IBP1871_18

MACHINE LEARNING METHODS APPLIED TO

RATE OF PENETRATION PREDICTION – A
TECHNICAL REVIEW
Luís Felipe F. M. Barbosa , Andreas Nascimento 2,
1

João A. de Carvalho Jr3 , Mauro H. Mathias4

Copyright 2018, Brazilian Petroleum, Gas and Biofuels Institute - IBP
This Technical Paper was prepared for presentation at the Rio Oil & Gas Expo and Conference 2018, held between
24 and 27of September, in Rio de Janeiro. This Technical Paper was selected for presentation by the Technical
Committee of the event according to the information contained in the final paper submitted by the author(s). The
organizers are not supposed to translate or correct the submitted papers. The material as it is presented, does not
necessarily represent Brazilian Petroleum, Gas and Biofuels Institute’ opinion, or that of its Members or
Representatives. Authors consent to the publication of this Technical Paper in the Rio Oil & Gas Expo and
Conference 2018 Proceedings.

Abstract

Bilgesu et al. (1997) published one of the first works that used neural networks to predict the
Rate of Penetration (ROP) of oil and gas wells with high accuracy. After a decade without
further development, other researchers started caring out similar studies on prediction of this
important drilling parameter. The ROP models can be classified as traditional models (based
on physics-models), or data-drive models, using methods from statistics, such as multiple
regression, or machine learning methods. This paper presents the preliminary results of a review
on machine learning methods applied to ROP prediction. It is shown that the use of machine
learning techniques outperforms other methods (regression or traditional models) in ROP
prediction.

Keywords: Drilling Optimization. Machine Learning.

1. Introduction

The estimation of Rate of Penetration (ROP) is an important step to optimize drilling

activities. For that, it is necessary to establish a model to determine how the drilling variables
affect the ROP. However, this task is quite complex, because, as pointed out by Mitchell and
Miska (2011), (i) the ROP depends on many variables, and (ii) the manner in which those
variables affect ROP is complex and only partially understood.
Many researches have been conducted to establish mathematical equations of ROP
behavior as a function of the most important drilling variables. Eren and Ozbayoglu (2010)
showed in a timeline that the scientific period of drilling optimization research started in 1950s,
emerging with subsequently many ROP models. One of the most important studies on drilling
optimization was developed by Bourgoyne and Young (1974), where a multiple regression
analysis of the drilling data was conducted to achieve minimum drilling costs by maximizing

______________________________
1
Mechanical Engineer - São Paulo State University (UNESP), School of Engineering, Department Energy,
Guaratinguetá - Av. Ariberto P. da Cunha, 333 – Guaratinguetá, SP CEP 12510410, Brazil
2
Ph.D., Petroleum Engineer – Petroleum Engineering Program, Department of Engineering and Technology,
Federal University of Espírito Santos (UFES), São Mateus, 29932-900, Brazil
3
Ph.D., Aerospace Engineer – São Paulo State University (UNESP), School of Engineering, Department Energy,
Guaratinguetá - Av. Ariberto P. da Cunha, 333 – Guaratinguetá, SP CEP 12510410, Brazil
4
Ph.D., Mechanical Engineer - São Paulo State University (UNESP), School of Engineering, Department
Mechanics, Guaratinguetá - Av. Ariberto P. da Cunha, 333 – Guaratinguetá, SP CEP 12510410, Brazil
 Rio Oil & Gas Expo and Conference 2018

the ROP in a typical operation. This model and other mathematical equations are referred along
the paper as traditional ROP models (see Section 4.1).
Relying solely on traditional ROP models leads to not high accurate ROP prediction,
since these models need empirical coefficients related, e.g., to formation characteristics. One
example of application of a traditional ROP model was carried out by Nascimento et al. (2015),
who applied Bourgoyne and Young ROP model (BYM) for presalt layer, and showed that
normalizing the factor of BYM formulation and allowing a wider range of applicable drillability
coefficients could decrease the relative error of the ROP simulation to the real data from 46 %
to 27 %, respectively. Other researchers (AMAR; IBRAHIM, 2012; BATAEE; IRAWAN;
KAMYAB, 2014; HEGDE et al., 2017) have shown that the use of machine learning methods
results in more accurate ROP prediction than those predictions obtained only from traditional
models. For that reason, the application of different techniques from the artificial intelligence
field has been applied in drilling parameters prediction in the past years, especially for ROP
prediction. The Figure 1 illustrates the evolution of the amount of published technical papers
per year related to ROP prediction/estimation, that employed other models different than the
traditional models. It is observed an increase of publications in last decade, although one of the
first works that applied neural networks to predict the ROP was performed by Bilgesu et al.
(1997).

