SPE-197157-MS

Drilling in the Digital Age: An Aproach to Optimizing ROP Using Machine

Learning

Peter Batruny, Hanif Yahya, Norazan Kadir, Amir Omar, Zahid Zakaria, Saravanan Batamale, and Noreffendy
Jayah, PETRONAS

Copyright 2019, Society of Petroleum Engineers

This paper was prepared for presentation at the Abu Dhabi International Petroleum Exhibition & Conference held in Abu Dhabi, UAE, 11-14 November 2019.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Low rates of penetration (ROP) were experienced in an area with well-known lithology. The vast drilling
experience and similarity of drilling conditions in the area, provided the operator with enough data to
improve the well schedule and cost performance through the use of machine learning.
Machine learning, specifically artificial neural networks (ANN), is a statistical tool to find relations
between multiple inputs. Details that would have been missed or considered outliers by a mathematical
model can be accounted for and explained in the ANN model. The ANN was trained on thousands of real
time data points recorded from selected wells in a specific depth interval. Typical drilling parameters such
as weight on bit, rotary speed, bit hydraulics, lithological properties, and dogleg severity were the input
parameters chosen in the model to generate ROP. Once the model was calibrated to historical data, it was
used to find the best parameters to maximize ROP.
R squared factors were 0.729 and 0.675 for 12.25 in. and 17.5 in. sections repectively. This was achieved
with an ANN structure of 2 hidden layers consisting of 5 nodes each. Sensitivity analysis identified bit
hydraulics, weight on bit, and rotary speed as the major parameters impacting ROP. The ROP model was
used to conduct a “virtual drill-off test” to identify drilling parameters that maximize ROP. ROP dependency
on weight on bit and lithological analysis suggests bit design can be further improved. Bit hydraulics showed
that higher flow rate was needed in sections with higher overbalance. Optimum drilling parameters were
tested on four wells and resulted in more than 50% higher ROP compared to original field data.
In an industry increasingly dominated by big data, separating the clean data from the “noise” will be a
vital topic. This paper aims to provide a blueprint for the use machine learning to optimize ROP in a manner
that is simple and easily replicated.

Introduction
Drilling days per thousand feet (DDPTF) is a measure of drilling performance of a well. It is compared with
the benchmark value set by the company to determine the performance of a particular well in relation to
other wells with similar complexity. DDPTF is heavily influenced by the rate of penetration (ROP). Low
ROP results in more days required to drill a well which in turn yields a lower DDPTF value.
2 SPE-197157-MS

The oil field in this case study is located onshore in the south of Iraq. The field contains a large number
of exploration and development wells. Formations in the studied oil field consist mainly of limestone and
shale, the field lithology is presented with respective true vertical depth (TVD) in reference to rotary kelly
bushing (RKB) (Table A—1). In this study, ROP shall refer to instantaneous ROP. The studied oil field
average ROP is 8.6 m/hr in the 12.25 in. section and 13.9 m/hr in the 17.5 in. section. An Unconfined
Compressive Strength (UCS) study on a section of the studied oil field showed that most of the UCS values
lie within 5000 and 10000 psi (Fig. A—1), which is low to medium strength formation. An investigation
was launched to determine the cause of low ROP and the possible solution to the problem.
All wells in the studied field are directional wells with either J or S shaped profiles, with the build in the
17.5 in. section and the drop in the 12.25 in. section (in S shaped profiles). All wells encounter the same
formation top true vertical depth (TVD), are drilled with similar polycrystalline diamond bits (6 blades with
16mm cutters), and use positive displacement motors (PDM) as the drive method.

Statement of Theory and Definitions

Several methods of optimizing ROP have been used and documented in the literature. These methods include
optimizing bit selection, and ROP mathematical modelling. The advancement of computing technology has
led to the transition to machine learning techniques such as Artificial Neural Networks (ANN) and Expert
System (Curry 2004), (Gidh 2011).

