. , Atyabi, A. , 2015. A comprehensive review of
USA, . https://doi.org/10.2118/150314-PA.
swarm optimization algorithms. PLOS ONE 10 (Number 5).
Chamkalani, A. , Shahri, M.P., Poordad, S. , 2013. Support vector machine model: a new
http://dx.doi.org/10.1371/ journal. pone. 0122827.
methodology for stuck pipe prediction. In: Paper SPE 164003 Presented at the SPE
Adesina, F.A.S., Abiodun, A., Anthony, A., Olugbenga, F. , 2015. Modelling the effect of
Middle East Unconventional Gas Conference and Exhibition, Muscat, Oman.
temperature on environmentally safe oil based drilling mud using artificial neural
Chhantyal, K. , Viumdal, H. , Mylvaganam, S. , 2017a. Soft sensing of non-Newtonian
network algorithm. Petroleum Coal J. 57 (Number 1), 60—70.
fluid flow in open venturi channel using an array of ultrasonic level sensors—AI models
Aanonsen, S.I., Naevdal, G. , Oliver, D.S., Reynolds, A.C., Valles, B. , 2009. The
and their validations. Sensors 17 (Issue 11). http://dx.doi.org/10.3390/s17112458.
ensemble Kalman filter in reservoir engineering — a review. Paper SPE 117274. SPE J.
Chhantyal, K. , Viumdal, H. , Mylvaganam, S. , 2017b. Ultrasonic level scanning for
393—412. Agu, C.E., 2014. Model Based Estimation of Drilling Mud Flow Using a
monitoring mass flow of complex fluids in open channels — a novel sensor fusion
Venturi Channel. TUC, Porsgrunn.
approach using Al techniques. Sensors, IEEE. http://dx.doi.org/10.1109/ICSENS.2017.
Agu, C.E., Lie, B. , 2014. Smart sensors for measuring fluid flow using a venturi channel.
8234010.
In: Proceedings from the 55th Conference on Simulation and Modelling (SIMS 55),
Chhantyal, K. , Viumdal, H. , Hoang, M., Mylvaganam, S. , Elseth, G. , 2016. Flowrate
Aalborg, Denmark, pp. 229—240.
estimation using ultrasonic level sensors using dynamic artificial neural networks
Agwu, O.E., Okon, A.N., Udoh, ED., 2015. A comparative study of diesel oil and
with real time recurrent learning — a comparative study of models and practical
soybean oil as oil-based drilling mud. J. Petroleum Eng Hindawi Publishing
implementation. In: Paper Presented at the 9th Eurosim Congress on Modelling and
Corporation. https://doi.org/10.1155/2015/828451.
Simulation, Finland.
Ahmadi, M.A., 2016. Toward reliable model for prediction of drilling fluid density at
Courteille, J.M., Fabre, M. , Hollander, C.R., 1986. An advanced solution: the drilling
wellbore conditions: a ISSVM model. Neurocomputing J. 211, 143—149. https://doi.
adviser. J. Petroleum Technol.
org/10.1016/j.neucom.2016.01.106.
Cristofaro, R.A.L, Longhin, GA., Waldmann, A.A., de Sá, C.H.M., Vadinal, R.B.,
Ahmadi, M.A., Shadizadeh, S.R., Shah, K. , Bahadori, A. , 2016. An accurate model to
Gonzaga, K.A., Martins, AL, 2017. Artificial intelligence strategy minimizes lost
predict drilling fluid density at wellbore conditions. Egypt. J. Petroleum. https://doi.
circulation non-productive time in Brazilian deep water pre-salt. In: Paper OTC-
org/10.1016/j.ejpe.2016.12.002.
28034-MS Presented at the Offshore Technology Conference. OTC Brasil, Rio de
Al-Baiyat, I. , Heinze, L. , 2012. Implementing artificial neural networks and support Janeiro, Brazil. https://doi.org/10.4043/28034-MS.
vector machines in stuck pipe prediction. In: Paper SPE 163370 Presented at the SPE
Deregeh, F. , Karimian, M. , Nezamabadi-Pour, H. , 2013. A new method of earlier kick
Kuwait International Petroleum Conference and Exhibition, Kuwait City, Kuwait.
assessment using ANFIS. Iran. J. Oil Gas Sci. Technol. 2 (Number 1), 33—41.
