From connected infrastructure to AI More then advanced ... · If your boss doesn’tbelieve you...
Transcript of From connected infrastructure to AI More then advanced ... · If your boss doesn’tbelieve you...
From connectedinfrastructure to AI
More thenadvanced statistics?
Ing. Verleden Korneel
Group automation engineer
Kortrijk
3/10/2019
Intro
140countries
6.000customers
Agriculture-
Automotive-
Commercial-
Construction-
Hygiene-
Industrial-
Public-
Residential
Global Company
PolymersProductsEngineered
SolutionsFlooring
Intro
29plants
> 5000employees
18sales offices &
distribution centers
17countries
FACTS & FIGURES
At a glance
2.000mio EUR turnover
1959Roger De ClerckFounder
1991Split into7 individual groups
2005Merger of 5 groups into B.I.G.
Plants
Sales Offices& Distribution Centres
Acquire talent for innovation for future technology & operational management!
Focus on innovation
Hiring, training, coaching & career
path
Competencedevelopment
program
Phd’s & Msc’shired since 2013
50of which 40 young
graduates
Innovation
RED: Research Engineering & Development
RED: Research Engineering & Development
- Senior innovation project & portfolio managers
- Process engineers
- Material specialists
- Chemistry engineers
- Bio-engineers
- PhD digital printing
- IP manager & patent engineer
- Engineering
- Utilities & infrastructure
- Electrotechnic & automation- Equipment engineer
State of the art machinery in BIG
Independent of machine vendors: BIG –IP!
Principles of Industry 4.0 in practice
Open controllers
Remote accessible
Manageable
More than “keeping the lights on”
Vision
6
Check investments for I4.0 readyness
Work on the group identity: ATF
Prolong life expectancy existing machines
Gather process knowledge &
put into code
Build/assemble our own new machines
Lead the way in new
automation technology
Mission
7
Build faith, trust and mutual respect
with business partners
Reverse engineer existing machines:
Excellent way to gather process knowledge
Put process knowledge in code:
Scientific and mathematic objective analysis
Lead the way in new technology
Build further to develop new capabilities
Lead the ATF: spread knowledge
Organise courses, share best practices
Leverage decisions with (machine)data
Maximize learnings from previous projects
Strategy
8
If your electrical cabinet looks like this: You are not AI-ready
If your control room looks like this: You are not AI-ready
If your control room looks like this: You are not AI-ready
If your boss doesn’t believe you have a problem here: You are not AI-ready
To become AI-ready: connect the dots
Make your OT – “AI-ready”
14
Bytes “in milliseconds”
Gigabytes“Per hour”
Step 1 to become AI-ready: upgrade infrastructure
Steps to become AI-ready: upgrade infrastructure
You connect to an open system: now next steps!
Wonderware Historian: Extract machine data
Extract data from sources- Wonderware Historian database- Wonderware MES database: BOM & batch numbers- Data from dosing units- Data from energy monitoring platform
Data is not (yet) information w/o a data scientist!Information+ insight + action = value
Wonderware Historian: Extract machine data
Transform data:
• Clean up• Missing values• Duplicates• Inconsistencies in format
e.g. data/time dd/mm to dd/mm• Fill gaps
• Metadata• Name variables (ISA95)• Add labels/attributes• E.g. Data/time format
• Merge• Filter• Sort
• Aggregate• Unique identifiers• Foreign keys• Similar to SQL
Transform data: From open source/freeware to 1.5 mio software
Source:BI magic quadrant Gartner
At this stage you’ll already get information & insights
Central infrastructure private cloud: ERP/BI/WW/OTConnect to ERP
• Recipe manager
• Monitoring of batches
Connect ERP to MES:
• Energy dashboards
• OEE Reporting
Connect MES to OT:
• Load parameters to machine
• Historise machine settings
Load data: train and validate models
24
Pick your battles:
- Optimize energy consumption
- Waste reduction
- Machine output
- Predict faillures
- Predict quality
Load data
25
Use or store the processed data for modelling & analytics:
- Machine learning:
- Clustering algorithms
- Random forest algorithms
- Neural networks
- Use historical data to predict future behaviour
- Decision support
- Anomaly detection
- Risk management
- Dashboarding
- Optimize processes
Deploy and operationalize
26
Clustering Decision tree
Write code back to controller Visualise
Deploy and operationalize
Next steps for Beaulieu:
Train neural networks/deep learning
Make tools avaiable for all PDE’s
Learn how to deploy more complex ML
Adapt infrastructure for realtime scoring
…
Even if you are on the right track,
You’ll get run overif you just sit there
Next steps for you: Ask yourself these questions
It takes more then just software to develop AI:
• Are your people ready?
• Is your equipment open? Will you be able to deploy?
• Is your equipment connected?
• Can we dedicate specialists & infrastructure to ETL?
• Will your company stick to an industry 4.0 plan?
Conclusion
You might find:
• Obvious insights
• Confirmation of presumptions
• The answer halfway
But you will:
• Know the process into detail
• Have a funded opinion
Excellence covered by peopleDelight our Customer • Strive for Excellence • Focus on People • Create added Value • Act with Integrity