Application of Artificial Intelligence
in Drilling Data Analysis
Supervised by: Dr. Alireza Salmachi (ASP)
Brett Jenkin ( a1645479)
Haziq Ahmed (a1661882)
Wilson Wan (a1639897)
1
Contents
1. Introduction
2. Aims and Objectives
3. Literature Review
4. Methodology
5. Results
6. Discussion
7. Recommendations
8. Conclusion
9. References 2
Introduction
Existing Well Data Apply Intelligence Create New Value
3
Aims and Objectives
100
1 2 3 4 5 6
Literature Data Data ANN Training Validation of Identify parameter
Review Collection Processing & ANN Model relationships to
Screening ROP
4
Literature Review
Wellbore Drilling
Parameters
Artificial Neural Previous Methods
Research
Networks Area of Drilling Data
Capabilities Analysis
Fundamentals of
Artificial Neural
Networks
5
Wellbore Drilling Parameters
Wellbore Drilling
Parameters
Formation Data Mud Log Data Drill Bit Data
6
Wellbore Drilling Parameters
Wellbore Drilling
Parameters
Formation Data Mud Log Data Drill Bit Data
Lithology Formation Name
7
Wellbore Drilling Parameters
Wellbore Drilling
Parameters
Formation Data Mud Log Data Drill Bit Data
Lithology Formation Name Bit Size Bit Type Bit Total Flow Area
8
Wellbore Drilling Parameters
Wellbore Drilling
Parameters
Formation Data Mud Log Data Drill Bit Data
Lithology Formation Name Bit Size Bit Type Bit Total Flow Area
Rate Of
Drilling Depth Weight On Bit Rotary Speed Torque Pump Flow Rate
Penetration
9
Previous Drilling Analysis Methods
Hareland et al.
Bourgoyne & Young
1965 1987
1974 2010 Our
Project
Well
Real-time
monitoring
Monitoring &
Bingham system Warren
Optimization 2018-
improves
7
Limitations of Previous Methods
✘
● Based on research done in a different region with
Inconsistent Predictions different lithology
● “Capped” accuracy depending on lithology similarity
✘
● Accurate but impractical
● Unrepresentative sample
Developed using Lab Results ● Uses too many variables as a control
● Does not make use of field data
✘
● Ignores many variables affecting ROP
Oversimplified ● Oversimplifies relationship between drilling
parameters
8
Benefits of ANN Method
✔
● Model was developed with vast amount of field data.
Developed using Field Data ● Field data provides a better representation of the
Cooper Basin.
✔
● Takes into consideration many drilling parameters.
Complex Model ● In depth understanding of the relationship of these
parameters.
✔
● Uses history matched data and an understanding of
Optimum Relationships parameter relationships to provide better parameter
recommendations.
9
Fundamentals of ANN
• A biological neuron is a brain cell that sends
Dendrites electrical impulses to each other to form neural
networks.
• Soma collects incoming impulses and when the
voltage reaches a threshold, an impulse will be
generated along the axon.
• In the artificial sense, each synaptic connection
has Weight, w.
Axon
• The neuron collects incoming signals, i, and
Soma
applies weights to determine the Output, o.
• The neuron can recall knowledge or “experience”
gained from the previous problems.
• Building a system of neurons that can classify
data and predict an outcome.
10
Fundamentals of ANN
• A biological neuron is a brain cell that sends
electrical impulses to each other to form neural
networks.
𝑖4 𝑖5 • Soma collects incoming impulses and when the
𝑤4 𝑤5 voltage reaches a threshold, an impulse will be
𝑖3 𝑤3 generated along the axon.
𝑜 = 𝑓𝑎𝑐𝑡 (σ5𝑛=1 𝑖𝑛 . 𝑤𝑛 )
• In the artificial sense, each synaptic connection
has Weight, w.
𝑤2 𝑜
𝑖2 • The neuron collects incoming signals, i, and
𝑤1 applies weights to determine the Output, o.
• The neuron can recall knowledge or “experience”
𝑖1 gained from the previous problems.
• Building a system of neurons that can classify
data and predict an outcome.