Alegre, L. , 1991. Potential applications for artificial intelligence in the petroleum
Doraisamy, H. , Ertekin, T. , Grader, A., 1998. Key parameters controlling the
industry. J. Petroleum Technol. 1306—1309 Society of Petroleum Engineers.
performance of neuro simulation applications in field development. In: SPE 51079,
Aliouane, L, Ouadfeul, S. , Boudella, A. , 2015. Pore Pressure prediction in shale gas Proceedings, SPE Eastern Regional Conference, Pittsburgh, PA.
reservoirs using neural network and fuzzy logic with an application to Barnett Shale.
Drilling Specialties Company (Online) The engineered lost circulation materials of
Geophys. Res. Abstr. 17, 2015-2723 EGU.
drilling specialties company. http://www.cpchem.com/bl/drilling/en-us/l)ocuments/
Ali, B.H., Adel, N.M., Afshin, T., Milad, A., Kim, P.L.T., Moonyong, L, Alireza, B.,
THEENGINEEREDLOSTCIRCULATIONMATERIALSOF.pdf. (Accessed 18
2016. Prediction of frictional pressure loss for multiphase flow in inclined annuli
September 2017).
during Underbalanced Drilling operations. Nat. Gas. Ind. B 3, 275—282. Elsevier
Eberhart, R.C., Shi, Y. , 2001. Particle swarm optimization: developments, applications
Publishers. https://doi.org/10.1016/j.ngib.2016.12.002.
and resources. In: Proceedings of the IEEE Congress on Evolutionary Computation
Alireza, M. , Mohammad, N. , Ahmad, A., 2011. Reducing consumed energy while
(CEO, Seoul, South Korea. vol. 1. pp. 81-86. http://dx.doi.org/10.1109/CEC.2001.
drilling an oil well through a deep rig time analysis. Adv. Petroleum Explor. Dev. CS
934374.
1 (Number 1), 22—31 Canada Publishers.
Ejimofor, J.C., 2016. Open Venturi-channel Flow Metering of Non-Newtonian Fluids: a
Anderson, R.L., Greenwell, H.c., suter, J.L., Jarvis, RM., Coveney, P.v., 2010. Towards
Case Study with Emulated Drilling Fluids. Masters Degree Thesis Submitted to the
the design of new and improved drilling fluid additives using molecular dynamics
Department of System and Control Engineering, Faculty of Technology. University
simulations. Ann. Braz. Acad. Sci. 82 (Number 1), 43—60.
College of Southeast Norway.
Anifowose, EA., Labadin, J. , Abdulraheem, A. , 2016. Hybrid intelligent systems in
petroleum reservoir characterization and modelling: the journey so far and the

319
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

Elkatatny, S. , 2017. Real-time prediction of rheological parameters of KCI water-based Lichun, J., Mian, C., Bing, H. , Then, S. , Yan, J. , 2014. Drilling Fluid Loss Model and
drilling fluid using artificial neural networks. Arab J. Sci. Eng. 42, 1655—1665. Loss
https://doi.org/10.1007/0369-016-2409-7. Dynamic Behaviour in Fractured Formations. Research Institute of Petroleum
Elkatatny, S. , Tariq, Z, Mahmoud, M. , 2016. Real time prediction of drilling fluid Exploration and Development, PetroChina, vol. 41. Elsevier BV Publishers, pp.
rheological properties using artificial neural networks visible mathematical model 105-112. Issue 1. https://doi.org/10.1016/S1876-3804(14)60012-4.
(white box). J. Petroleum Sci. Eng. 146, 1202—1210. Ljones, T., 2013. Drilling Mud Property Estimator. Masters Degree Thesis Submitted to
https://doi.org/10.1016/j.petrol.2016. the Department of Engineering Cybernetics. Norwegian University of science and
08.021. Technology.
Feng, Y. , Gray, K.E, K. E. , 2017. Modelling lost circulation through drilling-induced Lim, J. , 2005. Reservoir properties determination using fuzzy logic and neural networks.
fractures. In: Paper SPE 187945 —PA. SPE Journal. Society of Petroleum Engineers. J. Petroleum Sci. Eng. 49 (Issues 3—4), 182—192. https://doi.org/10.1016/j.petrol.
https://doi.org/10.2118/187945-PA. 2005.05.005.