Figure 1. A Compilation of a total of 45 works found by the authors on the literature considering thesis and
papers published in journals and congresses that predicted the ROP using other methods different than the
traditional models. Status: June 2018.

The use of artificial intelligence methods can be also applied to predict other drilling
parameters, such as weight on bit (KHOSRAVANIAN et al., 2016), pump pressure of the
drilling fluid (WANG; SALEHI, 2015), drilling fluid density (AHMADI, 2016), or to select the
optimum drilling parameters, e.g. the choice of the bit (YILMAZ et al., 2002). However, this
review will be limited to the ROP modeling. The main objective is to carry out a review of the
current progress on the use of machine learning methods to estimate the ROP, presenting the
preliminary results. Due to length limitation of this work, it is not possible to analyze and cite
all the 45 works that generated the Figure 1.

2. Searching Method

An extensive search was carried out on different databases, especially Scopus and One
Petro, in order to find those works that predict the ROP not using the traditional models,
reaching a total of 45 works as show in Figure 1. One challenge faced by the authors in this
stage was not a common keyword among the publications. In addition, using some keywords
2
 Rio Oil & Gas Expo and Conference 2018

led to publications from other areas, e.g. publications related to manufacturing engineering
appeared when searching for works with title “drilling”. That made not possible to find out all
of works showed in the timeline of Figure 1 with a simple search on any database. The process
of finding the articles started with a search on databases of Scopus and One Petro by using
different key words, such as, “drilling” & “optimization”, “drilling” & “machine learning” and
so on, what resulted in a few publications that predict the ROP using not the traditions models.
Then, by checking the references used in those works, additional papers could be found. Finally,
we checked in Scopus, which papers were citing those previous works. This approach made
possible to extend the amount of similar works.

3. Drilling Data

3.1. Drilling Daily Reports Versus Real-Time Drilling Data

In preliminary analysis, it was possible to observe that most of works used drilling data
from two sources: drilling daily reports (ANSARI et al., 2017, ESKANDARIAN et al., 2017)
or real-time drilling data (GANDELMAN, 2012, HEGDE et al. 2017, DIAZ et al. 2018)15.
The drilling data from daily reports are important source of information. However, the
downside when using daily reports is that only few observations are available to build predictive
models based on machine learning. For example, if it is required a data set with more than
hundred points, it will not always be possible to analyze drilling daily reports from an individual
well. While Diaz et al. (2018) used real-time drilling data from a drilled well with length of 4.6
km in South Korea with a total of 7043 observations in their study of ROP model, Ansari et al.
(2017) used drilling daily reports and needed to gather information from 19 wells drilled in
Persian Gulf in order to have available 248 points. The difference in the amount of observations
has a direct impact on training the predictive ROP model.

3.2. Drilling Data Domains

The real-time drilling data (RTDD) can be stored in time domain or in depth domain.
When using drilling data in time domain, the raw recording can also have data during non-
productive time. As consequence, manipulations are required to identify when the hole was
being drilled prior to assessing how the drilling variables affect the drilling rate, what is the
main goal of obtaining a ROP model. One of the few works that used RTDD in time domain to
predict drilling parameters was carried out by Fruhwirth et al. (2006), but the authors estimated
the pump pressure instead of the ROP. In their work, there is available only information about
data partition regarding the procedures for training, validation and testing, but no mention about
non-productive time in the original data set.
The depth domain is the common way of geologists to plot the logs against the depth.
Using the depth as index simplifies the data preparation, since only observations while drilling
are used in modeling. Some sampling rates found of RTDD when stored in depth domain were
0.25 ft (HEGDE al., 2017, HEGDE; GRAY, 2017) or 0.5 ft (NASCIMENTO et al. 2015,
NASCIMENTO et al., 2016).