Mathematical Models
Currently, several mathematical models for ROP exist in the literature. These models are based on a
mathematical relationship between ROP and the parameters that influence it.
Motahhari ROP model (Eq.1) is a function of bit wear, bit coefficient, bit diameter, RPM, rock strength,
and weight on bit (WOB) (Motahhari 2010).
(1)
Where, RP is the ROP in m/hr, WB is the weight on bit 1000 pound force, DB is the bit diameter in
inches, S is the confined rock strength pounds per square inch, G is the bit coefficient determined by bit
design, cutter size, bit geometry, and cutter-rock coefficient of friction, Nt is the combined surface and motor
RPM in revolutions per minute, and Wf is the non-dimensional wear function. Wf can be calculated using
different equations from (Liu 2014). In the Motahhari model, a decrease in Wf or G, while keeping other
parameters constant results in a decrease in ROP. The inverse happens when bit size or compressive strength
is decreased. The relationship between WOB, RPM, and ROP is not a straightforward one; the exponents
can yield an optimum value for WOB and RPM due to the exponential nature of the relationship.
Another mathematical model that is available is the Bingham model (Eq. 2).

(2)

The coefficeints a, b and c can be found by adjusting the coefficeints to fit historical drilling data. The
relationship between WOB, RPM, and ROP in the Bingham model is similar to that in the Motahhari model
suggesting a certain optimum value for WOB and RPM to maximize ROP.
Bourgonyne and Young developed a model (Eq. 3) that is a function of RPM, WOB, bit hydraulics, bit
wear, formation strength, compaction, and pore pressure. This model is the most complete mathematical
model because they used eight functions to represent the effect of all parameters in drilling. Initially this
model is developed for modelling roller cone cutter but it is proven that it can also be used in modelling
ROP for polycrystalline diamond bits (Bataee 2010).
(3)
SPE-197157-MS 3

Where dTVD is TVD in meters, ρ is mud weight (MW) in pouns per gallon, ρpore and ρECD are MW equivalent
pore pressure and circulating respectively. (WB/DB)t is the threshold pound force per inch of bit diameter
where the bit starts drilling, h is the tool dullness fraction and JF is the jet impact force (JIF) in lbf. Constants
a1 through a8 can be found by adjusting the coefficeints to fit historical drilling data. A comparison was
done between the Bourgonye and Bingham model in (Bataee 2010) and found that Bourgoyne model, while
much more complex, resulted in a more accurate ROP prediction than Bingham model.
All three models discussed above have common parameters that affect ROP: RPM, WOB, and bit size.
Other parameters are compressive strength, overbalance, jet impact force, bit wear, and bit design.

Machine Learning Models

Conventionally, ROP is predicted based on mathematical models or experience from similar parameters
in offset wells. Jamshidi (2013) developed two Artificial Neural Network models. One that determines bit
selection based on set drilling parameters and desired ROP as input. The other model uses optimum drilling
parameters and bit selection to determine optimum ROP. ANN is a machine learning model that is based
on the model of the human neuron that consist of dendrites, cell body, synapses and axon as shown in Fig.
1. Each component in the human neural network is represented in the artificial neural network model. The
dendrites act as input nodes, cell body represents activation function, synapse is the weightage of each input
and the axon out is the output node. The ANN is shown in Fig. 2.

Figure 1—Model of a human neuron

Figure 2—ANN Model

A typical artificial neural network usually consists of input, hidden, and output layers. ANN learns by
collecting and analyzing input data and finding correlation of input data and output. The artificial neural
network’s hidden layer iterates the weightage of each input and multiplies the input with the weightage to
get the output, this iterative process continues until a reasonably close correlation is achieved (Jamshidi
2013). The output is written as in Eq. 4.
(4)
4 SPE-197157-MS

