Abortion pills in Doha Qatar (+966572737505 ! Get Cytotec
SAS AX 2018 - Manufacturing Insights by William Nadolski
1. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
2. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
Manufacturing Insights:
Methods for Analyzing & Visualizing Industrial Sensor Data
Presented by William Nadolski
Senior Associate Analytical Consultant
SAS Institute
3. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
OVERVIEW
• Background and experience
• Sensor data challenges
• Sensor data opportunities
• Goals of this session
All images licensed under creative commons unless otherwise noted.
4. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Customer Testimonials
Justin Dye, Director of Honeywell Business Analytics
“I can tell you that the best money we spent was this SAS RAAS contract.
Without a doubt, we would not be where we are today without this.
Most organizations are happy with improvements in the single digits.
Two, three, four percent is a pretty big deal on an $11 billion business.
When we did this [implemented the recommendations], we got in
excess of a 30% yield improvement. A huge deal. It’s hard to overstate
how big a deal this is. This represents tens of millions of dollars.”
~Justin Dye, Director of Honeywell Business Analytics
http://players.brightcove.net/1872491364001/default_default/index.html?videoId=5573861692001
5. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Customer Testimonials
Tom Baumann, Senior Manager of Data Analytics for Honeywell IIOT
“The SAS team was able to identify a number of different issues that
were occurring during our process. They came up with 28 things that
we needed to do differently. We did that and our yield jumped from
70% to 95% over a period of five months. For us, that’s huge. In that
project alone, the savings were about four to five million dollars. That
gives you some sense of what you can do with the data.”
~Tom Baumann, Senior Manager of Data Analytics for Honeywell IIOT
http://players.brightcove.net/1872491364001/default_default/index.html?videoId=5766285180001
6. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
AGENDA
• Data Pre-Processing Techniques
• Useful Sensor Data Visualizations
• Dimensionality Reduction
• Feature Extraction Methods
• Time Series Clustering Approaches
• Motif & Discord Discovery in SAS Viya
7. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
DATA PRE-PROCESSING
Disclaimer: All recommendations are context-dependent
1. Identify run start and run end times
2. Map data to a common time index
3. Impute missing values
4. Waveform alignment
5. Data smoothing
8. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
SENSOR VISUALIZATIONS
• Run Overlay
• Panel Plots
• Spaghetti Plots
• Lasagna Plots
9. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Spaghetti & Lasagna Plots
Enables visualization of sensor behavior both within and across runs
Sensor Data 1: Spaghetti Plot Sensor Data 2: Lasagna Plot
10. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
DIMENSIONALITY REDUCTION
Why reduce the dimensionality of time series data?
• Simple Down Sampling
• Piecewise Aggregate Approximation (PAA)
• Piecewise Linear Approximation (PLA)
• Symbolic Aggregate Approximation (SAX)
11. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
FEATURE EXTRACTION
• Traditional batch-level summary statistics
• Summary statistics over discrete windows within run
• Discrete wavelet transformation coefficients
• Discrete Fourier transformation coefficients
• Representative cluster membership
12. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
CLUSTERING APPROACHES
Methods for Detecting Anomalies Across Batches
Requires a measure of distance between every pair of runs
• Euclidean Distance
• Dynamic Time Warping
• SAX Edit Distance
13. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Dynamic Time Warping
• DTW identifies the optimal warping path by:
- Calculating the distance from every element of series T to every element of series S
- Selecting the path that minimizes the distance between the two series
• Extremely computationally expensive
- Can speed up by enforcing a global search constraint
- Also has benefit of avoiding pathological warpings
14. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
DTW Clustering Demo
Trace Dataset
All Runs Plotted
Plot One Run for Each Label
(Raw Values)
Plot One Run for Each Label
(Normalized Values)
15. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
DTW Clustering Demo
Key Code: PROC SIMILARITY
All Runs Plotted by Resulting Cluster Membership
16. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
SAX Edit Distance
• Method of dimensionality reduction
• Converts numeric time series into a representational text string
• First convert signal via PAA then map PAA values onto SAX symbols
• Can then compute the edit distance between run “words”
17. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
SAX Clustering Demo
• Typically, SAX breakpoints are chosen to ensure equiprobable regions
• Demo instead utilizes simple normalization
- Original Series: Each run comprised of 275 sensor measurements
capable of spanning an infinite number of values from 0 to 1
- SAX Representation: Each run represented by 11 SAX time units
of width 25 spanning a SAX alphabet of only 6 distinct letters (a-f)
- Can then use either Generalized or Levenshtein Edit Distance between
each pair of runs to compute the similarity metric used for clustering
Top Four SAX Sequences
PAA to SAX Symbol Mapping
Resulting Cluster Membership
(Normalized Values)
18. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
MOTIF & ANOMALY DETECTION
Methods for Detecting Within-Batch Anomalies/Patterns
• Motif Discovery
• Discord Detection
• Motif Scoring
19. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Discovery in SAS Viya
ECG Heartbeat Sensor Data
• How does a “typical” heartbeat behave?
• What is the most “similar” and frequent non-trivial subsequence of length L?
20. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Discovery
• Simply need to specify the desired motif length
21. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Discovery
• How should a prototypical heartbeat behave?
22. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Discord/Anomaly Detection in SAS Viya
• What is the most surprising or “dissimilar” subsequences of length L?
• What about the second most or third most?
23. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Discord/Anomaly Detection
• Wish to identify the 10 most dissimilar discords of length 125
• Are they all “true” anomalies?
24. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Discord/Anomaly Detection
• Top three anomalies of length 125 identified
• Zoomed comparison against previously identified motif centroid
25. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Scoring in SAS Viya
• Suppose we’re interested in a specific sub-sequence
- ECG behavior immediately after subject experiences a major premature ventricular contraction
• Wish to locate other “similar” instances within the series
26. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Scoring
• Using our identified sub-sequence as a target,
we can identify the top 10 most “similar” subsequences
27. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Motif Scoring
• Closest (non-self) sub-sequence corresponds to previously
identified minor premature ventricular contraction
• Consider the potential manufacturing applications
28. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Key Takeaways
• Understand the process
• Collaboration is key
• Power of visualizations
• Data quality over quantity
• Importance of end-to-end part tracking
• Explanatory vs. predictive modelling
29. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
Questions?
30. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
Thank You!
Presented by William Nadolski
Senior Associate Analytical Consultant
SAS Institute
William.Nadolski@sas.com
https://github.com/nadolsw/
31. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Your Feedback Counts!
Don't forget to complete the mobile app session survey!
1. Go to the Agenda icon in the conference app.
2. Find this session title and select it.
3. On the Sessions page, scroll down to Surveys and
select the name of the survey.
4. Complete the survey and click Finish.
32. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
#AnalyticsX
Image & Data Attribution
Unless otherwise noted, images licensed under CC0
• Slide 3: “Communication Internet of Things” by geralt
• Slide 6: “Presentation Icon Blue” by Iconbaandar Team
• Slide 7: “Applications-databse” by dracos on Wikimedia Commons
• Slide 8: “Infographic Design” by UnboxScience
• Slide 10: “Cut crop selection tetris” by Clker-Free-Vector-Images
• Slide 11: “Icon Mathematical Plot” by Inductiveload on Wikimedia Commons
• Slide 12: “K Means Example Step 4” by Weston.pace on Wikimedia Commons
• Slide 18: “Sound Note Melody” on SVGSilh
• Trace Dataset by D. Roverso
• ECG Data hosted by Eamon Keogh, Jessica Lin, and Ada Fu
• A debt of gratitude goes out to Dr. Eamonn Keogh of UC Riverside for his approachable research on many of these techniques
• All code used in this presentation can be found at: https://github.com/nadolsw/SAS/tree/master/AX_2018