4. Methods to predict the ROP

15
Only some works were cited due to length limitation of this paper, as already mentioned in Section 1. Citing
solely all publications used to generate the Figure 1 would result in a reference section of roughly four pages long.

3
 Rio Oil & Gas Expo and Conference 2018

Hegde et al. (2017) proposed a simple classification of the ROP models, which will be
used in this work. The ROP models can be classified as: traditional (physics-based) models or
data-driven models (regression models and machine learning methods).

4.1. Traditional Models

Here, those ROP models which try to stablish mathematical equations among the
drilling variables are called as traditional models, because some of them appeared in the initial
phase of the scientific research of drilling optimization. In additional, those models do not rely
solely on the drilling data, as the machine learning methods. As consequence, it is possible to
find out other researches (HEGDE et al. 2017) calling them as physics-based-models, or as
simple as drilling models (HARELAND; HOBEROCK, 1993).
There are many models describing the effects of several parameters on ROP. Eren and
Ozbayoglu (2010) provided a literature survey, explaining some of the traditional models. The
readers can also refer to other sources (HARELAND; HOBEROCK, 1993, MITCHELL;
MISKA, 2011) in order to see the details of the traditional models.
Some of traditional models which are worth of citing, because they are used in some
papers for comparison purpose with data-driven models, are the following: Graham and
Muench (1959); Mauer (1962); Bingham (1965); Young (1969), Bourgoyne and Young (1974);
Warren (1987); Hareland and Rampersad (1994).
As already mentioned, the actual relationship among the drilling variables is not very
well understood (MITCHELL; MISKA, 2011). Therefore, some efforts (MOTAHHARI et al.,
2010, DENG et al., 2016) have been made recently to understand better the relationship
between the drilling variables and how they affect the ROP. Deng et al. (2016) proposed a
theoretical model for determining the ROP for roller cone bit, and this model was validated
with lab drilling results. The authors used the rock dynamic compressive strength instead of
static compressive strength, what increased the accuracy of the theoretical model.

4.2. Multiple Regression Applied to Predict the ROP

Multiple regression can be applied to model the ROP as functions of drilling

parameters. In this case, the dependent variable (ROP) is a function of independent variables
(X1, X2, … Xk), which are representing the selected drilling variables as inputs into the ROP
model. Moraveji and Naderi (2016) applied the full quadratic form of multiple regression,
having linear, quadratic and interaction coefficients. The ROP was modeled as function of six
drilling parameters, as showed in Table 1.

Table 1. ROP modeling using multiple regression carried out by Moraveji and Naderi (2016).

Drilling Data Source Amount of Inputs Model Accuracy

1732 cases from one Full quadratic form of R2 = 71.5 %
6
southern Iranian gas field multiple regression R2 adj = 71,1%

It was also found other works (ARABJAMALOEI; SHADIZADEH, 2011;

ESKANDARIAN et al., 2017, HEGDE et al., 2017) that used multiple regression, with either
linear coefficients or both linear and non-linear coefficients, in order to compare the results
obtained with learning techniques from artificial intelligence.
4
 Rio Oil & Gas Expo and Conference 2018

4.3. Machine Learning applied to predict the ROP

In our preliminary analysis, it was possible to observe that several methods from
artificial intelligence have been applied to model the ROP. Here, we will focus mainly on those
studies that compared the ROP prediction using machine learning techniques with other
methods. A total of 10 works from 45 are summarized in the Table 2.

Table 2. Some works that used Machine Learning Methods to predict the ROP.