Firoozjaee, R.A., Khamehchi, E. , 2014. A novel approach to assist history matching Li, M. , Zhang, G. , Xue, J. , Li, Y. , Tang, S. , 2014. Prediction of the wall factor of
using artificial intelligence. J. Chem. Eng. Commun. 202 (Issue 4) Taylor and arbitrary particle settling through various fluid media in a cylindrical tube using
Francis Online. https://doi.org/10.1080/00986445.2013.852977. artificial intelligence. Sci. World J Hindawi Publishing Corporation.
Ghaffari, H. , Shakeri, H. , Sharifzadeh, M. , 2009. Developing of new facets of indirect https://doi.org/10.1155/ 2014/438782.
modelling in the geosciences. In: Proceedings of the International Conference on Luchian, H. , Breaban, M.E., Bautu, A. , 2015. On meta-heuristics in optimization and
Rock Joints and Jointed Rock Masses, Tucson, Arizona, pp. 1—8. data analysis. application to geosciences. In: Cranganu (Ed.), Artificial Intelligent
https://espace.curtin.edu. au/hand1e/20.500.11937/33454. Approaches in Petroleum Geosciences. Springer International Publishing.
Goldstein, E.B., Coco, G. , 2014. A machine learning approach for the prediction of Magalhaes, S. , Scheid, CM. , Calcada, LA., Folsta, M. , Martins, AL., Marques deSa,
settling velocity. Water Resour. Res. 50, 3595-3601. http://dx.doi.org/10.1002/ C.H., 2014. Development of on-line sensors for automated measurement of drilling
2013WR015116. fluid properties. In: Paper SPE-167978-MS Presented at the IADC/SPE Drilling
Gu, Y. , Oliver, D.S., 2007. An iterative ensemble Kalman filter for multiphase fluid flow Conference and Exhibition, Fort Worth, Texas, USA, .
data assimilation. In: Paper SPE 108438, SPE Journal. Society of Petroleum https://doi.org/10.2118/167978-MS.
Engineers, pp. 438—446. Magalhäes, S.C., Calçada, LA., Scheid, C.M., Almeida, H. , Waldmann, A.T.A., 2016.
Ho, T.L., Ehara, S. , 2007. Oil Reservoir Properties Estimation by Fuzzy-neural Improving drilling performance with continuous online measurements of electrical
Networks. stability and conductivity in oil based drilling fluids. J. Petroleum Sci. Eng. 146,
Memoirs of the Faculty of Engineering, vol. 67 Kyushu University Number 3. 369-379. https://doi.org/10.1016/j.petr01.2016.05.045.
Hoang, M., 2016. Tuning of Viscosity and Density of Non-Newtonian Fluids through Maghrabi, S. , Kulkarni, D. , Teke, K. , Kulkarni, S.D., Jamison, D. , 2014. Modelling of
Mixing Process Using Multimodal Sensors, Sensor Fusion and Models. Masters shaleerosion behaviour in aqueous drilling fluids. In: Paper Presented at the
Degree Thesis Submitted to the Faculty of Electrical, it and Cybernetics. University SPE/EAGE European Unconventional Resources Conference and Exhibition.
College of Southeast Norway. Mahdianfar, H., Pavlov, A. , Aamo, O.M., 2013. Joint unscented Kalman filter for state
Huang, H. , Azar, J.J., Hale, A.H., 1998. Numerical simulation and experimental studies and parameter estimation in managed pressure drilling. In: European Control
of shale interaction with water based drilling fluid. In: Paper IADC/SPE 47796 Conference (ECC), Zürich, Switzerland.
Presented at the 1998 IADC/SPE Asia pacific Drilling Conference, Jakarta, Manshad, A.K., Rostami, H. , Niknafs, H. , Mohammadi, A.H., 2017. In: Heavy,
Indonesia, https://doi.org/10.2118/47796-MS. Oil, Mohammadi, Amir H. (Eds.), Integrated Lost Circulation Prediction in
iSCOOP, 2016. Artificial intelligence (Al) and cognitive computing: what, why and Drilling Operation. Nova Science Publishers, Inc.
where. https://www.i-scoop.eu/artificial-intelligence-cognitive-computing/, Mathworks (online) User Stories: Halliburton Makes Oil Exploration Safer Using
Accessed date: 5 June 2017. MATI_AB and Neural Networks.
