http://predixtransform.com
Understand how to develop analytics models using the Asset and Analytics services within Predix. We'll start with a quick tour of the conceptual framework, and then dive deep into actual modeling and deployment examples that you can use. This session will include demo and code walk-through.
3. 3PREDIX TRANSFORM
“GE Renewable recently launched at the
Renewable APM Prognostics App, a Predix
based application that is powered by GE’s
Digital Twin technology built and run on GE
Digital’s Predix platform. It uses wind
turbine operating, maintenance and
inspection data to project future operating
conditions and predict turbine component
reliability.” – GE Renewable Energy
Digital Twin
Digital Twins transform Digital
Industrial Businesses with high-
fidelity, digital replicas of assets to
help predict, plan and optimize
business outcomes
Analytics/Models
-Math
-Physics
-Statistical
(machine learning/AI)
Operational
data
Digital Twin technology
“Why Digital Threads and Twins Are the
Future of Trains” – Jamie Miller, President &
CEO at GE Transportation
Metadata
(structure)
Context data
4. 4PREDIX TRANSFORM
Digital Twin Technology
Digital Twin Class Digital Twin Instances
Created by DT Builders Used in Apps by Developers
Analytics
Models
ML/AI
Physics
Domain
Asset
Model
Gold
Data
Connected
Digital Twins
SR & SE
Connectors
Asset Data
Operational Data
Context Data
APIs
Events
Handlers
Asset
Class
5. 5PREDIX TRANSFORM
Key Enabling Technologies for Digital twin
Continuous Data, Low Events
Event Data:
• Low number of failure events
• High cost of events &
transactions
Discrete Data, High Events
Event Data:
• 13 M Ad Clicks / day
• 5 B Amazon items / year
• 7.2 M Apple App downloads / day
• 12.7 B Alibaba orders / year
Consumer vs. Industrial
Internet
Per asset model Continuously tuned – new data / insights
Scalable – MMs assetsBusiness outcomes Adaptable – new …
Domain Data
Capabilities
Physical + Digital
Engineering Model
Industrial Analytics
Machine Learning & AI
Automated Data Pre-
Processing
Predix
Inspection Capabilities
Digital Thread
Life & Operational Behavior
Performance
Model Management
Model Generation &
AutomationKnowledge Extraction
00110
10010
11001
6. 6PREDIX TRANSFORM
Digital Twin Ecosystem
Predix builds, executes and manages Digital Twin at Scale
Models
Model
Management
Marketplace
Data
Management
Knowledge
Management
Execution
Stewardship
Data Sets
00110
10010
11001
7. 7PREDIX TRANSFORM
Rapid Model Building
SDK
Drag-and-Drop
Parallel Training
Iterate and
Feedback
Deploy
On Infrastructure:
Catalog,
Analytics,
Data, Asset,
Edge,
Security,
BizOps
Fully Integrated with
Predix Services
Intelligent Industrial
Applications
Consume
Model Execution
Monitoring
Model Management
Build, Run, and Manage
11. 11PREDIX TRANSFORM
Asset Features
Graph DB – Objects, Relationships
Open
Model Schema
Audit Service
Query Engine
Scripting Engine –
Custom Logic
REST APIs
Validation
&
Conformance
Common
Model
Template
Service
Time
Machine
System
Admin
•Extensible
•Relationship
s
•Advanced
Query •Fine grain
•Event
grouping
•“What was”
•Scripts
•Triggers
•Sync vs
Async
•Javascript
12. 12PREDIX TRANSFORM
Predix Analytics
Develop Deploy Validate Orchestrate Execute
μ
μ
μ
Upload Configure
Test Run
Manage, operationalize, scale IIOT analytics
Model
• I/O
• Params
Model
• I/O
• Params
Model
• I/O
• Params
17. 17PREDIX TRANSFORM
Step-by-step
• Define Asset
Model
• Link Asset to Time
Series
• Prepare Analytic for
Upload
• Define I/O
Mappings
• Define
Orchestration
• Link Orchestration
to Assets
18. 18PREDIX TRANSFORM
Asset Model
• Asset Model JSON (Turbines)
• Asset/Time Series integration (Tags)
{
uri: /turbines/abc-123-klm-987,
manufacturer: GE Energy,
model: 2.3-116,
attributes: {
nominalPower: {
value: 2300,
unit: kW
},
numBlades: {
value: 3
}
}
{
uri: /tags/tuv-456-mno-012,
name: Temperature,
description: Motor Temperature,
tagType: Sensor,
sourceKey: 40E503.Temp1
unit: DegC,
tagType: /tagTypes/pqr-345-xyz-567,
monitoredAsset: /turbines/abc-123-klm-987
}
{
uri: /tagTypes/pqr-345-xyz-567,
name: Temperature,
description: Motor Temperature,
tagType: Sensor,
unit: DegC,
}
Time Series tag ID
19. 19PREDIX TRANSFORM
Predix Graph Expression Language
(GEL)
/turbines?filter=attributes.nominalPower=2000..*
/tags?filter=(manufacturer=GE Energy)<monitoredAsset
/tagTypes?filter=((manufacturer=GE Energy)<monitoredAsset)>tagType
Turbine Tag TagType
monitoredAsset tagType
Turbines with
nominal power 2000
or higher
Sensor tags on all
GE manufactured
turbines
Sensor tag types on
all GE turbines
25. 25PREDIX TRANSFORM
What’s Next
E2: Data Services in Predix
E3: Edge and Cloud Connectivity
E4: Building Your First Predix App
E5: Predix Security with ACS/UAA
IIA9: Digital Twin & Industrial Machine Learning
http://www.predix.io/resources
- Core concepts of Digital Twins
- Set of Predix platform services that enable building twins
- Focus on two core services
- Other sessions will cover Edge, Data, building App with Predix UI components
- extensible asset model that provides the context for industrial apps
- rich set of features for managing the lifecycle of assets
- integrate asset with data and analytics to predict behavior – single asset and fleet of assets
- finally, leverage extensibility of the model, you can create single integrated view across system of records, even knowledge extraction
- Open extensible graph model, advanced query features
- Fine grain change tracking, with event grouping, time machine
- Scripting engine, custom logic tied to asset model events, validation/data maintenance
Platform to manage and scale industrial analytics runtime
Build and manage a catalog of analytics across lang, technologies
Automate deployment and data handling
Create complex analytics workflow
Scale to run with push of a button
Support a variety of languages and technologies
Make cloud ready – buildpacks, security, tuning/scaling
BPMN support, out of the box data handling, scheduler to automate execution
Native data source support, custom data handler
Sample twin application
Application to manage reliability of wind turbines across a wind farm
Inventory of Assets
Sensor Data
Cumulative Damage Model
Goal - predict repair/replacement
Full JSON support, complex structures
Unique object ID – URI
Relationship using URI attributes
Link asset to time series tags
Query via REST resource end points
Simple
Relationship operator
Entry point signature – JSON input string
Input parsing
Output to JSON string
Full BPMN support
Analytics reference naming convention
fieldId maps to tag name in Asset model
Combine with I/O mapping template, look up tag ID, fetch time series data, feed into analytics at runtime
Automate the running of the orchestration against a set of asset instances