Machine Testing
Drilling Data Amount of Compared with
Authors Learning Best Model Accuracy
Source Inputs in ML other methods?
Method of ML
Laboratory Data - 1st case –
Neural
8000 points from 1st case - 10 R = 98.2%
Networks - No -
rig floor 2nd case - 6 2nd case –
Bilgesu SLFN
simulator R = 95.5%
(1997)
Neural 1st case –
USA - Field Data
Networks - 1st case - 10 R = 96.5%
from several well No -
SLFN 2nd case - 8 2nd case –
(500 points)
R = 98.0%
Arabjamaloei Iran - 35 wells Yes, with
Neural
and drilled in Ahwaz multiple
Networks - 7 Yes R2 = 74.0%
Shadizadeh oilfield (330 regression and
SLFN
(2011) cases) BYM
Mediterranean
Neural
Amar and Sea - Offshore,
Networks - APRE =
Ibrahim provided by Eren 7 Yes, with BYM Yes, RBF
RBF and 9.6 %
(2012) and Ozbayoglu
ELM
(2010)
Adaptive
Turkey – 7
neuro-fuzzy Yes, with linear RMSE =
Basarir et al. boreholes drilled
inference 4 multiple Yes 0.33
(2014) in 6 different
system regression (validation)
regions
(ANFIS)
15 wells (1800 Neural Yes, with
Bataee et al. R2 = 85.7 %
points for Networks - 5 Bingham, BYM, Yes
(2014) (validation)
training) MLP Warren
Support
Ansari et al. Persian Gulf - 19 8 after feature
vector No - R = 90.6%
(2017) wells (248 points) selection
regression
1st case – 5 Yes, with 1st case –
Neural
Bezminabadi Iran – Azadegan 2ns case – 9 multivariate R = 75%
Networks - Yes
et al. (2017) Oilfield (added rock nonlinear 2nd case –
SFLN
parameters) regression R = 86%
4 or 6 inputs
3 methods - R2 = 80.1%
Iran - 5 wells in after feature Yes, with linear
Eskandarian Cubist, RF Yes, MON- (10 fold
South West (226 selection of 13 multiple
et al. (2017) and MON- MLP cross-
points) non-constant regression
MLP validation)
variables
Yes, with
Normalized
traditional
Error
models
Hegde et al. USA – 1 vertical (Median for
RF 4 (Bingham, Yes, RF
(2017) well all
Motahhari,
formation)
Hareland) and
~ 0.13
linear regression
5
 Rio Oil & Gas Expo and Conference 2018

Table 2. Some works that used Machine Learning Methods to predict the ROP (Continuation)

Machine Testing
Drilling Data Amount of Compared with
Authors Learning Best Model Accuracy
Source Inputs in ML other models?
Method of ML
South Korea - 1 Neural
Diaz et al. 7 after feature R2 = (90%,
well (7034 Networks - Yes, with BYM Yes
(2018) selection 99%)
points) MLP

SFLN: single hidden-layer networks (a MLP with one hidden-layer)

RBF - radial basis function networks
ELM - extreme learning machines
APRE - absolute percent relative error
RF – random forest
MON-MLP – monotone multi-layer perceptron
MLP – multi-layer perceptron (with more than one hidden-layer)

5. Discussion on ROP models

Although 10 of 45 works were briefly analyzed, interesting findings can be already

mentioned, as follows.

5.1. Machine learning algorithms outperforms other methods

In the Table 2, it is possible to observe that the 8 out of 10 works compared machine
learning techniques with other techniques (traditional models and/or regression models) in ROP
modeling. In all works that performed this comparison, the use of learning algorithms provided
a better prediction of the ROP. The reason for that is the capability of those models to capture
non-linear relationship among the variables.

5.2. Machine learning methods applied to ROP prediction

In this selected sample of works (Table 2), most of them applied neural networks to
predict the ROP. However, it is not possible yet to affirm which method is preferred or more
employed in this type of study due to small amount of works that are being analyzed here. Other
methods, such as support vector regression and Random Forest (RF), were employed also in
the ROP prediction tasks.
While the use of neural networks results in predictive model with good generalization
capability as seen in Table 2, the use of them leads to complex models, which are not easy to
be understood due to amount of weights and biases, which are expressed in matrices equations.
For that reason, the neural networks are commonly referred as black-box models. As pointed
out by Eskandarian et al. (2017), “rule-extraction is one of the benefits of using RF models in
comparison to black-box models such as ANN [Artificial Neural Networks]” (p. 612). Basarir
et al. (2014) employed adaptive neuro-fuzzy inference system, from which some simple rules
can be extracted, proving valuable information about how the drilling variables affect the ROP.

