1
Lloyd F. Colegrove
Mary Beth Seasholtz
Bryant LaFreniere Chief Analytics Officer East Coast forum.
A 2015 Golden Mousetrap Award Winner
Design Tools Hardware & Software: Analysis & Calculation Software
Dow Chemical for NWA Focus EMI solution from Northwest Analytics
Read more: http://www.cnbc.com/id/102415149
2015 Manufacturing Leadership Award
Big Data and Advanced Analytics Leadership
Winners in this category will have transformed the mountains of data generated by the
typical manufacturing enterprise into actionable insights that can be used to achieve
competitive advantage. Winners, for example, will have assembled the platforms, tools,
data models, applications, processes, and skills needed to mine meaningful and timely
information from data
http://www.dow.com/news/press-releases/article/?id=10743
Enterprise Manufacturing Intelligence
Mountains of Data
Wisdom
Planning
Improvement
Internal Marketing Trailer
Enterprise Manufacturing Intelligence-
Mountains of Data
Wisdom
Planning
Improvement
Internal Marketing Trailer
Data: Is it any good? How do you know?
UCL or USL
LCL or LSL
UCL or USL
LCL or LSL
Data must be analyzed in context.
Data in context.
Y
X
Z
Is that all there is to your data?
Data in context.
X
Y
Z
Data in context
Can “rules of thumb” apply to your data?
Y
X
L.H.C.
11/10/2
0089
The
Dow
Chemic
al
Compa
ny
Pg. 9
Data in context:
“The Signals in-between the Data”
32
34
36
38
40
42
44
46
48
Variable3
8 16 24 32 40 48 56 64 72 80 88 96 104
Avg=40.13
LCL=33.30
UCL=46.96
Multivariate analysis can reveal a
change in the correlation
structure not visible with
univariate analysis.
7
8
9
10
11
12
Variable1
35 36 37 38 39 40 41 42 43 44 45 46
Variable 3
6
7
8
9
10
11
12
13
14
Variable1
8 16 24 32 40 48 56 64 72 80 88 96 104
Avg=10.02
LCL=6.70
UCL=13.34
Enterprise Manufacturing Intelligence.
The Journey Out of Darkness
Wisdom
Planning
Improvement
Internal Marketing Trailer
Enterprise Manufacturing Intelligence.
The Journey Out of Darkness
Wisdom
Planning
Improvement
Internal Marketing Trailer
Current Data Use – Poor coordination,
no obvious plan. We work, data sits.
Manufacturing
Products
Monitor
Safety
Product
ReleaseProcess Control
Data Data Data
R&D
Reports
System.
Historic
“Local”
knowledge
Newly
generated
knowledge
Future Data Use:
Data will work for us!
Manufacturing
Products
Data Data Data
Analytics
Platform (aka
Focus EMItm)
R&D
Reports
System.
Historic
“Local”
knowledge
Newly
generated
knowledge
Motivation
• Use DATA to
– Justify* actions to FIX
– Guide* actions to IMPROVE
– Prescribe* actions to make BREAKTHROUGH CHANGES
“Largest impediment to becoming more data-driven is lack of
understanding of how to use analytics*” “*Analytics: The New Path to Value”,
MIT Sloan Management Review, October 2010
What this means to us is …
– We must learn how to better
listen to the signals that our
plants are sending us and
how to respond to them.
Journey to the SOLUTION….
AnalyticComplexity
SIMPLE
COMPLEX
Dashboards for
Improvements
Organized
Data$ $$$$
Data
Alarms
Automated
Actionable
Analytics
Manufacturing
Analytics
Knowledge
Enterprise
Information
Value Delivery
Implementation of LIMS / Data Historian/ Etc.
Data
Establish new rules as
to how the data “lives”
Guiding Principles: (1) data lives in one spot only and (2)
every piece of data is owned by one entity and
uniquely identifiable.
Reveal data and new relationships
Why was this graph so hard to make?
D from 100% is good product
being flushed away
Looking at more than
Control Charts
– Need next step of what all of this data
means in the bigger context
• More than linear grabbing of data
• It is the relationship/interaction of
the data among the business
information, collaborative
troubleshooting, and other
important aspects in the
plant/process.
– Clay Shirky: “… It’s not information
overload. It’s filter failure …”
• Need to cull out the relationships
Many Control ChartsControl Charts
•Good info, useful BUT…
•Only answers questions
about individual variables
Future Workflow – as dreamed up on a paper napkin
Retrieve
Data
Analyze
Data
Join
Data
Quality
Analyst
A
Wonderful
tool
SIMCA-P
Matlab
“Services Layer”
This services Layer will
know how to interact with
all the different databases
(1) Discover what is available
& show it to the user
(2) Retrieve data once user
says what s/he wants
Manually or unattended.