10
ANN Capabilities
Improve Analysis of Support Financial Risk
Drilling Rock Matrix Management Forecast Prevention
Performance Strength Decision
11
Methodology
Data Collection Data Processing Analysis of Rock Correlation Coefficient Neural Network
Matrix Strength Analysis Algorithm 12
Data Collection
13
Data Processing
Screening
Clustering Identifying
Compiling Validating 14
Rock Matrix Strength
Bit weight exponent
Weight on Bit (klbs)
𝛼 Rock formation exponent
𝑊 Experimentally equals 1
𝑅 = 𝐾 𝑁𝛽
𝑑𝑏
Rotation per Minute (RPM)
Rate of Penetration (ft/hr)
Bit Diameter (in)
Rock Matrix Strength
(Bingham Equation, 1965) 15
Correlation Coefficient
𝑟 = Correlation Coefficient
𝑛 σ 𝑋𝑌 − σ 𝑋 σ 𝑌 𝑋 = Input Variable
𝑟=
2 𝑌 = Output Response
𝑛 σ 𝑋2 − σ 𝑋 2 [𝑛 σ 𝑌 − σ 𝑌 2 ൧ 𝑛 = Number of samples
Y Y Y
X X X
Strong Positive Correlation Strong Negative Correlation Weak Correlation
r = 0.9 r = -0.9 r = 0.1 16
Artificial Neural Network
Arrange Train Neural Analyse
Dataset Network Results
17
Results – Hutton Data Collection
• Substantial quantity of data over a significant range
• To better suit the pattern recognition, we reduced the range and identified wells with common
18
data
Results – Hutton Group 1
• Selected wells from within the Coonatie Group
• The intention was that by being located within the same region, the lithology would be more
consistent 19
Results – Hutton Group 2
• The selection was based upon including the data that was from the same depth region as the
Coonatie wells identified in Group 1 20
Results – Hutton Group Comparison
Coonatie Well Group Expanded Well Group
-4.75%
-3.11% -1.58%
-3.21%
21
Results – Combined Hutton Group
-2.91% Coonatie Well Group -1.30% Coonatie Well Group
+1.90% Expanded Well Group +1.90% Expanded Well Group
Poor data trends require further analysis and an alternate technique
22
Results – Neural Network Fit
ROP (ft/hr)
0 50 100 150 200 250 300 350
6600
6800
7000
Data used to
train NN
7200
7400
Depth (ft)
7600
Data used to
7800 validate NN
8000
Data used to
8200
test NN
8400
8600
23
Results – Adori Data Collection
• Substantial quantity of data over a significant range with the majority between 5900ft – 7400ft
• No significant grouping of wells such as the Coonatie group used in the Hutton analysis 24
Results – Adori Data Groups
6450ft – 6850ft Well Group 6850ft – 7330ft Well Group
• The selection was based upon identifying a depth range with a high concentration of data that
with reduced ROP ranges 25
Results – Adori Group Comparison
6450ft – 6850ft Well Group 6850ft – 7330ft Well Group
-0.0075%
-4.93% -4.48%
-0.003%
26
Results – Combined Adori Group
-10.28% 6450ft – 6850ft Well Group -6.77% 6450ft – 6850ft Well Group
-9.54% 6850ft – 7330ft Well Group -6.47% 6850ft – 7330ft Well Group
Outlier effect on relationship identification is increased when they are
present in higher quantity
27
Results – Rock Matrix Strength
28
Results – Parameter Comparison
Hutton Sandstone Parameter Correlation Adori Sandstone Parameter Correlation
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
-0.10 -0.10
-0.20 -0.20
-0.30 -0.30
-0.40 -0.40
-0.50 -0.50
Poor Bottomhole Cleaning Poor Mud Rheology 29
Results – ROP Errors
Adori Sandstone Poor Data Examples
500
450
400
350
ROP (ft/hr)
300
250
200
150
100
50
0
6200 6300 6400 6500 6600 6700 6800 6900 7000 7100 7200
Depth (ft)
• Grouping of ROP values along consistent rates
• Range of values to excessive or inconsistent to effectively be utilised by the Neural Network
30
Results – Torque Errors
Adori Sandstone Torque Comparison
14000
12000
10000
Torque (ft/hr)
8000
6000
4000
2000
0
5800 6000 6200 6400 6600 6800 7000 7200 7400 7600
Depth (ft)
• Torque metering faults
• Range of values too uniform or inconsistent to effectively be utilised by the Neural Network
31
Results – Consistent Trend data
300
250
200
ROP (ft/hr)
150
100
50
0
5800 6000 6200 6400 6600 6800 7000 7200
Depth (ft)
• Data values following easily identifiable trends
• Outliers are present on some data sets, however are easily identified from the well trend and
can be managed manually before running the Neural Network fit. 32
Results – Refined Adori Group
+14.5% Combined Adori Group Data +16.0% Combined Adori Group Data
Reduction in ROP value faults significantly improves correlation
33
Results – Node Optimisation and Comparison
+17.4% +17.7%
Increasing the Neural Network
nodes enables improved trend
matching
+2.91% +1.64%
34
Results – Adori Parameter Comparison
Adori Correlation Coefficients Adori Sandstone Correlation After Data Analysis
0.30 0.3
0.20 0.2
0.10 0.1
0.00 0
-0.10 -0.1
-0.20 -0.2
-0.30 -0.3
-0.40 -0.4
-0.50 -0.5
Mud rheology effect clear,
Low confidence in Correlations WOB and RPM inefficiencies also identified 35
Future Work
AB 12
C 3
Comparing Different Normalise Dataset Converting Non- Neuron Sensitivity
Models Numeric Values Analysis
36
Conclusion
Hutton Adori
Potential to Improve Bit Improve Mud Improve Data Access could
improve Drilling Cleaning Design Metering improve Technical
Feedback
37