Where Y is the output of neuron, wi is the weightage of i-th input, xi is the input value of i-th input, and F
is a non-linear activation function. The input layer consists of input nodes, each input node represents one
parameter. The hidden layer consists of hidden nodes; there can be more than one hidden layer in an ANN.
Bilgesu (2001) concluded that in the case of ANN for selecting bit based on desired ROP to be achieved,
a 3 layered hidden network with 10,10,1 structure gives the lowest prediction error when compared to 2-
layered and 4 layered Neural Networks.
Valisevich (2015) trained ANN with ROP as an output when bit wear is an input and another model
with bit wear as an output when ROP is an input. Inputs that don’t change are WOB, RPM, flow rate, rock
properties, and bit wear.
Hegde (2018) reports that drilling parameters can be divided into three categories: controllable
parameters, uncontrollable parameters, and responding parameters. Control parameters are the input that can
be manipulated such as WOB, RPM, and flow rate. Uncontrollable parameters cannot be changed such as
lithology, unconfined rock strength, principal stresses, and maximum pump output. Responding parameters
are the parameters that are affected by the controlable and uncontrollable input parameters such as ROP,
drillstring vibration, TOB, and Minimum Specific Energy.

Description and Application of Equipment and Processes

Data Collection
ASCII drilling data is produced from every well. This data log contains crucial information on the surface
from sensors. Data from twelve wells in six wellpads are collected. The data is then split into data from
17.5 in. sections (between 600 and 1500 mTVDRKB) and 12.25 in. sections (between 1500 and 2400
mTVDRKB). Spliting the data into two sections eliminates bit size as an extra input. Different sections from
different wells are combined to produce two datasets: 14,817 and 14,049 depth points from the 17.5 and
12.25 in. sections respectively. Data cleaning was done by removing incomplete, inconsistent and corrupted
data. Inconsistent data such as zero RPM with ROP value of more than zero is removed assuming drilling is
always done in rotary or sliding mode. Data with negative overbalance reading is removed from the dataset,
because the scope focuses on conventional overbalance drilling. Outliers can be identified by values that
are more than the maximum value and less than the minimum value. Maximum and minimum values can
be calculated with Eqs. 5 through 7.
(5)
(6)
(7)
Where xmax and xmin are the new maximum and minimum data points, Q1 and Q3 and are the first and
third quartile values, and >Q3,1 is the interquartile range.
Based on the literature review, common parameters that influence ROP are identified. These parameters
are WOB, lithology, bit size, downhole RPM, bit wear, rock strength, overbalance, flow rate, and JIF.
Because this research looks at deviated wells, dogleg severity (DLS) is also added as a paremeter to consider.
Following Hedge’s classification method these parameters are classified into 3 categroies. The controllable
parameters defined as parameters that can be controlled from surface facilities, are WOB, flow rate, surface
RPM, bit total flow area (TFA), and mud weight (MW). The uncontrollable parameters are pore pressure,
and rock strength. The responding parameters are overbalance, JIF, bit horsepower per square inch (HSI),
bit wear, DLS, and motor RPM.
Bit hydraulics such as TFA and HSI are related to MW, TFA, flow rate, and bit pressure loss (BPb). Eqs.
8 through 11 show the steps to calculate HSI and JIF (Wright 2003) (Warren 1986).
SPE-197157-MS 5

(8)

(9)

(10)

(11)

where q is flow rate in gallons per minute, Vn is the average nozzle velocity feet per secod, and An is the
TFA in square inches.
TVD, WOB, surface RPM, MW, and flow rate are input parameters acquired directly from raw data.
Survey data is taken every 30m. To match the required depth resolution in the ASCII data, an industry
directional drilling software is used to interpolate, inclination, azimuth, and DLS.

