IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

Bit selection using field drilling data and mathematical investigation

To cite this article: M S Momeni et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 328 012008

View the article online for updates and enhancements.

This content was downloaded from IP address 181.214.187.91 on 16/03/2018 at 00:36

ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

Bit selection using field drilling data and mathematical

investigation

M S Momeni1, S Ridha1, S J Hosseini1,*, B Meyghani2 and S S Emamian2

1
Department of Petroleum Engineering, Universiti Teknologi PETRONAS, 32610 Bandar Seri
Iskandar, Perak Darul Ridzuan, Malaysia
2
Department of Mechanical Engineering, Universiti Teknologi PETRONAS, 32610 Bandar Seri
Iskandar, Perak Darul Ridzuan, Malaysia

*Corresponding Author: [email protected]

Abstract. A drilling process will not be complete without the usage of a drill bit. Therefore, bit
selection is considered to be an important task in drilling optimization process. To select a bit is
considered as an important issue in planning and designing a well. This is simply because the
cost of drilling bit in total cost is quite high. Thus, to perform this task, aback propagation ANN
Model is developed. This is done by training the model using several wells and it is done by the
usage of drilling bit records from offset wells. In this project, two models are developed by the
usage of the ANN. One is to find predicted IADC bit code and one is to find Predicted ROP.
Stage 1 was to find the IADC bit code by using all the given filed data. The output is the Targeted
IADC bit code. Stage 2 was to find the Predicted ROP values using the gained IADC bit code in
Stage 1. Next is Stage 3 where the Predicted ROP value is used back again in the data set to gain
Predicted IADC bit code value. The output is the Predicted IADC bit code. Thus, at the end,
there are two models that give the Predicted ROP values and Predicted IADC bit code values.

Keywords--Artificial Neural Network (ANN), International Association of Drilling Contractors

(IADC), Rate of Penetration (ROP).

1. Introduction

1.1. Background

The basic form of bit selection is normally done based on cost per foot. This method is simply choosing
the bit that will provide the lowest cost per foot over the upcoming interval. In addition to that, other
factors are taken into consideration as well such as offset, journal angle, and other design aspects. This
differentiates one bit to another according to the specific environments. Therefore, understanding bit
types is a vital step before moving on to bit design as well as bit selection.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

1.2. Bit Types

Rotary drilling bits can be generally classified as either drag bit or rolling cutter bits according to the
design features. Drag bits have fixed cutter blades in common. These blades are integrated within the
body of the bit. The rotation takes place with the drill string as one unit. On the other hand, the rolling
cutter bits normally have 2 or more cones which have basic cutting elements. These cutters rotate about
the axis of the cone during the bottomhole rotation [1]. There are two main types of bits which are Drag
Bit (PDC Bit) and Rolling Cutter Bit (Tricone Bit).

1.3 Bit Design

1.3.1 Bit Design for Rolling Cutter Rock Bits

Basically, there are 3 basic types of bearings:

 The Non-Sealed Roller Bearings (NSRB)
 The Sealed and Lubricated Roller Bearings (SLRB)
 The Sealed and Lubricated Journal Bearings (SLJB)

1.3.2 Bit Design for PDC Bits

The PDC bit is known to be a solid one piece tool holding polycrystalline diamond cutters. The synthetic
diamonds are molded into a thin layer to be attached to a tungsten carbide disc. This is done via a HPHT
process. The shock load propagates through entire cutting via the random orientation of the cleavage
planes that indirectly allows reduction in breakage. PDC bits are known to shear the rock which helps
save energy. Thus, optimized drilling can be gained with the usage of less WOB. Given a favorable
formation, PDC bits are acknowledged to perform longer and harder. The effectiveness is about 3 times
than conventional rolling cutter bit [2]. Having all these plus points, PDC bits are quite expansive and
can be destroyed by gumbo type formations. Therefore, a proper geologic analysis together with PDC
bit compatibility must be done before drilling. The detailed field analysis for PDC bit has not been
completed and this report is based on the data limitations [3].

