"How Pirelli uses Domino and Plotly for Smart Manufacturing" by Alberto Arrigoni, Senior Data Scientist, Pirelli (pirelli.com)
Abstract:
Pirelli, a global performance tire manufacturer, uses data science in its 20 factories to improve quality and efficiency, and reduce energy consumption. For this “Smart Manufacturing” initiative, Pirelli’s data science team has developed predictive models and analytics tools to monitor processes, machines and materials on the factory floors. In this talk we will show some of the solutions we deploy, demonstrate how we used Domino’s data science platform and Plot.ly to build these solutions, and discuss the next steps in this journey towards predictive maintenance.
Bio:
Alberto Arrigoni is a data scientist at Pirelli, where he works to process sensors and telemetry data for IoT, Smart Factories and connected-vehicle applications.
He works closely with all major business units such as R&D, industrial engineering and BI to develop tailored machine learning algorithms and production systems.
He holds a PhD in biostatistics from the University of Milan Bicocca and prior to joining Pirelli was a staff data scientist at the National Institute of Molecular Genetics (Milan), as well as a Fulbright student at the Santa Clara University and visiting PhD student at Pacific Biosciences (Menlo Park, CA).
3. Smart Manufacturing - Industry 4.0
“ […] the current trend of
automation and data
exchange in manufacturing
technologies. It includes
cyber-physical systems, the
Internet of things and cloud
computing? - Wikipedia
4. Smart Manufacturing - Industry 4.0
“ The current trend of
automation and data
exchange in manufacturing
technologies. It includes
cyber-physical systems, the
Internet of things and cloud
computing? - Wikipedia
5. Real Time Analytics
Pirelli Smart Manufacturing Vision
Predictive
Manufacturing
Advanced Data
mining
Data products
Prescriptive
Manufacturing
Predictive Models
Algorithms
Detect trend, outliers,
issues
in (near) real time
Build and deploy models
that can forecast product
quality from process data
M2M communication for
Process tuning and
resource allocation
With the goal of maximizing
quality and efficiency
ML + Smart
integrated
communication
Virtual
Factory
7. Controlled user group Factory users
Production Deployment
Iteration loops
Fast prototyping
Data Products Development
8. Smart AlertingKPIs visualization
and analytics
- Data is not human interpretable
- Large Volume
- Non straightforward KPIs
- Machine Learning / Algos
Examples:
- Trends detection
- Anomaly detection in
production process
Alerts triggers actions that can
be validated by visualization
- Data is human Interpretable:
- Small volume
- Straightforward KPIs
- Descriptive Analytics
Examples:
- Imbalance detection
(for production cycle times)
- Production efficiency KPIs
visualization
Real time Visualization triggers
actions
Data Products Categories
9.
10. Fitting density distributions on
cycle time data
Detection of discrepancies
between different distributions
GOAL: Analyze process time
discrepancies on different
machines
Curing timeMachine 1
Machine 2
Machine 3
Example of KPI visualization: Cycle time Machine imbalance
11. Product
category 1
Product
category 2
Product
category 3
Cycle Time
Results are ordered
according to the
Discrepancy calculated
between the
distributions
Cycle-time density
distributions
Machine 1
Machine 2
Machine 3
Imbalance: Prioritization of intervention
12. Final products
Uniformity KPIs
(high dimensionality data)
Step A Step B
Trend detection by
product category
Step C
Products processed in Step B within a time delta
contributed to the low final quality.
M-01
M-02
M-03
M-04
Example of Smart Alerting: Quality assessment
13. An alert is sent
every time a trend is
detected.
Python code
running on
renders reports for
decision support
Plotly with Pandas
(Cufflinks) served
via a
14. For each KPI we can identify trends using a Sliding window on a rolling basis
(4 hours batches, analysis is run every hour)
Time
Trend detection on uniformity KPIs
How did we implement and deploy it?
15. Implement tests (almost) from
scratch using Numpy/Scipy Use tests implementation from the R
packages (served by a Domino API
endpoint)
Option 1 Option 2
Deploying trend detection
Our codebase is mostly Python
Lots of R packages for Time series Analysis
16. This is all it takes to start an R API endpoint from Domino:
*Mann-Kendall test for monotonic trend in a time series z[t] based on the
Kendall rank correlation of z[t] and t (Hipel and McLeod, 2005)
Example of Python/R integration: trend detection on uniformity KPIs
17. Input time series
Test for trend significance
Example of Python/R integration: trend detection on uniformity KPIs
18. Machine learning Model (One-class
SVM with RBF kernel)
Batch model training
(~ once a week) on reference
data -> inliers’ dataset
Anomaly/novelty detection: i.e. classifying new data as similar or different
to the training set
Anomaly detection in production process: ML approach
19. We want to identify not the anomaly in “some” process parameters but we
want to label the process overall as an outlier
Normalized reference distributions for process parameters (training set)
p1 p2 …
Anomaly detection: a machine learning approach
21. 1. User centric approach: factory folks are the key
2. Provide tools for data exploration to factory folks
130 people Trained
Trained from exploratory data analysis to deploy a web app
Smart manufacturing: training for people in the factories
22. Visualization plays a big role in factories as mean to convey key information to
the workforce in the field
Domino + plotly have provided a nice combination for:
• Fast prototyping / iterating with users in the exploratory phase
• Combine output from algorithms and Machine learning models into
interactive web-based visualizations to be used in production
Next steps:
• Expand towards predictive and prescriptive manufacturing
Summary