Model
Due to the amount of input parameters and data acquired, using a mathematical relation to predict and
optimize ROP will be difficult. Therefore, the ANN method is chosen to predict ROP based on the various
inputs.
Input values are normalized to avoid reaching the maximum limit of iteration, normalization also brings
positive effect to training time of Artificial Neural Network. This study uses the MinMax normalization
method represented in Equation (12)).
(12)
where xnorm is the normalized value of the input x. The 17.5 and 12.25 in. section datasets are normalized,
randomized, then split into training and testing data by 90% and 10% respectively. The purpose of
randomization is to ensure that depth dependant parameters are not separated by depth when the training
and testing data are split from the parent datases.
The software used in the study is R Studio which uses the R programming language (R Core Team 2013)
and the neural network package used by this study within R Studio is NeuralNet (Beck 2018). The ANN
within NeuralNet takes user defined input and ouput parameters. The architecture, such as number of hidden
layers and nodes, and threshold value specifying the threshold for the partial derivatives of the error function
as stopping criteria are also specified by the user. The separated training data is then fed to the ANN to find
the correlation between ROP as output and the different parameters as inputs. The resultant model is then
applied to the test data and compared to the actual test data values to determine the accuracy of the model.
Multipe benchmarks for accuracy can be used, in this study the R2 correlation method is used. The coefficient
of multiple correlation (R2) can be determined by Eqs. 13 through 15 extracted from (Bilgesu 2001).

(13)

(14)
(15)
SSE and SSy are sum of square error and sum of square of the ANN. The perfect ANN will return a
coefficient of multiple correlation (R2) of one and a good fit will return a value close to one (Bilgesu 2001).
The input parameters, hidden nodes, hidden layers, and threshold value can be manipulated in an iterative
process to achive a correlation value close to one. Once this value is achieved, generalized weights for every
input parameter value can be plotted. These generalized weights show the magnitude and relationship of the
6 SPE-197157-MS

input parameter to ROP. A generalized weight value of zero means the input parameter value has no effect
on ROP. A positive or negative generalized weight value means that ROP is positively or negatively affected
by the input parameter value respectively. This relationship is important when simulating ROP because it
has to conform to the relationships in the literature to some degree. The model can find statistical relations
between input parameters and ROP that might work for a particular set of data, but it might not be physically
correct. The ANN model would then have to be rerun or the ANN architecture would have to be changed.
This can be a time consuming task. For the purpose of simplification, this study starts by selecting the
number of input parameters that results in the highest correlation factor with one hidden layer and one node.
Once the input parameters are set, the number of nodes are added while the correlation factor continues
to increase. Once the number of nodes that achieves the highest correlation factor is found, a second layer
is added and the number of nodes in that layer are increased until the correlation factor stops increasing.
One or more input parameter generalized weights are verified with mathematical models mentioned in the
literature review, and the model is finalized.

Optimization
Once the final model is achieved, a sensitivity analysis can be conducted. Sensitivity analysis is conducted
by varying one parameter while keeping other parameters constant. This has been done in the field by the
driller and is referred to as a drilloff test. While the drilloff test in the field can be time consuming, using
the model requires no rig time and the parameters can be given to the driller before the start of the section.
In this study, the value of the fixed parameters is the most frequent value in the dataset. The variable
input value lies between the minimum and maximum values of that parameter in the dataset. This ensures
that the optimum parameters lie within the historical data and can be applied on existing equipment without
exceeding the equipment rating.
After the optimum parameters for both 17.5 and 12.25 in. sections are obtained, they are modeled
and compared with the model of the frequently occurring parameter values. The comparison provides a
prediction of the potential ROP increase if the optimum parameters are to be maintained while drilling.
The optimum parameters are applied on the field and the resultant ROP data is plotted against predicted
optimum ROP and most frequent ROP values.

Presentation of Data and Results

Model Performance
The network architecture that results in the highest correlation factor within the time limits of this study is
an ANN with a correlation factor of 0.729 for the 12.25 in. section, and 0.675 for the 17.5 in. section. This
was achieved with an input layer consisting of seven inputs: TVD, WOB, Total RPM, DLS, JIF, HSI, and
Overbalance. While the correlation factor might not be considered high, other factors such as model fit to
trends in ROP as well as conformity to physical relationships are taken into account (Jones 2018).
The resultant networks for 17.5 in. and 12.25 in. are detailed in Figs. 3a and 3b respectively.
SPE-197157-MS 7

Figures 3a & 3b—17.5 in. and 12.25 in. Section Neural Network Architecture

Each individual well data is run through the model and the actual ROP compared to the simulated ROP.
summarizes the results of the errors for all 10 wells from both 12.25 in. and 17.5 in. sections.

