Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Engineer 2.0:
How Human and Artificial Intelligence can Reshape the Future of Asset Management
“It is not the most intellectual of the species that survives; it is not the strongest that survives; but the species that survives is the one that is able best to adapt and adjust to the changing environment in which it finds itself.”
1963 Leon C. Megginson (not Charles Darwin….)
Our Workplaces are Changing…
Different Sensors
Changes to asset surveysFaster, higher
capacity data transfer
Large volumes of storage available
Digital design & manufacture
Global company communications
Instant digital financial transactions
The Evolution of Asset Management
The 3 Laws of Digital:
Moore's Law: The power of chips doubles about every 18 months
Kryder’s Law: The size of storage doubles about every 12 months
Nielsen’s Law: The size of effective bandwidth doubles every 21 months
Asset Information
What is your average time to deliver into BAU a mid – large scale technology programme?
Asset Information
Lifecycle Delivery – Acquire/Dispose
Adidas SpeedFactory, Ansbach Germany
3 more planned (US, GBR, FRA)
Efficiency driver:
• Higher production
• Lower Costs
• Easier logistics (lower risk)
160 jobs created at Oechsler Motion GmbH
Strategic Partner of Adidas who built and now operate the plant
Lifecycle Delivery – Operate / Maintain
First “lights out” factory for Adidas
Organisation & PeopleNEW ORGANISATIONAL MODELS
Characteristics of New Nature of Organisations
More receptive, adaptive and generative -- always focused on meeting the needs of stakeholders.
New forms of organizations often exhibit the following characteristics:
1. Strong employee involvement 2. Organic in nature3. Authority based on capability4. Alliances 5. Teams 6. Flatter, decentralized organizations 7. Mindfulness of environments, changes, patterns and themes
Network Structure Self Organising
Self Managed Teams Learning Organisation
The Evolution of Artificial Intelligence
40’s: Goldfish computers
Before 1949 computers lacked a key requisite for intelligence: • they couldn’t store commands, only
execute them
In other words, computers could be told what to do but couldn’t remember what they did
60’s: Learning from the eyes of flies
Frank Rosenblatt, a psychologist at Cornell, was working on understanding the simple decision systems present in the eye of a fly
use light patterns to determine whether to fly away from trouble or not.
60’s: Learning from the eyes of flies
Mark I Perceptron Model:
Simple input output relationship
A Perceptron takes in inputs, takes a weighted sum and returns ‘0’ if the result is below a threshold and ‘1’otherwise.
Inputs:Continuous light intensity values
Outputs:1 = Fly away0 = Remain
80’s: Neocognitron
Inputs:Image of an octagon
Outputs:Recognizesoctagon
Process:Breaks shape down into “features”
Kunihiko Fukushima proposes a hierarchical, multi-layered Neural Network.
Precursor for modern handwriting, image and pattern recognition.
Fast forward to 2018: Google Duplex
Google demo’d AI that could make appointments for users• Knew when to use speech disfluencies
(i.e umms, mhmms)
Contextual awareness is key for accurate bookings and efficiency
Content Recommendation
Monitors user behaviour, browsing history and clicks
2 3Machine Learningprofiles users based on interests
Recommends similar content based on user score
Content based versus Collaborative filtering
Sample Input Data: IP addressDevice, OS, browserImpressionsClick-through-rates%time on pages
Fraudulent transactions
Monitors baseline financialtransactions 3 Fraudulent
transaction is detected
Point based anomaly:- Distance based on value (i.e. $),location (lat/long of transaction)
Contextual anomalies:- Distance based on clustered behaviour
Archetypes of today’s industry challenges
Requires deep awareness of contextWith so much data, it’s hard to differentiate between noise and peaks, behavioral shifts versus equipment failure.
Imbalanced datasetsMore ”normal” (>99%) data than “anomalous” (<1%) data
Complex, hard to model behaviourHumans are arguably the most complex systems, we behave irrationally yet ML has enabled us to model/predict/anticipate our behaviour. Can we adapt this for industrial systems?
The Blended Workforce of the Future….
ENGINEER 2.0
• High volume transactional activities
• Always on (24/7)
• Train once, repeat often
• High speed to value
• Direct corporate knowledge
• Help to de-risk skills shortage
• Less baggage!
• Travels well
Forward Looking, Evidence Based Decisions
DATA
INFORMATION
INSIGHT
DECISION
Forward Looking, Evidence Based Decisions
DATA
INFORMATION
INSIGHT
DECISION
Forward Looking, Evidence Based Decisions
DATA
INFORMATION
INSIGHT
DECISION
Connecting to HARVI…
Archetypes of today’s industry challenges
Google Duplex: HARVI:
Consumer Virtual Assistant
Content Recommendation
Fraud Detection
Industrial Virtual Assistant
Intelligent Control & Maintenance
Event Detection
Event Detection
What is an event?
Events can be defined as any change in behaviour in a system. This could be a result of:
- process shifts (i.e. new influent streams), - asset failures (i.e. bursts, transients), - sensor failure (i.e. sensor drift or fault)- maintenance activities (i.e.
scheduled/reactive, nearby construction)
Sensornoise
Processshift
Aeration Lane Data: What’s wrong with it?
Aeration Lane Data: Expected Behaviour
PassiveUse statistical techniques to automatically detect if a real-time signal is significantly different from expected baseline behaviour.
This method does not have any explicit understand of the system domain, it is pure statistics. It cannot make causal inferences.
ActiveUse SME’s field knowledge and past experience to automatically detect if a signal is anomalous. The SME labels anomalous events that have occurred in the past - i.e. when they knew a burst occurred before. HARVI then “remembers” these events. In the future, if HARVI sees signals that resemble that past event, it flags the signals proactively as an event.
By using SME insight, HARVI now understands the causal link between events and signals.
How can we detect events?
Active Event Detection based on signal shape
Pressure Data: Identifying Business As usual
Pressure Data: Identifying Business As usual
Pressure Data: Identifying Possible Events
Turn data into information fast and consistently
Works with SMEs to be taught how to recognise failure patterns and asset behaviour
Provides unprejudiced decision support
Reshaping Asset Management
ENGNEER 2.0
A new mix of workforce is upon us….
TRUST?
CONTACT US
MARK KANEY
CHRISTOPHER STEELE
THOUHEED GAFFOOR
MOHAMAD VEDUT