Nathan ZeneroTeradata
AI IS FULL OF PROMISE
BUT CAN IT DELIVER?
DATA SCIENCE RISK
North Stars Data Quality Walled Gardens Gaps Partners
Agenda
North Stars
©2018 Teradata North Stars
1
SAFE SMOOTH PRECISE EFFICIENT
No one gets
hurt.
Know where
trouble is likely
Deliver wellbores
with
Lower LOE
Cheaper M&R
Hit the geologic
target
…and nothing else
Mitigate
NPT, ILT
Material Waste
2 3 4
Operations Goals (for drilling)
DIGITAL NORTH STAR
Analytics Platform
Data Standards
Interoperability Standards
Sensor Ingest
Micro Service Platform
Integrated Data
Digital Twin
Automated Geosteering
DATA QUALITY
Understand Risk
ValueAnd Cost
Treat DataAs a
Manufactured Good
IADC/SPE 178776-MS • Iron Roughneck Make Up Torque—Its Not What You Think!• Nathan Zenero
-10%
-25%
+25%
+10%
5000
10000
15000
20000
25000
30000
35000
40000
5000 10000 15000 20000 25000 30000 35000 40000
Iro
n R
ou
ghn
eck
Re
po
rte
d T
orq
ue
[ft-
lbs]
IRTT Measured Torque [ft-lbs]
UNDER TORQUED REGION
OVER-TORQUED REGION
When Sensors Lie (or are Missing)
Sensor Attributes
• Range
• Accuracy
• Precision and Sigma Level (Repeatability)
• Sensitivity
• Resolution
• Linearity
• Hysteresis
• Reliability
System Attributes
• Calibration/Validation
• Sampling/Conversion
• Smoothing/Filtering
• Transparency/Manipulation
• Timeliness
• Fidelity
Data Properties
Manufacturing
Instrument Traceability (Calibration/Validation, Verification)
Controls/Data Acquisition (RAW->Measurement) Aggregation, Transmission, Storage Reporting, Operations, Analytics
Engineering Design (Plant, Well, etc.)
Device SpecsOperating LimitsFactory Calibration
Process SpecsInstallation/CommissioningStandards and Practices
SamplingFilteringSmoothing
ScalingNoise ReductionCalculations
CompressionUnit ConversionCalculations
Error Correction/ModelingEnrichment/AugmentationCalculations
Analog to DigitalSpecs
Device
Traceability
Analog Pre-Processing
DSP
Scaling
Intrinsic
Properties
TransformationTime Sync
Error Correction
Co
nte
xt
(or
Err
or) Finance
ERP
Engineering
Master
Beyond Heuristics
0
20
40
60
80
100
120Opportunity Cost ($USD Per Well)
Iterations 1,000
Mean $44,578.64
St Dev $6,932.23
P(10) $35,694.63
Rigs 10
Wells/Year (total) 100
Cost/Rig/Day $199
ROI 391%
Total Savings $2,843,113
Value of Information (www.ogdq.org)
Walled
Gardens
WITSML
API
Vendor
WITSML
Stores
Semantic Layer + Data
Services
Well
MasterERP
Geology
Geoscience
Wellbore
Position
Well
Planning
Sensor
Data
No Walls
16
Job Setup
(MDM+RDM)
Teradata Database Connections
HTTPS
HTTPS
HTTPS
Vendor A
Vendor B
Vendor C
Normalized WITSML
Channel Map, User Permissions
Pu
blic C
lou
d (
AW
S o
r A
zu
re)
Any Application
Any User
Any Partner
Vie
ws
An
aly
tics
VantageData
StoreMachine Learning
Engine
GraphEngine
Hig
h S
pe
ed
Fa
bric
SQLEngine
Open Data Foundations
17
• Sensor Agnostic
• Any sensor, in any location, with any amount of redundancy
• Ordinality is controlled by SMEs, abstracted from data science
• Vendor Agnostic
• Can switch data vendors on-the-fly with no interruption to service or additional complexity
• Business Process Agnostic
• Every rig, business unit, or customer can have their own mnemonics and taxonomy and still be able to talk in a common language
• Time Agnostic
• Data can different clocks with error, drift, and dilation
• All clocks can be tracked, and corrected
Data Foundation Considerations
18
TIMEHOLE
DEPTH
BIT
DEPTH
BLOCK
HEIGHT
DELTA
P
HOOK
LOAD
PUMP
RATE
PUMP
PRESSURERPM TORQUE WOB
SLIP
STATE
RIG
STATE
STATE
CODE
MACRO
STATE
BIT
STATE
01/24/2017 4:30:03 PM 1348.62 1347.04 1.71 -1117 5.296 0 39 42 4.15 14897 IN SLIPSDRILLING_GENERI
C01120110110 0
NEAR
BOTTOM
01/24/2017 4:30:04 PM 1348.62 1347.04 1.97 -1121 5.179 0 33 46 4.19 15015 IN SLIPSDRILLING_GENERI
C01120110110 0
NEAR
BOTTOM
01/24/2017 4:30:05 PM 1348.62 1347.04 2.37 -1121 5.179 0 33 50 3.065 15015 IN SLIPSDRILLING_GENERI
C01120110110 0
NEAR
BOTTOM
01/24/2017 4:30:06 PM 1348.62 1347.04 2.38 -1123 5.649 0 30 51 2.591 14545 IN SLIPSDRILLING_REAMIN
G01120110111 0
NEAR
BOTTOM
01/24/2017 4:30:07 PM 1348.62 1347.04 2.39 -1123 5.883 0 30 51 2.44 14310 IN SLIPSDRILLING_REAMIN
G01120110111 0
NEAR
BOTTOM
01/24/2017 4:30:08 PM 1348.62 1347.04 2.39 -1123 5.883 0 30 2 2.398 14310 IN SLIPS DRILLING_IN SLIPS 01110010111 0NEAR
BOTTOM