5.3. Parametric Influence

The rule extraction is one way to assess the influence of the drilling variables on the
ROP. Another way is to perform parametric influence analysis, varying some drilling variables

6
 Rio Oil & Gas Expo and Conference 2018

while the others remain unchanged, as done by Eskandarian et al. (2017), understanding the
relationship among the variables.

5.4. The most important inputs

Some studies (BILGSEU, 1997, ANSARI et al., 2017, BEZMINABADI et al., 2017,
ESKANDARIAN et al., 2017) analyzed which combination of inputs provided a better ROP
prediction. This type of analysis is commonly known as feature selection, because, through
some method, only those variables with significant impact on the prediction are selected. Hegde
et al. (2017) showed a comprehensive way of assessing the most important parameters in the
regression by plotting the importance of the inputs. However, it is possible to find some studies
that only mentioned the use of the best combination of inputs (ARABJAMALOEI;
SHADIZADEH, 2011, DIAZ et al., 2018), or even that did not provide any information
regarding the selection of the drilling variables (AMAR; IBRAHIM, 2012).

5.5. Data partition

A common approach of those works from the Table 2 was to split the drilling data into
two data set: one for training (when required, part of this training set is used in the validation
process during the training) and another for testing to assess the generalization capability of the
predictive model. This data partition is randomly carried out, and can be applied as post-analysis
or prior to drilling a well with similar conditions of those used to build the ROP models.
Hegde et al. (2017) proposed another way to partition the data into training and testing
data set, so that it can be used in a real-time environment, and the trained model can be
employed to optimize the drilling activities by finding out the optimum values of the
controllable drilling variables, such as weight on bit and bit rotation speed, which maximize the
ROP (HEGDE; GRAY, 2017). Expect for this try, it was not found other works that tried to
build the ROP model while the drilling data are generated and sent to the rig crew.

6. Conclusions

As seen in the Table 2, the works that compared different methods to predict the ROP
concluded that the machine learning methods outperformed other methods (regression or
traditional models) in this type of task.
The use of machine learning techniques is possible to extract rules that represents the
relationship among the drilling variables, what can be help the drill crew to select the optimum
drilling variables. This can be done through parametric influence analyses.
Most of works carried out historical analysis for wells drilled in a similar region, what
can be employed as post-analyses or prior to drilling a similar well. However, few tries have
been conduct to build the ROP model while the drilling data are being generated, as done by
Hegde et al. (2017) and Hegde and Gray (2017).

7. Acknowledgements

The authors would like to thank the financial support from CAPES and CNPQ.

7
 Rio Oil & Gas Expo and Conference 2018

8. References

AHMADI, M. A. Toward reliable model for prediction drilling fluid density at wellbore
conditions: A lssvm model. Neurocomputing, v. 211, p. 143–149, 2016. DOI:
https://doi.org/10.1016/j.neucom.2016.01.106

AMAR, K.; IBRAHIM, A. Rate of Penetration Prediction and Optimization using Advances in
Artificial Neural Networks, a Comparative Study. In: Proceedings of the 4th International
Joint Conference on Computational Intelligence, Barcelona, Spain, 2012, p. 647-652. DOI:
https://doi.org/10.5220/0004172506470652

ARABJAMALOEI, R.; SHADIZADEH, S. Modeling and Optimizing Rate of Penetration

Using Intelligent Systems in an Iranian Southern Oil Field (Ahwaz Oil Field). Petroleum
Science and Technology, v. 29, n. 16, p. 1637–1648, jul 2011. DOI:
https://doi.org/10.1080/10916460902882818

ANSARI, H. R.; HOSSEINI, M. J. S.; AMIRPOUR, M. Drilling rate of penetration prediction

through committee support vector regression based on imperialist competitive algorithm.
Carbonates and Evaporites, v. 32, n. 2, p. 205–213, 2017. DOI:
https://doi.org/10.1007/s13146-016-0291-8

BASARIR, H.; TUTLUOGLU, L.; KARPUZ, C. Penetration rate prediction for diamond bit
drilling by adaptive neuro-fuzzy inference system and multiple regressions. Engineering
Geology, v. 173, p. 1–9, 2014. DOI: https://doi.org/10.1016/j.enggeo.2014.02.006