Isebor, O.J., Echeverría, CD. , Durlofsky, L. , 2014. Generalized field-development https://www.mathworks.com/company/user_stories/ halliburton-makes-oil-
optimization with derivative-free procedures. SPE J. 19 (Number 5) (SPE-163631- exploration-safer-using-matlab-and-neural-networks.html. (Accessed 26 February
PA). 2018).
https://doi.org/10.2118/163631 -PA. McCarthy, J., 2007. What Is Artificial Intelligence? Computer Science Department,
Jahanbakhshi, R. , Keshavarzi, R. , 2015. Quantitative and qualitative analysis of lost Stanford University, Stanford, CA 94305. http://www-formal.stanford.edu/jmc/
circulation in natural and induced fractured formations: the integration of whatisai.pdf, Accessed date: 12 June 2017.
operational conditions and geomechanical parameters. Eur. J. Environ. Civ. Eng. Medsker, L.R., 1995. Hybrid Intelligent Systems. Kluwer Academic Publishers.
19 (Number 4), 418—444. https://doi.org/10.1080/19648189.2014.949872. Meng, Y. , Xu, C. , Wei, N. , Li, G. , Li, H. , Duan, M. , 2015. Numerical simulation and
Jahanbakhshi, R. , Keshavarzi, R. , Jalili, S. , 2014. Artificial neural network-based experiment of the annular pressure variation caused by gas kick/injection in wells. J.
prediction and geomechanical analysis of lost circulation in naturally fractured Nat. gas Sci. Eng. 22, 646—655. Elsevier BV publishers. https://doi.org/l O. 1016/j.
reservoirs: a case study. Eur. J. Environ. Civ. Eng. 18 (Number 3), 320—335. jngse.2015.01.013.
https://doi. org/10.1080/19648189.2013.860924. Meschi, M.R., Shahbazi, K. , Shahri, M.P., 2010. A new criteria to predict stuck
Jahanbakhshi, R. , Keshavarzi, R. , Shoorehdeli, M.A., Emamzadeh, A. , 2012. Intelligent occurrence. In: Paper SPE 128376 Presented at the SPE North Africa Conference
prediction of differential pipe sticking by support vector machine compared with and Exhibition, Cairo, Egypt, . https://doi.org/10.2118/128376-MS.
conventional artificial neural networks: an example of Iranian offshore oil fields. Miri, R. , Sampaio, J.H.B., Afshar, M. , 2007. Development of artificial neural networks
SPE Drill. Complet. https://doi.org/10.2118/163062-PA. to predict differential pipe sticking in Iranian offshore oil fields. In: Paper SPE-
Jeirani, Z. , Mohebbi, A. , 2005. Artificial neural networks approach for estimating 108500MS Presented at the International Oil Conference and Exhibition in Mexico,
filtration properties of drilling fluids. J. Jpn. Petroleum Inst. 49 (Number 2), 65—70. 27-30 June, Veracruz, Mexico, . https://www.onepetro.org/search?q = affiliation
https://doi.org/10.1627/jpi.49.65. A. https://doi.
Jin, L, Jamili, A. , Huang, L, Perez, F. , 2017. Modelling the mechanisms of clay damage
by molecular dynamic simulation. J. Geofluids Hindawi Publishing Corporation. org/10.2118/108500-MS.
Moazzeni, A. R. , Nabaei, M. , Jegarluei, S.G., 2010. Prediction of lost circulation using
https://doi.org/10.1155/2017/1747068.
virtual intelligence in one of Iranian oilfields. In: Paper SPE 136992 Presented at the
Kalman, R. E. , 1960. A new approach to linear filtering and prediction problems. Trans.
34th Annual SPE International Conference and Exhibition Held in Tinapa —
ASME-Journal Basic Eng. 82 (Series D), 35-45. https://doi.org/10.1115/1.3662552.
Calabar, Nigeria, https://doi.org/10.2118/136992-MS.
Kamari, A. , Gharagheizi, F. , Shokrollahi, A., Arabloo, M. , Mohammadi, A.H., 2017.
Moazzeni, A. , Nabaei, M. , Jegarluei, S.G., 2012. Decision making for reduction of
Estimating the drilling fluid density in the mud technology: application in high
nonproductive time through an integrated lost circulation prediction. Petroleum Sci.
temperature and high pressure petroleum wells. In: Mohammadi, A.H. (Ed.), Heavy
Technol. 30 (Number 20), 2097-2107. https://doi.org/10.1080/10916466.2010.
Oil. Nova Science Publishers, Inc, pp. 285—295.
495961.
