Weitere ähnliche Inhalte Ähnlich wie Real-time Predictive Analytics in Manufacturing - Impetus Webinar (20) Mehr von Impetus Technologies (20) Kürzlich hochgeladen (20) Real-time Predictive Analytics in Manufacturing - Impetus Webinar1. Real-Time Predictive Analytics
in Manufacturing
Vivek A. Ganesan, Principal Architect
Yue Cathy Chang, Sr. Director, Business Development
Impetus Technologies, Inc.
3. The Future of Manufacturing
INTELLIGENT DATA DRIVEN
MANUFACTURING
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
4. Quality of
Management Decisions
More Data, Better Quality
Intuition
Amount of Data Analyzed
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
5. Quality of
Management Decisions
More Data, Better Quality
Relevant
Data
Intuition
Amount of Data Analyzed
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
6. Quality of
Management Decisions
More Data, Better Quality
Accurate
Relevant
Big Data
Data
Intuition
Amount of Data Analyzed
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
7. Quality of
Management Decisions
More Data, Better Quality
Quick?
Accurate
Relevant
Big Data
Data
Intuition
Amount of Data Analyzed
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
8. We Have a Big Data Situation
When traditional information systems cannot …
Store
Process
Analyze
© 2013 Impetus Technologies
Volume
Velocity
Variety
COST
TIME
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
9. Big Data in Manufacturing
Volume
• Sensors
• Machine data
• “Internet of
Things”
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
10. Big Data in Manufacturing
Volume
• Sensors
• Machine data
• “Internet of
Things”
© 2013 Impetus Technologies
Velocity
• Drinking from
the fire hose!
• Consume or
collapse!!
• Analyze at the
speed of data?
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
11. Big Data in Manufacturing
Volume
• Sensors
• Machine data
• “Internet of
Things”
© 2013 Impetus Technologies
Velocity
• Drinking from
the fire hose!
• Consume or
collapse!!
• Analyze at the
speed of data?
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
Variety
• Dealing with
diagnostics
• Binary data +
log data
• Ability to
analyze variety
of data formats
13. Big Data and the Fourth „V‟
The fourth „V‟ is „Value‟
The Value of Big Data in manufacturing is in Analytics
Gartner defines four kinds of Analytics
Descriptive
What? Who? How? Why?
Diagnostic
What if? Why not? Who else?
Predictive
What will happen when?
Prescriptive
What can I do about it?
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
14. Big Data Predictive Analytics
The formula is simple:
1. Collect data at every stage of the manufacturing process
2. Store data on a Big Data store
• Economical, Accessible, Distributed, and Scalable
3. Process data
• Manage the variety and complexity of the data
4. Analyze data
• Apply mathematical models to make predictions
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
15. Predictive Analytics Context
D
A
T
A
Sensors
© 2013 Impetus Technologies
Control
Predictions
Instruments
I
N
G
E
S
T
Analytics
•
•
•
•
•
•
Fix
Throttle
Alert
Adjust
Optimize
Abort
Model
•
•
•
•
•
Represent
Learn
Predict
Iterate
Improve
Batch
Historical
Iterative
Real-Time
Immediate
Feedback
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
Instantaneous
16. Big Data, Big Value
In manufacturing, the greatest value is in :
Real-Time Predictive Analytics
Prescriptive Analytics is possible but depends on :
Good Predictions
Fast Feedback Loop
Real-Time Predictive Analytics is the first step towards :
Intelligent Data-driven Manufacturing
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
17. Manufacturing Analytics Quadrant
REAL TIME
BATCH
Historical
"What happened"
Hindsight
DIAGNOSTIC
© 2013 Impetus Technologies
PREDICTIVE
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
18. Manufacturing Analytics Quadrant
REAL TIME
BATCH
Near-term
"What is happening"
Insight
Historical
"What happened"
Hindsight
DIAGNOSTIC
© 2013 Impetus Technologies
PREDICTIVE
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
19. Manufacturing Analytics Quadrant
REAL TIME
BATCH
Near-term
"What is happening"
Insight
Historical
"What happened"
Hindsight
Inferential
"What may happen"
Foresight
DIAGNOSTIC
© 2013 Impetus Technologies
PREDICTIVE
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
20. Manufacturing Analytics Quadrant
REAL TIME
BATCH
Near-term
"What is happening"
Insight
Influential
"Make it happen"
Intelligent
Historical
"What happened"
Hindsight
Inferential
"What may happen"
Foresight
DIAGNOSTIC
© 2013 Impetus Technologies
PREDICTIVE
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
21. Opportunities and Challenges
Business Opportunities
• Preventive Maintenance
• Real-time/near real-time actionable
response
• Improve Productivity/Margins
• Reduce wastes, improve efficiency
• Improve Yield
• High Ingestion Rates
• Sensor/tool data with subsecond ingestion requirements
• Millions of writes per second
• Complex Log Formats
• Semi-structured data
• Huge Amount of Data
• Supply Chain
• Optimize supply chain
© 2013 Impetus Technologies
Technical Challenges
• TB/PB of data storage for
deeper analytics
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
25. Input Data: Raw Logs
• Meta-data about the prospective
product line is created at a
factory site.
– E.g., number of sensors emitting log
files or readings.
• Various log files are generated:
– Containing Text.
– Containing specific Sensor readings,
continuous as well as binary values.
– At each time-step, a specific pass/fail.
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
26. Input Data: Parsing
This data is parsed into a matrix
representation
• Columns representing sensor logs
• Rows representing the time
In our dataset:
• 590 attributes, approximately every
minute
Machine 1,
sensor 1-120
– Missing Data from logs
– Failure : success :: 1 : 15
• 50,000 time-steps, i.e., 29.5 million values
(“parsed”, not raw) in a month
© 2013 Impetus Technologies
Time
• Also has labels: {+1,-1} for failure/success
from for each time-step
Machine 12,
sensor 1-92
35. Key Takeaways
•
•
•
•
Measure and Collect Everything
Process, Diagnose, and Predict
Get Real with Real-Time
Generate Actionable Intelligence
© 2013 Impetus Technologies
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
36. Talk to us about Impetus Solutions and
Services for Manufacturing Big Data
Assessment
Objectives &
Strategy
Model
Solution
Modeling
BUSINESS
PROCESS
MANAGEMENT
Analyze &
Optimize
Solution
Analysis
People,
Process,
Technology
Impact
Business Analytics
and Data Science
Solution Architecture, POC
and Production planning
Technology strategy, Use Case
development & Validation
bigdata@impetus.com
Big Data Platform Implementation
© 2013 Impetus Technologies
bigdata.impetus.com
Recorded webinar is available at http://lf1.me/hqb/
For more Info contact bigdata@impetus.com
Operations and Visualization