http://Predixtransform.com
Understand the developer skills required to become a great data scientist, hear case studies (scrap forecast, machine health monitoring, and energy price forecasting) of data science projects, and find how to successfully deploy data science apps on Predix.
2. 2PREDIX TRANSFORM
Agenda
Model development2
Case study: Energy price forecast4
Case study: Machine health monitoring5
Case study: Scrap forecast6
Important services in Predix1
Model deployment & usage3
3. 3PREDIX TRANSFORM
Important services in Predix
UAA
Analytics
Catalog
Analytics
Runtime
Analytics
UI
Authenticate
users
Manage
model
Execute
model
Upload and
test model
6. 6PREDIX TRANSFORM
Deploy & use – Own catalog
Development
Develop & wrap
model as per
specs
Service instantiation
Create service
instances in developer
org
Usage
Create/use apps in
developer org
Deployment
Upload & deploy
to developer
Catalog instance
7. 7PREDIX TRANSFORM
Deploy and use – foundational catalog
On-boarding &
Stewardship
Development
Develop and
wrap model per
specs
PDS Process
Certify, optimize,
standardize for
Foundational
catalog
Deployment
Make available for
subscription through
Foundational catalog Usage
Clone to user org on
subscription
9. 9PREDIX TRANSFORM
Energy price forecast
Overview
• GE Manufacturing plant
• Temperature: 10° F - 97° F
• Variable pricing and ration charges
by utility
Problem
• Manufacturing during peak times
increases cost
Solution
• Predict energy price
• Predict ration charge
• Plant plans production per
price/charge prediction
10. 10PREDIX TRANSFORM
Machine health monitoring
Overview
• GE Manufacturing plant
• Multiple machines with
maintenance incidents
• SME knowledge key to resolution
Problem
• Time lost in determining resolution
for recurring incidents
• SME knowledge lost over time
Solution
• Automate incident fix
recommendation
• Map SME knowledge for input to
recommendation logic
11. 11PREDIX TRANSFORM
Scrap forecast
Overview
• GE Manufacturing plant
• Multiple manufacturing lines
• Varied Scrap levels per process
• $ values associated with scrap per
part
Problem
• Provide means to identify relevant
root causes for future scrap
Solution
• Provide forecast of scrap levels per
part
• Tie forecasts to scrap root causes
12. 12PREDIX TRANSFORM
Five lessons to take away
• Keep models reusable
• Pick deployment option based on intended reach
• Know the environment: Analytic Catalog, Runtime, UI
• Get to know Cloud Foundry
• Help is always at hand: Predix documentation, forums,
support contacts (see next page for links)
13. 13PREDIX TRANSFORM
The toolbox
• Python, Java, Matlab | (tons of documentation out there!)
• Basic Cloud Foundry commands | docs.cloudfoundry.org
• Analytics catalog | www.predix.io/docs/#A5cFZF2V
• Analytics runtime | www.predix.io/docs/?r=212627#h6rTgHDW
• Analytics UI | www.predix.io/docs#ODbwpgV
• UAA | www.predix.io/docs#Xq8fdAir
• Predix forums | www.predix.io/resources
• Reach the Predix Data Science Products team at:
predixdatascience@ge.com