Kamyab, M. , Dawson, R. , Farmanbar, P. , Corval, Al, 2016. A new method to determine
Mohaghegh, D.S., 2000. Virtual- intelligence applications in petroleum engineering: part
friction factor of cuttings slip velocity calculation in vertical wells using Neural
1 — artificial neural networks. SPE 58046. J. Petrol. Technol.
Networks. In: Paper SPE - 182359- MS Presented at the SPE Asia Pacific Oil and
Morais, A.F., Seybold, H. , Herrmann, H.J., Andrade Jr., J.S., 2009. Non-Newtonian fluid
Gas Conference and Exhibition Held in Perth Australia, . https://doi.org/10.2118/
flow through three-dimensional disordered porous media. Phys. Rev. Lett. 103 (Issue
182359-MS.
19). https://doi.org/10.1103/PhysRevLett.103.194502.
Keshavarzi, R. , Jahanbakhshi, R. , 2013. Real-time prediction of pore pressure gradient
Murillo, A. , Johanna Neuman, J. , Samuel, R., 2009. Pipe sticking prediction and
through an artificial intelligence approach: a case study from one of middle east oil
avoidance using adaptive fuzzy logic modelling. In: Paper SPE-120128-MS
fields. Eur. J. Environ. Civ. Eng. 17 (Issue 8), 675-686. https://doi.org/10.1080/
Presented at the SPE Production and Operations Symposium, Oklahoma City,
19648189.2013.811614.
Oklahoma, . https:// doi.org/10.2118/120128-MS.
Khamehchi, E., Rahimzadeh Kivi, I. , Akbari, M. , 2014. A novel approach to sand
Naderi, M. , Khamehchi, E. , 2017. Well placement optimization using metaheuristic bat
production prediction using artificial intelligence. J. Petroleum Sci. Eng. 123, 147—
154. https://doi.org/10.1016/j.petr01.2014.07.033. algorithm. J. Petroleum Sci. Eng. 150, 348—354. https://doi.org/10.1016/j.petrol.
Lea, G. , 2015. The Struggle to Define what Artificial Intelligence Actually Means. 2016.12.028.
Popular Science. Bonnier Corporation. http://www.popsci.com/why-we-need- Nikravesh, M. , Kovscek, A. R. , Jonston, R.M., Patsek, T.W., 1996. Prediction of
legaldefinition-artificial-intelligence, Accessed date: 12 June 2017. formation damage during the fluid injection into fractured low permeability
Legg, S. , Hutter, M. , 2007. A collection of definitions of artificial intelligence. In: reservoirs via neural networks. In: SPE 31103, Proceedings, SPE Formation Damage
Proceedings of the 2007 Conference on Advances in Artificial General Intelligence: Symposium, Lafayette, LA, . https://doi.org/10.2118/31103-MS.
Concepts, Architectures and Algorithms. IOS Press, Amsterdam, pp. 17—24. Nilsson, N. , 2010. The Quest for Artificial Intelligence. Cambridge University Press,
Leskens, M., de Kruif, B. , Belfroid, S. , Smeulers, J. , Gryzlov, A. , 2008. Downhole Cambridge, pp. 13.
multiphase metering in wells by means of soft-sensing. In: Paper SPE 112046 Oladunni, 0.0. , Trafalis, T.B., 2011. Single-phase fluid flow classification via learning
Presented at the 2008 SPE Intelligent Energy Conference and Exhibition, models. Int. J. General Syst. 40 (Number 5), 561—576.
Amsterdam, Netherland. Osman, E.A., Aggour, M.A., 2003. Determination of drilling mud density change with
pressure and temperature made simple and accurate by ANN. In: Paper SPE 81422

320
O.E. Agwu et al. Journal of Petroleum Science and Engineering 167 (2018) 300—315

Presented at the SPE 13th Middle East Oil Show and Conference, Bahrain, . https:// Sheremetov, L. , Batyrshin, I.Z., Filatov, D. , Martinez, J. , Rodriguez, H. , 2008. Fuzzy
doi.org/10.2118/81422-MS. expert system for solving lost circulation problem. J. Appl. Soft Comput. 14—29.
Ouyang, L. , Hill, A.D., Zhu, D. , 2013. Theoretical and numerical simulation of Elsevier Publishers. https://doi.org/10.1016/j.asoc.2006.11.003.