Table 1—Summary of Model Performance

The mean error % seems to exhibit a linear relationship with the range of the values (Fig. 4). The larger
the ROP spread, the higher the average percent error. The model in this study had an average absolute error
of 38% which is lower than that obtained by Bingham, Bourgoyne, and Warren models in other studies
(Bataee 2010).

Figure 4—Percent Error Variation with ROP Data Range

The model was also able to simulate ROP trends that might seem random to the naked eye. Fig. 5 shows
the comparison between actual and simulated values of the test data sets for 12.25 in. and 17.5 in. sections.
8 SPE-197157-MS

Figure 5—Actual vs Simulated ROP Using Test Data

Even though the R2 values were not greater than 0.9, the combination of R2 values and model performance
in terms of error and prediction showcase the validity of the model.

Sensitivity Analysis
The model’s validity justifies its use to conduct a sensitivity analysis on the input parameters and the
variables that influence these parameters. To conduct sensitivity analysis, a base case must first be defined.
Mean values of each parameter to be studied are taken as inputs to the base case. A summary of these values
can be found in Table 2.

Table 2—Base Case Input Parameter Values for 17.5 in. and 12.25 in. Sections

All parameters in Table 2 are changed except of DLS, because DLS is fixed in order to achieve the target.
In order to vary total RPM, the motor RPM is changed by changing revs/gallon and surface RPM is kept
constant due to motor limitations on surface RPM. HSI and JIF values are varried by changing TFA in order
to keep flow rate constant. Figs. 6 and 7 showcase the results of the sensitivity analysis, plotting the percent
change in a parameter and its effect on ROP. In the 17.5 in. section (Fig. 6), ROP was most sensitive to JIF,
SPE-197157-MS 9

HSI, and flow rate showing a positive change with increasing value of these paramters. Total RPM seems to
decrease ROP in the 17.5 in. section and WOB has no effect. Decrease in ROP with increasing Total RPM
can be attributed to increasing mechanical specific energy (Alali 2012) especially because 17.5 in. section
motor bend is 1.5 degrees to achieve required build rate at KOP.

Figure 6—Sensitivity Analysis Plots for 17.5 in. Section

In the 12.25 in. section (Fig. 7), JIF, HSI and mud weight had the highest impact on ROP. The positive
impact of the higher mud weight might be due to the stabilizing effect that mudweight has in the shale
formations (L, M, and N), ROP reduced with increasing JIF and HSI values, and increased with increasing
mud weight values.
10 SPE-197157-MS

Figure 7—Sensitivity Analysis Plots for 12.25 in. Section

Conventionally, ROP should increase with increasing JIF and HSI values. Such a negative impact on
ROP from high HSI and JIF values requires further investigation. Further analysis is done by varying flow
rate and TFA, increasing TFA to accommodate a higher flow rate (Table 3).

Table 3—Change in ROP, HSI, and JIF with Various combinations of flow rate and TFA

It becomes apparent that it is not high values of HSI or JIF that are contributing to lower RPM, but the
pump pressure limitations on increasing flow rate. Equations 8 through 11 show that increasing TFA has
a greater effect on JIF and HSI than an increase in flow rate. This causes the ANN model to relate higher
HSI and JIF values to lower ROP. This brings up an important flaw in the application of machine learning:
The results, while statistically correct, cannot be taken at face value and must be scrutinized against well
established concepts.
In both 17.5 in. and 12.25 in. sections it would seem that hydraulics is the biggest contributing factor
to ROP.

Optimization
Once the main contributing inputs are identified, their optimum values must be specified within the operating
limits of downhole and surface equipment facilities. The summary of parameters and resultant average ROP
SPE-197157-MS 11

values for the field, modelled optimum, and actual wells tested are presented in Table 4 for the 17.5 in.
section and Table 5 for the 12.25 in. section.

