Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020
Data Driven Drilling and Rate of Penetration Optimization
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020
Background
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 3
Drilling
Pump flow rate (GPM)
Rotary speed (RPM)
Drilling Rig
Weight on bit (WOB)
Drilling bit
Rate of Penetration(ROP)
Drilling Fluids
Formation 2
Formation 1
Sect
ion
vPump flow rate (GPM)
Rate of Penetration (ROP)
vRotary Speed (RPM)
vWeight on bit (WOB)
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 4
ROP OptimizationThe Rate of Penetration (ROP):
𝜔𝑖 : a weight given for each drilling parameter.
The optimum ROP is defined as the fastest ROP for a particular well, drilling through a particular formation, while minimizing drilling Non-Productive Time (NPT).
)𝑹𝑶𝑷 = 𝒇(𝝎𝟏𝑾𝑶𝑩,𝝎𝟐𝑹𝑷𝑴,𝝎𝟑𝑮𝑷𝑴 .
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 5
Motivation
No single system satisfactory predicts/optimize ROP globally
v Physics Approaches [Bingham, Maurer, Teale]: Does not incorporate static and dynamic parameters. Requires data we do not have, such as rock Uniaxial Compressive Strength (UCS).
v Data Based Approaches [Moran, Alkatatny, Ahmed] : Previous work is done for one specific field/formation.
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 6
Objective
Need for a robust ROP optimization in real-time to recommend the controllable dynamic drilling parameters
(WOB, RPM, GPM).
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020
The Proposed Approach
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 8
The Proposed ApproachAn integrated Two-phase data-driven rate of penetration optimization system.
Phase One [Historical]: Geologically Driven and Historical Data Based
Phase Two [Real Time]: Automatic Drill-Off Test
Data Selection Cleansing and Processing
Geological Calibration
Inverse Distance Weight (IDW).
Phase One (input) Constrained Random Search (CRS) Update [WOB, RPM]
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 9
The Proposed Approach
At 4,000 ft
Well with a drilling problem.
Wells contributing to the model in this section, no NPT.
Wells contributing to this depth step.
wi A weight for this well.
Well 1:WOB = 50 klb.𝜔1 = 1.6
Well 3:WOB = 48 klb.𝜔3 = 0.8
𝝎𝟑
𝝎𝟐
𝝎𝟏
𝝎𝟒 OutputWOB = 49.7 klb.
Well 2:WOB = 52 klb.𝜔2 = 1.2
Well 4:WOB = 45 klb.𝜔4 = 0.4
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020
Validation
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 11
Phase One ModelData Health Index
(Uncertainty)
Formation Tops
Optimum Value
GPMRPM WOB GPM
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 12
Phase Two Update
A visual of the automated drill-off test window, that illustrate the different pairs of WOB and RPM with the
resulted ROP
Jump/adjustment of phase one recommendation as we entered a new geological formation.
ROP ft/hr
WOB
RPM
Heat Map ROP
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 13
Results
WOB GPMRPMROP
Trial DataPhase one and Phase two
Optimal
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 14
Model Generated
𝑅𝑂𝑃! = &"#$
%
𝑊&" 2/(1 + 𝑒'& ('(,'()*+('(,*,-.+('(,+/-.+0'( ) − 1) + 𝑏&
A new relationship was derived from the ANN model using the biases and weights of neurons-connections among the input, hidden and output layers. The new ROP correlation is shown in the following Equation:
fitnetWOBRPMGPM
W
b+
tansig W
b+
tansig
ROP
Output
Inpu
t
17-neuron
Hidden 1-Layer Output Layer
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 15
Act
ual R
OP
vs
Pred
icte
d R
OP
RMSE=9 RMSE=23 RMSE=44 RMSE=69
Comparison with Other ModelsC
orre
latio
n C
oeffi
cien
t
R= 0.41 R= -0.2R= 0.94 R= 0.76
ROP (ft/hr) ROP (ft/hr) ROP (ft/hr) ROP (ft/hr)
ActualBingham
ActualANN
ActualMaurer
ActualAhmed
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 16
25% Higher ROP Achieved, compared with the top 10 offset wells.
Rat
e of
Pen
etra
tion
(ft/h
r)
Depth (ft)
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 17
Thank You
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 1818
Longitude
Latit
ude
WO
B
50.0
New Well Location
Depth (ft): 5551 ft
WOB Kriging: 50 klb
WOB IDW: 58 klb
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 1919
WOB at Depth (ft): 5551 ftWOB IDW: 58 klb
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 2020
WOB at Depth (ft): 5551 ftWOB Max Entropy: 47.9 klb
WOB IDW: 58 klb
Den
sity
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 2121
IDW
∑!"#$"%𝑤!𝑛
= 1
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 2222
WOB at Depth (ft): 5551 ft
54.8
Normal distributionSigma=3
Normal Kernel (the normal kernel is often used, which means K(x) = ϕ(x), where ϕ is the standard normal density function.)
Smoothing=4
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 2323
v ANN Method Development
Data distribution 65 / 35 70 / 30 75 / 25 80 / 20 85 / 15
# of Neurons 20 - 0 20 - 10 10 - 10 10 - 20 15 - 10 15 - 10 17 - 0 18 - 0
Training functions trainlm trainbr trainbfg traingd traingdm trainb trainoss trains
Transfer functions tansig hardlims poslin radbas compet purelin softmax tirbas netinv
Network functions fitnet feedforwardnet cascadeforwardnet
fitnetWOBRPMGPM
W
b+
tansig W
b+
tansig
ROP
Output
Inpu
t
17
Hidden Layer Output Layer
MIT EARTH RESOURCES LABORATORY ANNUAL FOUNDING MEMBERS MEETING 2020 24
• HIGHT ROP AT WHICH CUTTINGS ARE BEING GENERATED TOO FAST TO BE CLEANED FROM THE ANNULUS;
• HIGH WOB THAT WILL GENERATE EXCESSIVE TORQUE FOR THE TOP DRIVE;
• HIGH WOB THAT WILL GENERATE EXCESSIVE TORQUE FOR THE DRILL PIPE; AND
• HIGH RPM THAT CAUSES EXCESSIVE VIBRATION OF THE DERRICK.
• HIGHT GPM THAT WILL INDUCE LOSSES.
24
https://www.iadc.org/
: