AI-Driven

Production Optimization Platform

AWS Canada Roadshow

Calgary, AB

June 2017
ambyint.com
Ambyint Introduction & Business Overview
Overview
Customer Pain Points
Solution
Value Proposition
Case Studies

Alex Robart
CEO
 Drilling and Completion So why are we using the same
Technology has Changed Math and Technologies to pump it?
 Production Operations Pain Points

#1: low quality data + not enough data

Manual data from 30-minute data
daily pumper visits OR via SCADA

#2: 20-year old control logic

POC VFD
Gibbs Equation = 1960 Speed control logic = if/then

#3: too many wells, too little time

1 production engineer 80/20 rule
150-200 wells 20% of wells get 80% of attention
 Problem & Solution
Problem: Production Operations not set up for success
Solution: Take ‘pump by exception’ to the next level

Every well, every day Pump by exception

Timer + Operator + Greasebook VFD + SCADA + XSPOC Adaptive Control Solution

 Lagging Automation Maturity in Artificial Lift
Higher value wells,
ESP Automation maturity and growth Current Growth
generally have telemetry

Moderate adoption of 20-

Rod Current Growth year old automation
Pump technologies

Niche application means

PCP Current Growth
lower technology adoption

Weak infrastructure and

Plunger Current Growth well economics have limited
adoption

“Re-emerging” lift type and

Gas Lift Current Growth complex system has seen
limited adoption
 Complementing & Advancing Legacy Systems

1990 25+ years of innovation 2017

Software

Modern Software UI Ambyint

Big Data
XSPOC
Machine Learning
Well Lifecycle Management
Artificial Intelligence
Torque Based Fillage Analytics

Dynocard Screening (ML)

Hardware

Smartphone Incident Characterization (ML)

POC + VFD
M2M Comms Anomaly Detection
Edge Computing
SCADA Distributed AI Physics-Based Analytics
 Automated & Intelligent Production Optimization

Production Optimization Platform (POP) High Resolution Adaptive Controller (HRAC)

SaaS Software Platform Intelligent Devices

Optimizing Production Operations through Physics-based

Analysis complemented with AI-Driven Monitoring & Control of
Artificial Lift Systems
 Seamless Wellsite Integration
Built with existing operations in mind: integrates seamlessly with any customer well
hardware/instrumentation
Embedded comms
Real-time acquisition of load transmitting to cloud software
and position sensor data to platform delivers monitoring,
drive analytics management, optimization

Real-time crank arm motion Seamless integration with

sensing to drive motion EPM/motor
analytics Seamless integration with
POC/VFD acquire data in real-
Intelligent stuffing box leak time and deliver enhanced,
detection and prediction patented control

Order of magnitude lower cost automation solution than SCADA, making business
case compelling, even on stripper wells
 End-to-End Technology Platform

Hi-Resolution Adaptive Controllers POP

Hardware Control/Optimize Communications Software

Intelligent devices Edge computing/ Integrated comms Cloud platform:

analytics monitoring,
Integration with Push-based operations &
existing Over-the-air architecture optimization
automation analytics updates
Data compression Automated well
Over-the-air Real-time physics + & encryption diagnosis
firmware updates data science
Predictive +
prescriptive
 High-Impact Analytics: Need Physics + Data Science
XSPOC considerations for modern analytics: 1) domain expertise (physics) and 2) “analytics limited
by low quality data”

Physics-based Data Science &

Analysis Artificial Intelligence
Dover/XSPOC Flutura
Ambyint
traditional lift Weatherford Spark “big data” statistics
optimization Cognition
GE/Lufkin

Deep physics-based analytics expertise + proprietary hardware to generate better data

5.8 x 1013 observations: 67,000x more than nearest competitor
Deploying actionable modern data science and AI in artificial lift
Source: Terry Trieberg, Theta at ALRDC Gas Well Deliquification Workshop 2/22/17. Presentation titled, “Well Production Automation and Diagnostics: Past, Present, and Future”.
Modern Software for Field Operations & Optimization
 Changing the Game: Stripper Well Automation TCO
5 bopd well, no
automation, 70% uptime:

● Ambyint improves
$56,000
uptime to 80%,
reduces Opex = 4-
month payback

● Conventional
$3,000 automation improves
uptime, reduced Opex
= NPV negative

Ambyint solves key economic hurdle preventing adoption of conventional automation

in low value or remote wells
 Value Prop: Reduce Opex + Increase Production

Effective Base Reduce Intervention Increase Operational

Management Cost Leverage

Increase uptime Lower failure rates Automate routing

workflows
Reduce failure cycle Move from reactive to
time proactive Focus on higher-value
activities
Mitigate under- Minimize over-pumping
pumping Bridge field and office
 Effective Base Management: Reducing Wellsite Visits