Table 4—Optimization results for 17.5 in. Section

Table 5—Optimization Results for 12.25 in. Section

Analyzing the real time data from three wells (X6-2, X3-2, and X3-3), it is possible to adjust the
parameters as the well is drilled subject to operational limits. These limits include pressure losses associated
with high flow rate, small TFA and mud weight, achieving required angle adjustment from motor bend, and
downhole motor pressure loss limitations.
In the 17.5 in. section (Table 4), the well with the smallest motor bend (X3-2) and closest flow rate to the
optimized flow rate achieved the highest average ROP improvement (14.05%). Well X6-2 achieved a lower
improvement of 10.18% and was drilled with a higher motor bend angle and lower flow rate. 14% and 10%
might not be high enough for statistical significance, however, motor bend and flow rate are identified in
the sensitivity analysis as significant parameters affecting ROP. This validates the model further, because
the actual behaviour of ROP did not vary significantly from the modelled ROP behaviour. Flow rate in this
section could not have been pushed further due to pump limits. However, this can be offset by reducing
the TFA size in future wells.
In the 12.25 in. section (Table 5), the highest average ROP improvement of 52% occurs in well X3-2.
Well X6-3 maintained similar parameters to X6-2, however, difficulty maintaining angle caused an increase
in motor bend which resulted in a decrease in ROP. Well X3-2 drilled with lower mud weight, smaller TFA,
and higher bend angle than recommended optimum, however, got closest to the recommended flow rate.
This resulted in well X3-2 achieveng the lowest improvement in average ROP (25%).

Conclusions
The machine learning model created and implemented in this study resulted in an average % error in ROP
between 24% and 70% for over twenty-four thousand data points. This error seems to linearly depend on
on the range of ROP in a well. Overall, the model performed better than other mathematical models in the
literature over a broader testing range.
12 SPE-197157-MS

The sensitivity analysis that used the model to vary input parameters and measure their effect on ROP
identified key parameters that, when optimized can increase ROP. The optimization of these parameters
resulted in a max improvement of average ROP of 52% in the 12.25 in. section and 14% in the 17.5 in.
section. An increase of 50% and 14% in ROP reduced drilling time and all time-dependent drilling costs
by 30% and 12% respectively.
The aim of this study was to experiment with and provide a blueprint for the application of machine
learning in real time drilling operations. The accuracy and subsequent success in using the model to
optimize drilling inspires confidence in applying more machine learning techniques to different aspects of
the industry. However, care must be taken when compiling, arranging, and modelling data. Because the
machine does not know about drilling mechanics, it can “learn” incorrectly by forming connections between
parameters that are not there.
Machine learning proved to be much more efficient than mathematical models in the case of ROP
modelling due to the speed of analysis of large volumes of data over a broad range.

Acknowledgments
The Authors would like to thank the Wells Department in PETRONAS Carigali Iraq BV. for their support
through data gathering and field trials.
The Authors would also like to thank PETRONAS for providing resources to conduct this study.