1.4 Bit Selection Methods

A bit cost might be relatively small in a well’s budget which is approximately 5% of the total expense
of a well, but the impact of bit performance on overall well cost might end up being considerably large
[4-6]. Bit selection is basically classified into three categories namely:

• Cost Analysis
• Offset Well Log Analysis
• Bit Performance Modeling Using ANN and Genetic Algorithm

2. Methodology

ANN is a computational technique used to solve complex problems. MLP network is one of the most
popular neural network architectures for modeling process [7]. It consists of input layer of source nodes,
hidden layer of computation nodes (neurons), and output layer. The number of nodes that is being used
in the input as well as the output layer is completely dependent on the number of input and output
variables being used respectively [8]. The figure shown is a schematic of MLP network. Further
theoretical details about MLP networks is presented by [9]. In this project, the MLP network is trained
by the usage of LM technique. One of the advantages of using the LM technique is that it produces
efficient and faster second order convergence rate and is capable to maintain the stability at the same

2
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

time [10, 11]. Based on the research papers, it is clearly stated that the single layer is good enough to
approximate a complex function [12].Therefore, in this study, a three layered Feed-Forward Network is
developed namely input, hidden, and output layers. Among the available data sets which are gathered
from different offset wells, 70% is used for training, 15% is applied for validation process and the
remaining 15% is used to test gained results from the bit being modeled as well as the ROP functions.
Stage 1 was to find the IADC bit code by using all the given filed data. This data includeSize(in), Flow
area (in2), Depth out (m MD), Bit meter (m), Rotation hours (hrs), ROP(m/hr), Minimum Rotation
(rpm), Maximum Rotation (rpm), Total bit revolution, Minimum Weight (kN), Maximum Weight (kN),
Minimum Flow (l/min), Maximum Flow (l/min), Minimum Pump Pressure (bar), and Maximum Pump
Pressure (bar). The output is the Targeted IADC bit code.

Table 1. Method in Obtaining ANN Results

Model Data Being Description
Used
ANN Drilling Data 10= 512X
20=434X
30=532M
Output Example Ex: ANN predicts
IADC Bit 10,20,30 the numbers as
Code perhaps 11.5 (10),
23.4 (20), and 38.7
(30)
The results of the three methods discussed above are considered for ROP prediction and improve
predicted IADC bit code value.

3. Results and Discussion

For the ANN simulation process, the Matlab software was used. The Neural Network tool which was
prebuild in this Matlab software was used to gain all the results and generate all the graphs needed. The
type of data used for this project comprises of drilling data. The table below shows an example of data
set that was used. The following table represents the range of the data which was used.

Table 2. Data Type Example

Type MITO Rotation min(rpm) 60
Size(in) 17,50 Rotation max 60
(rpm)
IADC Code 115M Total bit 86400
revolution
Input For ANN 1 Weight min (kN) 20
Flow area (in2) 0.994 Weight max (kN) 60
Depth out 68 Flow min (l/min) 2500
(m MD)
Bit meter (m) 55 Flow max (l/min) 2500
Rot hours (hrs) 24,00 Pump min/max 20
(bar)
ROP(m/hr) 2,3 Pump max (bar) 22

Table 3. Data Type Range

Type -------- Rotation 0-271
------- min (rpm)

3
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

Size(in) 4.13- Rotation 0-335

36.00 max
(rpm)
IADC -------- Total bit 0-
Code ------- revolution 1829000

Input For 1-150 Weight 0-240

ANN min (kN)
Flow area 0.000- Weight 0-400
(in2) 2.227 max (kN)
Depth out 52- Flow min 0-4000
(m MD) 4762 (l/min)
Bit meter 0-569 Flow max 0-4100
(m) (l/min)
Rot hours 0.44- Pump min 0-242
(hrs) 190.78 (bar)
ROP(m/hr) 0.00- Pump 0-297
50.00 max (bar)