Customer Pre-Ambyint Ambyint Impact

● Visiting wellsites every day on set ● Reduction of wellsite visits by 87%

routes, regardless of wellsite issues ● Redeployment/Reduction of contract
● No remote surveillance to enable operator workforce for direct and
pump-by-exception immediate Opex savings
● Heavily people-driven and truck- ● Reduction in windshield time improves
reliant HSE
● Heavily reliant on contract ● Integration of leak detection device to
operators identify potential leak hazards

Operator calculating 3-4 month paybacks based on Opex reductions

 Reduce Intervention Cost: Diagnose/Predict Gas Lock
Customer Pre-Ambyint Ambyint Impact
First
Repeat
Ambyint Screenshot occurrence
● E&P consistently experiencing gas
lock across large portion of wells in
field
● Pump actively pumping but not
producing
● Requires physical check of Ambyint Pattern Recognition

production tanks to confirm Automatic Recommendation Normal

● Causes equipment damage/failure operation

● If diagnose/predict, can intervene to

slow/stop pump Gas lock
operation

Operator calculating 1-3 month paybacks based on this value-prop alone

Increase Operational Leverage: Unlocking Optimization Time
Technology Enabled
Ambyint Impact
Optimization Opportunities

Adaptive Metrics Direction

control
Data science & BBLs
Predictive
Treatment
Physics analysis s
Logic control (fillage) Failures
Remote on/off control
Strokes
Real time visibility

High Quality Data Visits

Ambyint Technical Overview
History & Foundation
SCADA Data Issues
Data Science & AI
Cybersecurity
Ambyint Architecture
Leveraging AWS

Nav Dhunay
COO
 Exceptional Foundation

10+ … years of data gathered from... 1000+ … oil wells.

70/30 … percentage of horizontal / vertical wells in our data lake.

5 ms … sampling rate of high resolution data.

33M+ … dynamometer cards generated with expert classification!

Volume

Variety

Veracity

Velocity
 SCADA Data Limits Data Science

SCADA ● Regular polling

frequency of SCADA
misses insightful events
SCADA Fixed ● Uneconomic to ratchet
Polling up polling without large
Frequency CAPEX spend
Load

Sticking?
Pumpjack
duty cycle Time Tagging?

Ambyint ● Ambyint HRACs capture

Stroke Based rich, stroke level data
Observation ● HRACs send insightful
events into data science
● Transmission costs
managed by Ambyint
Ambyint
Load

Time
High resolution + stroke level + event-based data enables quality data science
 Predictive Analytics

Data Analysis Big Data

Field data coming in
IOT - Ambyint
from Ambyint devices - WellExpert
Recommendation Engine - Public Data

Examples:
Gas lock / gas interference Machine
1. Training sets
Pattern recognition 2. Models Learning
3. Real data Algorithms
Automated anomaly detection

Ambyint competitive advantages of data flows and analysis

1. Only integrated end to end adaptive control solution on the market
2. Continuous feed of new data with massive data lake from WellExpert to back-test new ideas
3. True closed loop machine testing and learning cycle
 Cognitive Card Recognition: A New Approach
Visual pattern recognition
(traditional approach)
diagnosis systems are
inherently flawed

1
Humans only see macro-level
changes, missing the subtle
details hidden due to poor
data quality or high friction
2

Ambyint data science utilizes

micro-pattern diagnosis +
expert-categorized data lake
to resolve this flaw

3
 Torque-Based Analytics
Teaching the machine to predict and diagnose gas lock in oil wells.

Ambyint pattern recognition

First Repeat occurrence:

Ambyint screenshot occurrence well shut down
Normal
operation

Gas lock
operation

Operator calculating 1-3 month paybacks

based on this value proposition alone
 Creating a POC from an Electric Motor

Electric Challenge: How do you predict pump fillage from Torque?

Prime Mover

1. Collect motor torque information THIS IS HARD

Current
1. Need to determine when a stroke starts and ends

1. Apply physics based algorithms to stroke cycle

Torque
1. Machine learning model generated around features of torque patterns
a. Use max torque as up stroke, and voted method to decide starting
point

Fillage Leveraging our 33mm dynocards in our data lake; 22mm with torque
 Cybersecurity - Typical SCADA

PLC/RTU Wireless Comm SCADA Server LAN / PC

● Unsecured ● Clear text ● Unencrypted ● Unencrypted LAN

local ports storage
● Sniffable ● No user level
● Susceptible wireless ● HDD security
to DDoS
● Uptime? ● Susceptible to
malware/ransom
● Backup? ware
 Cybersecurity - Ambyint

Ambyint Device Satellite AWS Any Device

● Secure ● Push ● Encrypted ● SSL Encryption

device storage
● Encrypted ● Browser based
● No local ● 99.99%
ports ● Native apps
● Highly
● DDoS attack available ● User level and app
immune level security
 Architecture Drivers

● On the Go Pricing
● Performance
● Deployment Speed
● Security
● Flexibility
● Do what you’re good at
Alex Robart
CEO
+1 434 294 1396
[email protected]

Nav Dhunay
COO
+1 403 830 3096
[email protected]