Join data depending
on goals:
• Continuous
• Batch
• Multiple plants
Pirouette
Etc.
What the User Sees: A Workflow Implementation Tool
Where to Start? Our First Hurdles:
Accessing and Joining Data
• Data available in
– instrument software
– Lab information systems
– process historians
– SAP-like product systems
• Data collected at different time intervals
– Indexed differently; some in time, some in batchID
• Data integrity impacted by e.g.
– Natural plant variation
– Inappropriate plant operation
– Vagaries of chemical processes (reaction kinetics,
etc.)
Once we create an appropriate “play space” for our data, what will we achieve?
From Very BIG Data to Very BIG Knowledge
Analyze
Report
Prepare/
Distribute
Capture
Data
Aggregation
Analyze
Report
Capture
V
A
L
U
E
Automated
Manual
Data + Analytics = Intelligence
Collaboration + Intelligence = Knowledge
Machine #1 Machine #2
Process #1
Instrumentation / Devices
HMI/SCADA
Historian
Machine #1 Machine #2
Process #2
Instrumentation / Devices
Laboratory
LIMS
Process
DCS
MES
Role-specific
clients/content
Executive Management
Business Unit
Management
Corporate
Engineering/Quality
Plant Management
Plant Quality
Process Engineers
Operators
Quality
System
NWAFocusEMI
Data Integration & Analytics
IntelligenceERP
Collaboration
Center
Knowledge
Base
Manufacturing Intelligence
Historian
QC Test
Stations
Intelligence
SCM
Partnership with Vendor
• Base Abilities
– Direct data-source connectivity
– Real-time data aggregation
– Comprehensive analytics
– Real-time, role-based dashboards
– Alarm & notification services
• “Accelerating” Modules
– Knowledge Base
• Key-word searchable enterprise-wide,
collective knowledge store
– Collaboration Center
• Fully-integrated, role-based, problem-
solving workspace (with rich-content
visual communications capabilities)
25
Discussion
triggered
by Data
between
Technical and
On-site
Persons
Consult
Existing
Knowledge
Agree on
Actions
Plant makes
Changes
Integrate
learning
into
enterprise
Real time
Tracking and
Notification
Dashboard
Alert!
The new cycle of data
usage…
Data, Calculations,
Predictive models
“Big Data”
Example of Culture Change
Jul 2013
Plant
Trip
Internal
Degradation
Post
Mortem
Analysis
Jan 2014
Plant
Trip
Dashboard
Alert !
Conversation Initiated
– how to protect the
internals.
Internals
Survives
just fine
• Dashboards for similar plants in two
countries
– Contains analytical & process data
• Calculations of relevant metrics
• Teaching SPC/SQC vs. specification
cutoffs for plant monitoring
• Research and Manufacturing are
engaged!
– Detected numerous plant drifts which
have initiated conversations and actions
– Developing a collaborative culture of
proactive intervention
• Situations being fixed before they become
a concerns
Initial Results, ROI
Proactive rather than Reactive!
Ta-daa!
When we started Now
28
“I work from what is in front of me. If I can see something flashing, then
I will deal with it. If it is not right in front of me, I don’t deal with it until it
becomes a crisis!” – Typical Run Plant Engineer
Why all that red at the start?
• The variables identified by Technology Team had not been
focused on historically
– We are looking at higher order things that the plant didn’t have
inclination or resources to look at before.
• Medium and Long term trends are not typically what a Run
Plant focuses on.
– Dashboard helps Technology Team show the plant these important
variables and calculations; plant can now internalize the learnings
from troubleshooting teams.
“When you’re up to your neck
in alligators, it’s easy to forget
that the original goal was to
drain the swamp.”
29
What engagement do you want to
facilitate?
30
Strategic: Large changes in capital or chemistry or
control in as systematic effects are revealed/discovered.
Made quarterly to yearly.
Tactical: Technical Staff & Local engineers: Decisions
on the weekly to monthly timeframe. Course corrections
optimizing across multiple variables and phenomena.
Transactional: Plant Operators are changing inputs to
the plant guided by plant procedures or automatic control.
One variable at a time decisions made at the ~hourly time
frame.
Get the data packaged right
31
Next Steps
– Roll-out of Enterprise systems to other BUs
– Continue to build our Knowledge Base concept
– Expand Collaboration Center usage
– Plot next steps to Manufacturing Analytics
– Continue to develop, partner and dream.