BATAEE, M.; IRAWAN, S.; KAMYAB, M. Artificial Neural Network Model for Prediction
of Drilling Rate of Penetration and Optimization of Parameters. Journal of the Japan
Petroleum Institute, v. 57, n. 2, p. 65–70, 2014. DOI: https://doi.org/10.1627/jpi.57.65

BEZMINABADI, S. N. et al. Effect of rock properties on rop modeling using statistical and
intelligent methods: A case study of an oil well in southwest of iran. Archives of Mining
Sciences, v. 62, n. 1, jan 2017. DOI: https://doi.org/10.1515/amsc-2017-0010

BILGESU, H. I. et al. A new approach for the prediction of rate of penetration (rop) values. In:
SPE Eastern Regional Meeting, 22-24 October, Lexington, Kentucky. Society of Petroleum
Engineers, 1997. DOI: https://doi.org/10.2118/39231-MS.

BINGHAM, M. G. A New Approach to Interpreting Rock Drillability. Petroleum Publishing

Company, 1965. 93 p.

BOURGOYNE, A.T.; YOUNG, F. S. A Multiple Regression Approach to Optimal Drilling and

Abnormal Pressure Detection. Society of Petroleum Engineers Journal, v. 14, n. 4, p. 371–
384, 1974. DOI: https://doi.org/10.2118/4238-PA

DENG, Y. et al. Theoretical and experimental study on the penetration rate for roller cone bits
based on the rock dynamic strength and drilling parameters. Journal of Natural Gas
Science and Engineering, v. 36, p. 117–123, nov 2016. DOI:
https://doi.org/10.1016/j.jngse.2016.10.019

8
 Rio Oil & Gas Expo and Conference 2018

DIAZ, M. B. et al. Drilling data from an enhanced geothermal project and its pre-processing
for rop forecasting improvement. Geothermics, v. 72, p. 348–357, 2018. DOI:
https://doi.org/10.1016/j.geothermics.2017.12.007

EREN, T.; OZBAYOGLU, M. E.. Real Time Optimization of Drilling Parameters During
Drilling Operations. In: SPE Oil and Gas India Conference and Exhibition, 20-22 January,
Mumbai, India, 2010. DOI: https://doi.org/10.2118/129126-MS

ESKANDARIAN, S.; BAHRAMI, P.; KAZEMI, P. A comprehensive data mining approach to

estimate the rate of penetration: Application of neural network, rule based models and
feature ranking. Journal of Petroleum Science and Engineering, v. 156, p. 605–615, 2017.
DOI: https://doi.org/10.1016/j.petrol.2017.06.039

FRUHWIRTH, R. K.; THONHAUSER, G.; MATHIS, W. Hybrid simulation using neural

networks to predict drilling hydraulics in real time. In: SPE Annual Technical Conference
and Exhibition, 24-27 September, San Antonio, Texas, USA. Society of Petroleum Engineers,
2006. DOI: https://doi.org/10.2118/103217-MS.

GANDELMAN, R. A. Predição da ROP e Otimização em Tempo Real de Parâmetros

Operacionais na Perfuração de de Poços de Petróleo Offshore (ROP prediction and real-
time optimization of operational parameters on drilling offshore oil Wells), MS thesis,
UFRJ, Brazil. (Federal University of Rio de Janeiro), 2012. (in Portuguese)

GRAHAM, J.; MUENCH, N. Analytical Determination of Optimum Bit Weight and Rotary
Speed Combinations. In:Fall Meeting of the Society of Petroleum Engineers of AIME.
Society of Petroleum Engineers, 1959. DOI: https://doi.org/10.2118/10.2118/1349-G

HARELAND, G.; HOBEROCK, L. Use of Drilling Parameters To Predict In-Situ Stress

Bounds. In: SPE/IADC Drilling Conference, 22-25 February, Amsterdam: Society
ofPetroleum Engineers, 1993. DOI: https://doi.org/10.2118/25727-MS

