Mesh Twin Learning - Optimization of Smart Factories with Mesh Twin Learning. Presentation by Maciej Mazur from the Rethink! Smart Manufacturing conference in Berlin 01.10.2019
Le nuove frontiere dell'AI nell'RPA con UiPath Autopilot™
Mesh Twin Learning - Optimization of Smart Factories with Mesh Twin Learning
1.
2. MACIEJ MAZUR · CHIEF DATA SCIENTIST
Mesh Twin Learning
Optimization of Smart Factories with Mesh Twin Learning
3. Agenda
− Key Enabler – Technology
− MTL – Practical Implementation
− MTL – Power and Benefits
− MTL – Where to start
4. Made with by PGS Software ·LINKEDIN.COM/IN/MM4ZUR
Maciej Mazur
Chief Data Scientist
10+ years of experience
At PGS Software since May, previously HPE and Nokia
Responsibilities
• Design and implementation of ML solutions
• Training junior data scientists at PGS Software
• Research focused on AI applications in security
• Active clients and team support
6. Made with by PGS Software ·
It’s no longer „if” but „when and how”
Surveys state that 93% of industry segments have
started adopting Industry 4.0.
Adoption of technology is growing
Each level of adoption at factories introduces new
technology, approach or pattern, which requires
investments into R&D’s and pilot projects
Evolution in three core areas
The biggest efforts are made to extend capabilities in
the areas of connectivity, intelligence and flexible
automation.
The Fourth
Revolution
And The Reality
We Must Face
7. Made with by PGS Software ·Deloitte University Press, The smart factory, 2017 7
Industry 4.0 – Adoption Levels
8. Made with by PGS Software ·
“ “
Richard Kelly, a McKinsey partner
The challenge is to roll out
successful pilot projects to
the entire organization.
That’s what makes the
transformation happen.
8
9. Made with by PGS Software ·
Where Are The Biggest Challenges?
Technology Stack
Amount of the technologies
increases. Skills are
required from multiple
areas: Cloud, IoT, Edges,
Digital Twins, ML/AI etc.
Operability
Changes are affecting all
aspects of the factory:
infrastructure,
networking, processes
and quality. Not without
issues.
Long-term vision
Companies are still suffering
from lack of clear
strategies and justified
business cases, that will
be expressed in ROI.
Security
Data-driven is a must. The
concern is how to protect
the critical data from
exposure or misuse at the
decision automation level.
9
10. Made with by PGS Software ·
“ “
Richard Kelly, a McKinsey partner
Despite this focus and enthusiasm,
companies are experiencing pilot
purgatory.
They have significant activities
underway. But they are not seeing
meaningful bottom-line results.
10
11. What Is The Key To Fully
Enabled Industry 4.0?
12. Building an Industry 4.0 Enabler
Amount of Technologies
Expertise from Cloud, Edge
Computing, IoT, Digital Twins
etc.
Industry 4.0 Requirements
Security, Strategy alignment,
Operability, Automation, ROI
and KPI’s
(Micro) Optimizations
Based on the work
presented by Taguchi in
the late `50s
Our story started during a
coffee break…
13. Made with by PGS Software · 13
Mesh Twin Learning – Digitalization of Data
14. Made with by PGS Software · 14
Mesh Twin Learning – Local Modeling
15. Made with by PGS Software · 15
Mesh Twin Learning – Full Cloud Integration
16. Made with by PGS Software · 16
Mesh Twin Learning – Mesh of Smart Factories
18. Made with by PGS Software ·
Machine Learning on Different Levels
Component Twin
§ Smart control of engine
speed, automated QA
gate
§ Alerts on component
failures
Process Twin
§ Optimization of energy
consumption at factory
level
§ Forecasting over factory
velocity
System Twin
§ Automated nests e.g.
QC station with
automated report
§ Conveyor belt predictive
maintenance
Asset Twin
§ Parameter optimization
e.g. furnace temperature
§ Steel baking controls
§ Liquid ingrediencies
mixer
18
19. Made with by PGS Software ·
Edge computing on AWS IoT Greengrass
Enables data aggregation and local machine learning
inference, using models created, trained and optimized
in the Cloud.
ML with AWS SageMaker Neo
Optimized models that run twice as fast, as regular
ones, with less than a tenth of the memory footprint.
Additional benefit: no loss in accuracy!
Data Science using AWS SageMaker
Service provides data scientists with the ability to
build, train, and deploy machine learning models quickly.
20. Made with by PGS Software ·
MTL
Solves Most
Technical
Challenges
§ We share light ML models, not vast amounts
of data (efficiency)
§ Solution not depended on IoT or machine
vendors (flexibility)
§ Automated optimization for target devices
(optimization)
§ Secured infrastructure, and full data
governance and control (safety)
§ ML models competition, based on simulation
within the Cloud (optimization)
§ Automated processes with quick distribution
(cost reduction)
20
25. Made with by PGS Software · 25
Use cases you can run on MTL
Production
optimization
Predictive
maintenance
Automated QA
Energy usage
optimization
Operator
augmentation
Anomaly
detection
27. Made with by PGS Software ·
Build up MTL
for 3 production
lines
Connect to the cloud,
add edge devices to
run local models and
analyze the results
Ready to grow
Ready made
framework that you
can share across
other plants and start
benefiting from scale
Assessment
What sensors and
actuators are
available already?
What infrastructure
and connectivity is
there?
Final QA gate
automation
Low hanging fruit, ie.
adding thermal cameras
can automate QA
inspection
27
Example of MTL Introduction
STEP 1
STEP 2
STEP 3
STEP 4
28. Made with by PGS Software · 28
Typical Project Timeline
Initial assessment and
data exploration
100 DAYS TO MTL
Day 1
Build full MTL
infrastructure for one
location
Day
60
Building MVP for main
use case
Day 10
Working MTL solution
you can scale to share
with other plants
Day 100
29. Made with by PGS Software · 29
MTL Outcomes
Unified strategy
for IoT 4.0
Competitive
advantage
Swarm
intelligence
1 2 3
30. Let’s get in touch
Maciej Mazur
mamazur@pgs-soft.com
@marowid
www.pgs-soft.com