Because our goal is still:
TOTAL Data Domination
Thank you for your kind attention!

Dow Chemical presentation at the Chief Analytics Officer Forum East Coast USA (#CAOForum)

  • 1.
    1 Lloyd F. Colegrove MaryBeth Seasholtz Bryant LaFreniere Chief Analytics Officer East Coast forum.
  • 2.
    A 2015 GoldenMousetrap Award Winner Design Tools Hardware & Software: Analysis & Calculation Software Dow Chemical for NWA Focus EMI solution from Northwest Analytics Read more: http://www.cnbc.com/id/102415149 2015 Manufacturing Leadership Award Big Data and Advanced Analytics Leadership Winners in this category will have transformed the mountains of data generated by the typical manufacturing enterprise into actionable insights that can be used to achieve competitive advantage. Winners, for example, will have assembled the platforms, tools, data models, applications, processes, and skills needed to mine meaningful and timely information from data http://www.dow.com/news/press-releases/article/?id=10743
  • 3.
    Enterprise Manufacturing Intelligence Mountainsof Data Wisdom Planning Improvement Internal Marketing Trailer
  • 4.
    Enterprise Manufacturing Intelligence- Mountainsof Data Wisdom Planning Improvement Internal Marketing Trailer
  • 5.
    Data: Is itany good? How do you know? UCL or USL LCL or LSL UCL or USL LCL or LSL Data must be analyzed in context.
  • 6.
    Data in context. Y X Z Isthat all there is to your data?
  • 7.
  • 8.
    Data in context Can“rules of thumb” apply to your data? Y X
  • 9.
    L.H.C. 11/10/2 0089 The Dow Chemic al Compa ny Pg. 9 Data incontext: “The Signals in-between the Data” 32 34 36 38 40 42 44 46 48 Variable3 8 16 24 32 40 48 56 64 72 80 88 96 104 Avg=40.13 LCL=33.30 UCL=46.96 Multivariate analysis can reveal a change in the correlation structure not visible with univariate analysis. 7 8 9 10 11 12 Variable1 35 36 37 38 39 40 41 42 43 44 45 46 Variable 3 6 7 8 9 10 11 12 13 14 Variable1 8 16 24 32 40 48 56 64 72 80 88 96 104 Avg=10.02 LCL=6.70 UCL=13.34
  • 10.
    Enterprise Manufacturing Intelligence. TheJourney Out of Darkness Wisdom Planning Improvement Internal Marketing Trailer
  • 11.
    Enterprise Manufacturing Intelligence. TheJourney Out of Darkness Wisdom Planning Improvement Internal Marketing Trailer
  • 12.
    Current Data Use– Poor coordination, no obvious plan. We work, data sits. Manufacturing Products Monitor Safety Product ReleaseProcess Control Data Data Data R&D Reports System. Historic “Local” knowledge Newly generated knowledge
  • 13.
    Future Data Use: Datawill work for us! Manufacturing Products Data Data Data Analytics Platform (aka Focus EMItm) R&D Reports System. Historic “Local” knowledge Newly generated knowledge
  • 14.
    Motivation • Use DATAto – Justify* actions to FIX – Guide* actions to IMPROVE – Prescribe* actions to make BREAKTHROUGH CHANGES “Largest impediment to becoming more data-driven is lack of understanding of how to use analytics*” “*Analytics: The New Path to Value”, MIT Sloan Management Review, October 2010 What this means to us is … – We must learn how to better listen to the signals that our plants are sending us and how to respond to them.
  • 15.
    Journey to theSOLUTION…. AnalyticComplexity SIMPLE COMPLEX Dashboards for Improvements Organized Data$ $$$$ Data Alarms Automated Actionable Analytics Manufacturing Analytics Knowledge Enterprise Information Value Delivery Implementation of LIMS / Data Historian/ Etc. Data
  • 16.
    Establish new rulesas to how the data “lives” Guiding Principles: (1) data lives in one spot only and (2) every piece of data is owned by one entity and uniquely identifiable.
  • 17.
    Reveal data andnew relationships Why was this graph so hard to make?
  • 18.
    D from 100%is good product being flushed away
  • 19.
    Looking at morethan Control Charts – Need next step of what all of this data means in the bigger context • More than linear grabbing of data • It is the relationship/interaction of the data among the business information, collaborative troubleshooting, and other important aspects in the plant/process. – Clay Shirky: “… It’s not information overload. It’s filter failure …” • Need to cull out the relationships Many Control ChartsControl Charts •Good info, useful BUT… •Only answers questions about individual variables
  • 20.