References
Alali, A., Akubue, V. A., Barton, S. P. et al 2012. Agitation Tools Enables Significant Reduction in Mechanical Specific
Energy. Paper presented at the SPE Asia Pacific Oil and Gas Conference and Exhibition, Perth, Australia, 22-24
October. SPE-158240-MS. https://doi.org/10.2118/158240-MS.
Bataee, M., Kamyab, M., and Ashena, R. 2010. Investigation of Various ROP Models and Optimization of Drilling
Parameters for PDC and Roller-Cone Bits in Shadegan Oil Field. Paper presented at the International Oil and Gas
Conference and Exhibition in China. Beijing, China, 8-10 June. SPE-130932-MS. https://doi.org/10.2118/130932-MS.
Beck, Marcus W. 2018. NeuralNetTools: Visualization and Analysis Tools for Neural Networks. Journal of Statistical
Software 85 (11). https://doi.org/10.18637/jss.v085.i11.
Bilgesu, H. I., Al-Rashidi, A. F., Aminian, K. et al 2001. An Unconventional Approach for Drill-Bit Selection.
Paper presented at the SPE Middle East Oil Show, Manama, Bahrain, 17-20 March. SPE-68089-MS. https://
doi.org/10.2118/68089-MS.
Curry, D. A., Perry, P. B., and Evans, J. M. 2004. New Method of Representing Rock Properties Over Entire Bit Run
Improves Computer Generated Bit Recommendations. Paper presented at the IADC/SPE Drilling Conference. Dallas,
Texas, USA, 2-4 March. SPE-87100-MS. https://doi.org/10.2118/87100-MS.
Gidh, Y. K., Ibrahim, H., and Purwanto, A. 2011. “Real-Time Drilling Parameter Optimization System Increases ROP
by Predicting/Managing Bit Wear.” Paper presented at the SPE Digital Energy Conference and Exhibition. The
Woodlands, Texas, USA, 19-21 April. SPE-142880-MS. https://doi.org/10.2118/142880-MS.
Hegde, C., and Gray, K. E. 2018. Evaluation of Coupled Machine Learning Models for Drilling Optimization. Journal of
Natural Gas Science and Engineering 56 (August): 397–407. https://doi.org/10.1016/j.jngse.2018.06.006.
Jamshidi, E., and Mostafavi, H.. 2013. Soft Computation Application to Optimize Drilling Bit Selection Utilizing Virtual
Inteligence and Genetic Algorithms. Paper presented at the International Petroleum Technology Conference. Beijing,
China, 26-28 March. https://doi.org/10.2523/IPTC-16446-MS.
Jones, A. G. 2018. Beyond Chi-Squared: Additional Measures of the Closeness of a Model to Data. Paper presented at the
SEG International Exposition and Annual Meeting, Anaheim, California, USA, 14-19 October: Society of Exploration
Geophysicists. https://doi.org/10.1190/segam2018-2998227.1.
Liu, Z., Marland, C., Li, D. et al 2014. An Analytical Model Coupled With Data Analytics to Estimate PDC Bit Wear.
Paper presented at the SPE Latin America and Caribbean Petroleum Engineering Conference. Maracaibo, Venezuela,
21-23 May. SPE-169451-MS. https://doi.org/10.2118/169451-MS.
Motahhari, H. R., Hareland, G., and James, J. A. 2010. Improved Drilling Efficiency Technique Using Integrated PDM
and PDC Bit Parameters. Journal of Canadian Petroleum Technology 49 (10): 45–52. SPE-141651-PA. https://
doi.org/10.2118/141651-PA.
R Core Team. 2013. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for
Statistical Computing. http://www.R-project.org/.
SPE-197157-MS 13

Valisevich, A., Ruzhnikov, A., Bebeshko, I. et al 2015. Drillbit Optimization System: Real-Time Approach to Enhance
Rate of Penetration and Bit Wear Monitoring. Paper presented at the SPE Russian Petroleum Technology Conference.
Moscow, Russia, 26-28 October. SPE-176517-MS. https://doi.org/10.2118/176517-MS.
Winters, W. J., and Warren, T. M. 1986. Field Application of Diamond-bit Hydraulic-Lift Principles. SPE Drilling
Engineering 1 (4): 277–287. SPE-11950-PA. https://doi.org/10.2118/11950-PA
Wright, J., Chukwu, G. A., Khataniar, S. et al 2003. An Economic Appraisal of Hole Cleaning Using Hydraulic
Horsepower and Jet Impact Force. Paper presented at the SPE Western Regional/AAPG Pacific Section Joint Meeting.
Long Beach, California, USA, 19-24 May. SPE-83496-MS. https://doi.org/10.2118/83496-MS.
14 SPE-197157-MS

Appendix A
Table A-1—Field Formation lithology and Depth

Figure A-1—Field UCS vs TVD