HerschelBulkley fluid flow in propped fractures. In: Paper SPE-163841-MS Shokouhi, S.V. , Skalle, P. , 2009. Enhancing decision making in critical drilling
Presented at the SPE Hydraulic Fracturing Technology Conference, the Woodlands, operations.
Texas, USA, . https://doi.org/10.2118/163841-MS. In: Paper SPE 120290, Prepared for Presentation at the SPE Middle East Oil & Gas
Ozbayoglu, E.M., Ozbayoglu, M.A., 2009. Estimating flow patterns and frictional Show and Conference Held in the Bahrain, . https://doi.org/10.2118/120290-MS.
pressure losses of two-phase fluids in horizontal wellbores using artificial neural Siruvuri, C. , Nagarakanti, S. , Samuel, R. , 2006. Stuck pipe prediction and avoidance: a
networks. Petroleum Sci. Technol. 27, 135—149. Taylor and Francis Group, LLC. convolutional neural network approach. In: Paper IADC/SPE 98378 Presented at the
https://doi. org/10.1080/10916460701700203. IADC/SPE Drilling Conference, Miami, Florida, U.S.A.
Petrus, J.B., Thuijsman, F. , Weijters, A.J., 1995. Artificial Neural Networks: an Skalle, P. , Sveen, J. , Aamodt, A., 2000. Improved efficiency of oil well drilling through
Introduction to ANN Theory and Practice. Springer„ UK. case-based reasoning. In: Proceedings of PRICAI 2000, Sixth Pacific Rim
Popa, A.S., Cassidy, S. , 2012. Artificial Intelligence for heavy oil assets: the evolution of International Conference on Artificial Intelligence. Melbourne.
solutions and organization capability. In: Paper SPE 159504 Presented at the SPE song, J., Yuan, Y., Gil, S., Yang, X., Yue, Y., cai, J., Jiang, G., 2017. 2D Numerical
Annual Technical Conference and Exhibition, San Antonio, Texas, USA, . Simulation of Improving Wellbore Stability in Shale Using Nanoparticles Based
https://doi. org/10.2118/159504-MS. Drilling Fluid. Energies, vol. 10http://dx.doi.org/10.3390/en10050651.
Rable, B. , 2017. The future is here: 3 ways Al roots itself in O&G in the surge Sorgun, M. , Ozbayoglu, A.M., Ozbayoglu, M.E., 2014. Support vector regression and
Magazine. http://thesurge.com/stories/future-artificial-intelligence-roots-oil-gas- computational fluid dynamics modelling of Newtonian and non-Newtonian fluids in
industry , Accessed date: 5 June 2017. annulus with pipe rotation. J. Energy Resour. Technol. 137 (Number 3).
Raja, H.Z., Sormo, F. , Vinther, M.L., 2011. Case-based reasoning: predicting real- Steinwart, I. , Christmann, A., 2008. Support Vector Machines. Springer Publishers, New
time drilling problems and improving drilling performance. In: Paper SPE- York
141598-MS Presented at the SPE Middle East Oil and Gas Show and Conference, Sui, D. , Nybø, R. , Gola, G. , Roverso, D. , Hoffmann, M. , 2011. Ensemble methods for
Manama, Bahrain, . https://doi.org/10.2118/141598-MS. process monitoring in oil and gas industry operations. J. Nat. Gas Sci. Eng. 3 (Issue
Rasouli, H. , Rashidi, F., Karimi, B. , Khamehchi, E. , 2015. A surrogate integrated 6), 748-753. https://doi.org/10.1016/j.jngse.2011.05.004.
production modelling approach to long-term gas-lift allocation optimization. Chem. Sung, W., Hanyang, U. , Yoo, I. , 1995. Development of HT-BP neural network system
Eng. for the identification of well test interpretation model. In: SPE 30974, Proceedings,
Commun. 202 (Number 5), 647-654. SPE Eastern Regional Conference and Exhibition, Morgantown, West Virginia.
Razi, M.M., Mazidi, M. , Razi, EM. , Aligolzadeh, H. , Niazi, S. , 2013. Artificial neural Tatar, A. , Halali, M.A., Mohammadi, A.H., 2016. On the estimation of the density of
network modelling of plastic viscosity, yield point, and apparent viscosity for brine with an extensive range of different salts composditions and concentrations. J.
waterbased drilling fluids. J. Dispersion Sci. Technol. 34 (Number 6), 822—827. Thermodyn. Catal. 7 (Number 167). http://dx.doi.org/10.4172/2160-7544.
http://dx. doi.org/10.1080/01932691.2012.704746. 1000167.