01/24/2017 4:30:09 PM 1348.62 1347.04 2.39 -1123 5.883 0 31 0 2.391 14310 IN SLIPS DRILLING_IN SLIPS 01110010111 0NEAR
BOTTOM
01/24/2017 4:30:10 PM 1348.62 1347.04 2.59 -1123 5.883 0 31 0 2.406 14310 IN SLIPS DRILLING_IN SLIPS 01120010111 0NEAR
BOTTOM
01/24/2017 4:30:11 PM 1348.62 1347.04 3.57 -1123 5.649 0 31 0 2.426 14545 IN SLIPS DRILLING_IN SLIPS 01120010111 0NEAR
BOTTOM
01/24/2017 4:30:12 PM 1348.62 1347.04 5.77 -1123 5.414 0 30 0 2.446 14780 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:13 PM 1348.62 1347.04 9.03 -1123 5.296 0 30 0 2.47 14897 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:14 PM 1348.62 1347.04 13.14 -1124 5.296 0 31 0 2.482 14897 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:15 PM 1348.62 1347.04 22 -1124 5.062 0 31 0 2.478 15132 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:16 PM 1348.62 1347.04 26.42 -1124 4.959 0 30 0 2.476 15235 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:17 PM 1348.62 1347.04 31.12 -1123 5.179 0 30 0 2.475 15015 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:18 PM 1348.62 1347.04 35.93 -1123 5.296 0 30 0 2.474 14897 IN SLIPSDRILLING_GENERI
C01121010110 0
NEAR
BOTTOM
01/24/2017 4:30:19 PM 1348.62 1347.04 40.75 -1124 5.179 0 30 0 2.472 15015 IN SLIPS INVALID 01122010110 0NEAR
BOTTOM
No Black
Boxes
WOB: 14897
Slip State: IN SLIPS
Rig State: ROTARY DRILLING
State Code: 01120110110
Marco State: 0
Bit State: ON BOTTOM
Hold Depth: 13148.87
Bit Depth: 13146.27
Block Height: 2.3
Hookload: 5.98
Pump Rate: 114
Pump Pressure: 39
Block Height: 2.37
Delta P: 1117
1/23/2019 4:30:03 pm
Interoperability Requires Modularity and TrustTrust <> unquestioned acceptance; Trust = seamless data flow of known quality
Dis
trib
ute
d C
on
tro
ls
Ab
stra
ctio
n L
aye
rIIo
TD
ata
La
ye
r (w
ith
se
nso
r d
ata
-qu
alit
y m
od
el
ba
sed
on
ISO
Sta
nd
ard
)
ERP
AccountingEngineering / Operations
IIoT Sensor Data
En
terp
rise
Da
ta S
erv
ice
s
(eff
ort
less
in
form
atio
n)
Digital Twin RTM
Virtual Controls
Automation
Data Sources
Sensors
Distributed Controls
Digital Transformation
Open Group
The Kaggle Conundrum
© 2017 Teradata
Limit Feature Extraction
QUANTITY of data
Controlled image from BitBox
Advanced Feature Extraction
QUALITY data
Filling the Gap
© 2017 Teradata
12/16/2019
Inverting Kaggle
Democratizing Data
©2018 Teradata
AI Suffers without Complete Data
Digital Partners
Can we automatically extract information from technical text fields?
Wellmaster (i.e. Wellview) text logs are verbose, technical and are tied to categorical fields (such as Task and Activity)
Operational Comments, contain multiple sentences, jargon, units, and
One comment may apply to a series of subsequent time-lo entries
Currently, logs are expanded through manual processing
Using Teradata Drilling SME and Data Science tools and skills in Natural Language Processing
Use (NLP) and machine learning techniques to process data and identify records that are a loss event
Algorithmically group loss events to automatically create the Loss Event #
Machine learning to extract Remediation Options Used
Associated Remediation details extracted from log entries with NLP
Automation, Accuracy, Repeatability
Automatic post-event identification is very possible – 93% accuracy achieved
Perfectly reproduced loss events by count and ordinal
Extracted remediation options and numeric details – LCM Volume reproduced with 95% precision
Performance improvements with more data – 10x data volumes; continuous time period (i.e. 2 years); larger geographic area; additional sensor datasets
Loss Event Identification
CHALLENGE SOLUTION BENEFITS
planning
collaboration
automation
operations
Partners Should Understand Eachother
Tota
l In
vest
me
nt
0% Completeness of Solution 100%
What we say we want
(20% POC)
What we are willing to use
(80% MVP)Upscaling is rarely a linear projection of POC investment.
SCALE IS NOT SIMPLE
CO
MM
MER
CIA
LISA
TIO
N G
AP
• Automated Visual Inspection/Forensics • Drillbits (partnering with drillbit manufacturers)• Tubulars (partnering with tubular manufacturers)• BHA components
• Streaming Ingest of All Operations Data• Completion (partnering with sensor package OEMs)• Workover Data (co-developing low-cost workover
sensor kits)
• Industry Leaders• SPE DSATS, OGDQ• OSDU Contributors and Committee Chairs• Energistics Members and Contributors