The following was done to gain the results that will be discussed below. The Azar Well data was used
in the ANN modeling process. The modeling process was done in three stages. Stage 1 was to find the
IADC bit code by using all the given filed data. This data includes Size(in), Flow area (in2), Depth out
(m MD), Bit meter (m), Rotation hours (hrs), ROP(m/hr), Minimum Rotation (rpm), Maximum Rotation
(rpm), Total bit revolution, Minimum Weight (kN), Maximum Weight (kN), Minimum Flow (l/min),
Maximum Flow (l/min), Minimum Pump Pressure (bar), and Maximum Pump Pressure (bar). The
output is the Targeted IADC bit code. Stage 2 was to find the Predicted ROP values using the gained
IADC bit code in Stage 1. This time, the data used as input in the ANN modeling process include
Targeted IADC bit code, Size(in), Flow area (in2), Depth out (m MD), Bit meter (m), Rotation hours
(hrs), Minimum Rotation (rpm), Maximum Rotation (rpm), Total bit revolution, Minimum Weight
(kN), Maximum Weight (kN), Minimum Flow (l/min), Maximum Flow (l/min), Minimum Pump
Pressure (bar), and Maximum Pump Pressure (bar). The output is the Optimized ROP. Next is Stage 3
where the Predicted ROP value is used back again in the data set to gain Predicted IADC bit code value.
The input parameters were Size(in), Flow area (in2), Depth out (m MD), Bit meter (m), Rotation hours
(hrs), Predicted ROP(m/hr), Minimum Rotation (rpm), Maximum Rotation (rpm), Total bit revolution,
Minimum Weight (kN), Maximum Weight (kN), Minimum Flow (l/min), Maximum Flow (l/min),
Minimum Pump Pressure (bar), and Maximum Pump Pressure (bar). The output is the Predicted IADC
bit code. The results were gained and they were used to plot regression graphs. The graphs are shown
below. In all the simulations done under the ANN modeling, the data was used with 70% for training
data, 15% for validation data, and 15% for test data.

For the simulation process, the number of hidden neurons can be varied from a range of 1 until 23. The
best simulation results were gained when the number of hidden neurons was set to 10. The following
table shows the Mean Square Error values and R values for Training Data, Validation Data, and Test
Data.

Table 4. MSE And R Value For 10 Number Of Hidden Neurons (Stage 1)

SAMPLES MSE R
TRAINING 71 498.46659 0.82799
VALIDATION 16 1287.67076 0.68085
TESTING 16 2095.87942 0.83949

4
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

Figure 1. Results Of The Regression Graphs For Output Of Optimized IADC Bit Code (Stage 2)

For the stage 3 simulation process, it is seen that the number of hidden neuron layers of 17 showed a
better response compared to number of hidden neuron layers of 10.

Table 5. Predicted IADC Bit Code Value And Predicted ROP Value Examples
IADC ANN Predicted ROP (ft/hr) Predicted
Bit Output IADC ROP
Code Bit Code Value
Value (ft/hr)
1,1,1 10 8.540809 19.263360 20.03738
1,1,1 10 10.49332 16.2142857 18.74017
1,3,4 70 74.78044 7.77586207 3.43065
2,1,4 90 93.24648 3.37209302 2.866403
2,1,4 90 85.00292 3.09090909 2.379206
1,3,4 70 65.81053 2.80991736 1.642444
1,3,5 80 83.09457 2.80991736 2.324059
2,1,4 90 90.54357 1.39534884 1.711341
2,1,4 90 70.67985 1.93478261 1.633869
2,1,4 90 71.96897 1.90697674 1.499116
2,1,4 90 71.90172 2.0000000 1.35653
2,1,4 90 72.16207 1.60655738 1.455876
2,1,4 90 70.22408 1.5106383 1.02429
2,1,4 90 88.04242 1.41176471 1.29425
2,1,4 90 83.33124 2.63157895 1.329831
1,1,1 10 6.796307 13.5396825 12.9523
1,1,1 10 8.319983 6.4000000 5.959551
1,1,1 10 6.990621 4.30769231 3.70865
1,1,1 10 9.344001 3.3800000 3.596977

These values can be concluded as a good approximation as the predicted values are almost near to target
values provided. The graph below shows the comparison of predicted IADC Bit Code versus the targeted
IADC Bit Code.

5
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

Figure 2. Results Of The Regression Graphs For Output Of Predicted IADC Bit Codes

Based on the second set of data set used to recheck the accuracy of ANN simulation, it can be said that
the ANN simulation process is quite accurate to be used as a prediction tool. Thus, a third set of data
was taken into consideration to confirm the accuracy of the ANN simulation process whereby the results
were predicted to be at least 95% accurate.

4. Conclusion

In conclusion, the objective of the project has been achieved. The ANN simulation modeling process
has been used to Predict ROP value and the Predicted IADC Bit Code has been found. Generally, the
training is done for about 10 to 15 times for each Number Of Hidden Neurons Layers to gain the best
output of Mean Square Error, MSE and Regression value, R. It has been seen during the simulation
process that the most suitable hidden neuron number is 10 and 17. For the Azar Well 1 and 2 data sets,
the final accuracy levels of the Predicted IADC Bit Code Values were 0.99 and the final accuracy level
of the Predicted ROP Values were 0.95. This clearly shows that the usage of ANN simulation as a
prediction tool can provide and accuracy level of more than 95 percent which is fairly accurate. In order
to reconfirm that these values were accurate enough, a new set of data was also used to see if it gives a
proper prediction. Once again, the final accuracy levels of the Predicted IADC Bit Code Values were
0.96 and the final accuracy level of the Predicted ROP Values were 0.95 for the second set of data set.
Therefore, it is clearly seen that this project has managed to utilize the ANN simulation to predict the
IADC Bit Code and ROP values quite effectively. However there are still some errors in the prediction.
This is simply due to certain reasons which are insufficient data, limited time frame to complete the
project, and also the absence of Log Data.

REFERENCES

1. Fasheloum M. Investigation of drilling parameters indicators: University of Nottingham; 1997.

2. Nygaard R, Hareland G, editors. How to select PDC bit for optimal drilling performance. Rocky
Mountain Oil & Gas Technology Symposium; 2007.
3. Oteri VA. Drilling Optimization- Drill Bit Performance Optimization Using DROPS Simulator.
THESIS. 2010 11-30.
4. Lummus J. Acquisition and analysis of data for optimized drilling. Journal of Petroleum
Technology. 1971;23(11):1285-93.
5. Yιlmaz S, Demircioglu C, Akin S. Application of artificial neural networks to optimum bit
selection. Computers & Geosciences. 2002;28(2):261-9.

6
ICMMPE 2017 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 328 (2018) 012008 doi:10.1088/1757-899X/328/1/012008
1234567890‘’“”

6. Mostafavi H, Jamshidi E, editors. Soft Computation Application to Optimize Drilling Bit

Selection Utilizing Virtual Inteligence and Genetic Algorithms. International Petroleum Technology
Conference; 2013: International Petroleum Technology Conference.
7. Feng S, Li L, Cen L, Huang J. Using MLP networks to design a production scheduling system.
Computers & Operations Research. 2003;30(6):821-32.
8. Pinar E, Paydas K, Seckin G, Akilli H, Sahin B, Cobaner M, et al. Artificial neural network
approaches for prediction of backwater through arched bridge constrictions. Advances in Engineering
Software. 2010;41(4):627-35.
9. Haykin S. Neural networks: a comprehensive foundation: Prentice Hall PTR; 1994.
10. Hagan MT, Menhaj MB. Training feedforward networks with the Marquardt algorithm. Neural
Networks, IEEE Transactions on. 1994;5(6):989-93.
11. Cigizoglu H, Kisi O. Flow prediction by three back propagation techniques using k-fold
partitioning of neural network training data. Nordic Hydrology. 2005;36:49-64.
12. Wilamowski BM, Iplikci S, Efe MO, editors. An algorithm for fast convergence in training
neural networks. Neural Networks, 2001 Proceedings IJCNN'01 International Joint Conference on;
2001: IEEE.