Reingold, E. , Nightingale, J., 1999. Artificial Intelligence Tutorial Review Ternyik, J. , Bilgesu, I. , Mohaghegh, S., Rose, D. , 1995a. Virtual measurement in pipes,
(PSY371)Symbolic Al. part 1: flowing bottomhole pressure under multi-phase flow and inclined wellbore
http://www.psych.utoronto.ca/users/reingold/courses/ai/symbolic. html, Accessed conditions. In: Paper SPE 30975, Proceedings, SPE Eastern Regional Conference and
date: 31 December 2017. Exhibition, Morgantown, West Virginia.
Rooki, R. , 2015. Estimation of pressure loss of Herschel—Bulkley drilling fluids during Ternyik, J. , Bilgesu, I. , Mohaghegh, S. , 1995b. Virtual measurement in pipes, Part 2:
horizontal annulus using artificial neural network. J. Dispersion Sci. Technol. 36, liquid holdup and flow pattern correlation. In: SPE 30976,.Proceedings, SPE Eastern
161-169. Taylor and Francis Group, LLC. https://doi.org/10.1080/01932691.2014. Regional Conference and Exhibition, Morgantown, West Virginia.
904793. Thorsrud, A.K., Ekeli, Ø., Hilbig, N.C.C., Bergsvik, O. , Zamora, M., 2000. Application
Rooki, R. , 2016. Application of general regression neural network (GRNN) for indirect of novel downhole hydraulics software to drill safely and economically a north sea
measuring pressure loss of Herschel—Bulkley drilling fluids in oil drilling. J. Meas. hightemperature/high-pressure exploration well. In: Paper SPE-59189-MS Presented
85, 184-191. https://doi.org/10.1016/j.measurement.2016.02.037. at the IADC/SPE Drilling Conference, New Orleans, Louisiana, . https://doi.org/l().
Rooki, R. , Rakhshkhorshid, M., 2016. Cuttings transport modelling in underbalanced oil 2118/59189-MS.
drilling operation using radial basis neural network. Egypt. J. Petroleum Elsevier Toreifi, H. , Rostami, H. , Manshad, A.K., 2014. New method for prediction and solving
Publishers. https://doi.org/10.1016/j.ejpe.2016.08.001. the problem of drilling fluid loss using modular neural network and particle swarm
Rooki, R. , Ardejani, ED., Moradzadeh, A. , Kelessidis, V.C., Nourozi, M. , 2012a. optimization algorithm. J. Petroleum Explor. Prod. Technol. 4, 371—379. https://doi.
Prediction of terminal velocity of solid spheres falling through Newtonian and non- org/10.1007/s13202-014-0102-5.
Newtonian pseudoplastic power law fluid using artificial neural network Int. J. Valipour, S. , Shokouhi, S., Skalle, P. , Aamodt, A., 2014. An overview of case-based
Mineral Process. reasoning applications in drilling engineering. Artif. Intell. Rev. 41 (Issue 3), 317—
110-111, 53-61. Elsevier Publishers. https://doi.org/10.1016/j.minpro.2012.03. 329 Springer Link Publishers.
012. Verikas, A., 2009. Soft Sensors for Monitoring. Intelligent Systems Lab, Halmstad
Rooki, R. , Ardejani, ED., Moradzadeh, A., 2014. Hole cleaning prediction in foam University, Halmstad, Sweden.
drilling using artificial neural network and multiple linear regression. Geomaterials 4, Wang, G. , Pu, X.L, Tao, H.Z., 2012. A support vector machine approach for the
47—53. https://doi.org/10.4236/gm.2014.41005. prediction of drilling fluid density at high temperature and high pressure. J.
Rooki, R. , Ardejani, ED., Moradzadeh, A. , Mirzaei, H. , Kelessidis, V. , Maglione, R., Petroleum Sci. Technol. 30 (Issue 5) Taylor and Francis Online.
Norouzi, M., 2012b. Optimal determination of rheological parameters for https://doi.org/10.1080/ 10916466.2011.578095.