    Future Workflow –as dreamed up on a paper napkin Retrieve Data Analyze Data Join Data Quality Analyst A Wonderful tool SIMCA-P Matlab “Services Layer” This services Layer will know how to interact with all the different databases (1) Discover what is available & show it to the user (2) Retrieve data once user says what s/he wants Manually or unattended. Join data depending on goals: • Continuous • Batch • Multiple plants Pirouette Etc. What the User Sees: A Workflow Implementation Tool
  • 21.
    Where to Start?Our First Hurdles: Accessing and Joining Data • Data available in – instrument software – Lab information systems – process historians – SAP-like product systems • Data collected at different time intervals – Indexed differently; some in time, some in batchID • Data integrity impacted by e.g. – Natural plant variation – Inappropriate plant operation – Vagaries of chemical processes (reaction kinetics, etc.) Once we create an appropriate “play space” for our data, what will we achieve?
  • 22.
    From Very BIGData to Very BIG Knowledge Analyze Report Prepare/ Distribute Capture Data Aggregation Analyze Report Capture V A L U E Automated Manual Data + Analytics = Intelligence Collaboration + Intelligence = Knowledge
  • 23.
    Machine #1 Machine#2 Process #1 Instrumentation / Devices HMI/SCADA Historian Machine #1 Machine #2 Process #2 Instrumentation / Devices Laboratory LIMS Process DCS MES Role-specific clients/content Executive Management Business Unit Management Corporate Engineering/Quality Plant Management Plant Quality Process Engineers Operators Quality System NWAFocusEMI Data Integration & Analytics IntelligenceERP Collaboration Center Knowledge Base Manufacturing Intelligence Historian QC Test Stations Intelligence SCM
  • 24.
    Partnership with Vendor •Base Abilities – Direct data-source connectivity – Real-time data aggregation – Comprehensive analytics – Real-time, role-based dashboards – Alarm & notification services • “Accelerating” Modules – Knowledge Base • Key-word searchable enterprise-wide, collective knowledge store – Collaboration Center • Fully-integrated, role-based, problem- solving workspace (with rich-content visual communications capabilities)
  • 25.
    25 Discussion triggered by Data between Technical and On-site Persons Consult Existing Knowledge Agreeon Actions Plant makes Changes Integrate learning into enterprise Real time Tracking and Notification Dashboard Alert! The new cycle of data usage… Data, Calculations, Predictive models “Big Data”
  • 26.
    Example of CultureChange Jul 2013 Plant Trip Internal Degradation Post Mortem Analysis Jan 2014 Plant Trip Dashboard Alert ! Conversation Initiated – how to protect the internals. Internals Survives just fine
  • 27.
    • Dashboards forsimilar plants in two countries – Contains analytical & process data • Calculations of relevant metrics • Teaching SPC/SQC vs. specification cutoffs for plant monitoring • Research and Manufacturing are engaged! – Detected numerous plant drifts which have initiated conversations and actions – Developing a collaborative culture of proactive intervention • Situations being fixed before they become a concerns Initial Results, ROI Proactive rather than Reactive!
  • 28.
    Ta-daa! When we startedNow 28 “I work from what is in front of me. If I can see something flashing, then I will deal with it. If it is not right in front of me, I don’t deal with it until it becomes a crisis!” – Typical Run Plant Engineer
  • 29.
    Why all thatred at the start? • The variables identified by Technology Team had not been focused on historically – We are looking at higher order things that the plant didn’t have inclination or resources to look at before. • Medium and Long term trends are not typically what a Run Plant focuses on. – Dashboard helps Technology Team show the plant these important variables and calculations; plant can now internalize the learnings from troubleshooting teams. “When you’re up to your neck in alligators, it’s easy to forget that the original goal was to drain the swamp.” 29
  • 30.
    What engagement doyou want to facilitate? 30 Strategic: Large changes in capital or chemistry or control in as systematic effects are revealed/discovered. Made quarterly to yearly. Tactical: Technical Staff & Local engineers: Decisions on the weekly to monthly timeframe. Course corrections optimizing across multiple variables and phenomena. Transactional: Plant Operators are changing inputs to the plant guided by plant procedures or automatic control. One variable at a time decisions made at the ~hourly time frame.
  • 31.
    Get the datapackaged right 31
  • 32.
    Next Steps – Roll-outof Enterprise systems to other BUs – Continue to build our Knowledge Base concept – Expand Collaboration Center usage – Plot next steps to Manufacturing Analytics – Continue to develop, partner and dream. Because our goal is still: TOTAL Data Domination
  • 33.
    Thank you foryour kind attention!