HARELAND, G.; RAMPERSAD, P. Drag - Bit Model Including Wear. In: SPE
LatinAmerica/Caribbean Petroleum Engineering Conference. Buenos Aires, Argentina:
Society of Petroleum Engineers, 1994. DOI: https://doi.org/10.2118/26957-MS

HEGDE, C., et al. Analysis of rate of penetration (ROP) prediction in drilling using physics-
based and data-driven models. Journal of Petroleum Science and Engineering, v. 159, p.
295–306, 2017. DOI: https://doi.org/10.1016/j.petrol.2017.09.020

HEGDE, C.; GRAY, K. E. Use of machine learning and data analytics to increase drilling
efficiency for nearby wells. Journal of Natural Gas Science and Engineering, v. 40, p.
327–335, 2017. DOI: http://dx.doi.org/10.1016/j.jngse.2017.02.019

KHOSRAVANIAN, R. et al. Weight on drill bit prediction models: Sugeno-type and

Mamdani-type fuzzy inference systems compared. Journal of Natural Gas Science and
Engineering, v. 36, p. 280–297, nov 2016. DOI:
http://dx.doi.org/10.1016/j.jngse.2016.10.046

9
 Rio Oil & Gas Expo and Conference 2018

MAURER, W. The "Perfect - Cleaning" Theory of Rotary Drilling. Journal of Petroleum

Technology, v. 14, n. 11, p. 1270–1274, nov. 1962. DOI: https://doi.org/10.2118/408-PA

MITCHELL, Robert F.; MISKA, Stefan Z. Fundamentals of Drilling Engineering.

Richardson, TX: Society of Petroleum Engineers, 2011

MORAVEJI, M. K.; NADERI, M. Drilling rate of penetration prediction and optimization

using response surface methodology and bat algorithm. Journal of Natural Gas Science
and Engineering, v. 31, p. 829–841, apr 2016. DOI:
http://dx.doi.org/10.1016/j.jngse.2016.03.057

MOTAHHARI, H.; HARELAND, G.; JAMES, J. Improved Drilling Efficiency Technique

Using Integrated PDM and PDC Bit Parameters. Journal of Canadian Petroleum
Technology, v. 49, n. 10, p. 45–52, oct 2010. DOI:
http://dx.doi.org/10.1016/10.2118/141651-PA

NASCIMENTO, A. et al. Mathematical Modeling Applied to Drilling Engineering: An

Application of Bourgoyne and Yound ROP Model to a Presalt Case Study. Mathematical
Problems in Engineering, v. 2015, Article ID 631290, p. 9, 2015. DOI:
http://dx.doi.org/10.1155/2015/631290

NASCIMENTO, A. et al. Reverse Engineering: A New Well Monitoring and Analysis

Methodology Approaching Playing-Back Drill-Rate Tests in Real-Time for Drilling
Optimization. Journal of Energy Resources Technology, v. 139 (1), p. 12902-12902-5,
2016. DOI: http://dx.doi.org/10.1115/1.4033067

PAYETTE, G. S. et al. A Real-Time Well-Site Based Surveillance and Optimization Platform

for Drilling: Technology, Basic Workflows and Field Results. In: SPE/IADC Drilling
Conference and Exhibition, 14-16 March, The Hague, The Netherlands, 2017. DOI:
https://doi.org/10.2118/184615-MS

WANG, Y.; SALEHI, S. Application of real-time field data to optimize drilling hydraulics
using neural network approach. Journal of Energy Resources Technology, v. 137, n. 6, p.
062903–062903–9, 2015. DOI: https://doi.org/10.1115/1.4030847

WARREN, T. Penetration Rate Performance of Roller Cone Bits. SPE Drilling Engineering,
v. 2, n. 01, p. 9–18, mar 1987. DOI: https://doi.org/10.2118/13259-PA

YILMAZ, S.; DEMIRCIOGLU, C.; AKIN, S. Application of artificial neural networks to

optimum bit selection. Computers and Geosciences, v. 28, n. 2, p. 261–269, 2002. DOI:
https://doi.org/10.1016/S0098-3004(01)00071-1

YOUNG, F. Computerized Drilling Control. Journal of Petroleum Technology, v. 21, n. 04,

p.483–496, apr 1969. DOI: https://doi.org/10.2118/2241-PA

10