Herschel—bulldey drilling fluids using genetic algorithms (GAs). Korea-Australia Wang, X. , Liu, B., Yu, P., 2015. Research on the preparation and mechanism of the
Theology J. 24 (Issue 3), 163-170. https://doi.org/10.1007/s13367-012-0020-3. organic montmorillonite and its application in drilling fluid. J. Nanomater Hindawi
Ross, T.J., 2010. Fuzzy Logic with Engineering Applications, third ed. John Wiley and Publishing Corporation. https://doi.org/l().1155/2015/514604.
Sons Ltd. Wang, Y. , Duan, M. , Wang, D. , Liu, J. , Dong, Y. , 2011. A model for deepwater
Rui, C. , Qian, L, Hu, Y. , 2013. Application of case-based reasoning technology in floating platform selection based on BP artificial neural networks. Int. Soc. Offshore
drilling risk prediction. CNKI J (China Academic Journals Electronic Publishers). Polar Eng.
Shahin, M.A., Jaksa, M.B., Maier, H.R., 2009. Recent advances and future challenges for Wolpert, D.H., Macready, W.G., 1997. No free lunch theorems for optimization. IEEE
artificial neural systems in geotechnical engineering applications. Adv. Artif. Neural Trans. Evol. Comput. 1 (Issue 1), 67-82. http://dx.doi.org/10.1109/4235.585893.
Syst. http://dx.doi.org/10.1155/2009/308239. Hindawi Publishing Corporation. Wu, J. , Torres- Verdin, C. , Sepehrnoori, K. , Delshad, M. , 2004. Numerical simulation
Schatsky, D. , Muraskin, C. , Gurumurthy, R. , 2014. Demystifying Artificial of mud-filtrate invasion in deviated wells. In: Paper (SPE 87919) Revised for
Intelligence. Publication from Paper SPE 71739, First Presented at the 2001 SPE Annual
What Business Leaders Need to Know about Cognitive Technologies. Technical Conference and Exhibition. SPE Reservoir Evaluation & Engineering,
Delloite University Press. https://dupress.deloitte.com/dup-us- New Orleans, pp. 143—154.
en/focus/cognitivetechnologies/what-is-cognifive-technology.html?id = Yunhu, L. , Mian, C. , Yan, J., Bing, H. , Lichun, J. , Hui, H. , 2011. Identification of
Leak Zone Pre-drilling Based on Fuzzy Control. Trans Century Training
Accessed date: 12 June 2017. programme Foundation. IEEE, pp. 353-356 pp.
Schlumberger, 2017. Mud Program. http://www.glossary.oilfield.slb.comnerms/m/ Zaimy, S.S., Rasaei, M.R., 2013. Reconstruction of porosity distribution for history
mud_program.aspx, Accessed date: 20 May 2017. matching using genetic algorithm. J. Petroleum Sci. Technol. 31 (Issue 11) (Taylor
Shadizadeh, S.R., Karimi, F. , Zoveidavianpoor, M. , 2010. Drilling stuck pipe prediction and Francis Online)
in Iranian oil fields: an artificial neural network approach. Iran. J. Chem. Eng. 7 Zhou, H. , Niu, X. , Fan, H., Wang, G. , 2016. Effective calculation model of drilling
(Number 4). fluids density and ESL) for HTHP well while drilling. In: Paper IADC/SPE-180573-
Shadravan, A. , Tarrahi, M. , Amani, M. , 2015. Intelligent tool to design fracturing, MS Presented at the 2016 IADC/SPE Asia Pacific Drilling Technology Conference,
drilling, spacer and cement slurry fluids using machine learning algorithms. In: Paper Singapore, https://doi.org/10.2118/180573-MS.
SPE 175238 Presented at the SPE Kuwait Oil and Gas Show and Conference, Thu, D. , Liu, G. , Zhang, Q. , 2013. Research of pre-warning pipe-sticking based on
Kuwait, . https://doi.org/10.2118/175238-MS. neural network. Appl. Mech. Mater. 325—326, 1734—1737.
Shahdi, A. , Arabloo, M. , 2016. Application of SVM algorithm for frictional pressure https://www.scientific.net/ AMM.325-326.1734.
loss calculation of three phase flow in inclined annuli. J. Petroleum Environ.
Biotechnol. 5 (Number 3).
Shelley, R. , Massengill, D. , Scheuerman, P. , McRill, P., Hamilton, R. , 1998. Granite
wash completion optimization with the aid of artificial neural networks. In: SPE
39814, Proceedings, Gas Technology Symposium, Calgary, Alberta, . https://doi.org/l
O.
2118/39814-MS.

321