how industrial big data & analytics can optimize assets and operations · 2018-08-13 · how...
TRANSCRIPT
powered by #IndustrialInternet
How Industrial Big Data & Analytics Can Optimize Assets and Operations
Build your digital industry
September 30, 2015
powered by #IndustrialInternet
Presented By
Matthew Denesuk
GE Digital
powered by #IndustrialInternet
Topics
① What is Industrial Data Science (iDS)?
② Key iDS applications & examples
③ How do you plan/do iDS
④ Digital Twin deep dive
powered by #IndustrialInternet
Sensor and asset health data are streaming in from:
• Wind Turbines • Medical Equipment • Gas Turbines • Steam Turbines • Generators • Locomotives • Oil and Gas • Aircraft Engines
GE Assets Are Generating Huge Volumes of Data
300,000+
Complex
Assets
Monitored
World Wide
1 gas turbine blade with
sensors: 500GB data/day
Illustrative
powered by #IndustrialInternet
What is Industrial Data Science?
powered by #IndustrialInternet
Industrial Data Science & the Industrial Internet
PHYSICAL & ANALYTICAL
BRILLIANT MACHINES
INDUSTRIAL BIG DATA
PEOPLE & WORK
Industrial Data Science is the
“magic” of the Industrial Internet. • Platform capabilities • Consulting & service delivery • Integration with software, UX,
& data management
powered by #IndustrialInternet
Event Equipment component failure
Ad click-thru Media purchase
Cost/value per event Huge tiny small
Freq of event tiny Huge Large
Diversity/# of “cause” variables Large small small
“Cleanliness” of data low Very high Very high
Quantity of data Widely varied Huge Large
Physics-based models Many, highly relevant ? ?
Legacy / process
integration Critical Minimal Minimal
… … … …
Therefore iDS & Industrial Internet Solutions must employ different:
1. ICT Infrastructures (data, connectivity, compute)
2. Types of people 3. Data Science techniques 4. Partnering approaches 5. Technology & business
strategies 6. Deployment approaches
Industrial Data Science is very different from traditional Data Science
powered by #IndustrialInternet 7
Physics/engineering-based models
• Need much less data
• Powerful, but difficult to maintain & scale
Empirical, heuristic rules & insights
• Easy to understand
• Capture knowledge of your experts
Data-driven techniques – machine learning, statistics, optimization, advanced visualization, …
• Usually not enough data
• Biased to parameter space of normal operation
• Easiest to maintain and scale – by far
The Three Basic Components of Industrial Data Science
powered by #IndustrialInternet 8
Physics
Knowledge
Experience
Data (lots)
Compute (lots)
Do
ma
in
Model
Data Science
DEVELOP
Optimizing Assets: From Idea to Impact
powered by #IndustrialInternet 9
Physics
Knowledge
Experience
Data (lots)
Compute (lots)
Do
ma
in
Model
Model
Data
(less, low-latency)
Compute
(less)
Workflow!!
Business
Results
Learning… .
Data Science
DEVELOP
Data Science
DEPLOY
Optimizing Assets: From Idea to Impact
powered by #IndustrialInternet 10
Sensor Data
Another Industrial Example: use advanced physical models to create new features for machine-learning (ML) approaches
Predicted Values and Ds
Variety of Machine Learning Techniques
Outcome data
New ML features :
1. Deviations from expected physics
2. Inferred or hidden parameter estimates
Provide much richer, less noisy data, resulting in much stronger predictions and models.
powered by #IndustrialInternet
“Analytics-based Maintenance” (ABM) is the future of equipment health modeling & performance management
1. A broader approach to the concept of “wear”
1. Hybrid approach for integrating & extending physics-based models
2. Adding explicit stochastic elements to “deterministic” models
3. Utilizing Massive-Scale Feature Augmentation to scale the integration of domain/physics with big data approaches
4. The optimized maintenance & management policies based on these things
powered by #IndustrialInternet
Automated
synthesis of multi-factor wear models.
Out-of-the-box ability to configure and create new ABM
models in any domain.
ABM Framework: Rapid creation of hybrid physics & big data models.
Analytics-based Maintenance (ABM): rapid, scalable, lifing & performance analytics
Historical Data
Time Series Data
Operational
Cumulative Damage
Index
Physics & Usage
Component ABM Models
Preventative Maintenance
Dynamic Pricing Operational Optimization
Replacement Forecast Inefficiency Identification
powered by #IndustrialInternet
Execution Strategies
powered by #IndustrialInternet
Predix integration & deployment
Data-science driven solutions (UX/DS/SW)
Early warning… engines, generator, pump,
motors, heat exchanger etc.
Anomaly Detection
Decision Support,
Prescriptive Apps
Diagnostics, Forecasting
& Lifing
Multi Dimensional Time Series Raw Data
Historical inspection report, M&D, & PACs
(text, drawing)
Part condition (Inspection Output)
Expert knowledge
Predix-based Applications
Design models
Environmental Data Weather, atm.
chemistry
Configuration
Shop Analytics & Mgmt.…Scrap detection, forecasting (shop
cost, parts etc.), work scope)
Reasoning & Root cause
Decision Support…Disruption management,
operational recommendations
CBM… fleet segmentation,
inspection recommendation
Customer Solutions Inputs
Model As a Service… fast
model development at a scale
Tactical data science field investigations
“GET CONNECTED” “GET INSIGHTS” “GET OPTIMIZED”
Reliability Response
Machine & Equipment Health
Maintenance Optimization
Anomaly Detection:
Suite I
Case Compariso
n
Degradation Assessment
Machine Health Early
Warning (basic)
Multi-model comparison
& ranking
Machine Health &
RUL Prediction
Asset Data Ingestion
Time-Series Data
Ingestion
Event Data
Ingestion
Data Services
Machine Health Early
Warning (Adaptive)
Case Management
Maintenance Prioritization & Scheduling
iDS Analytics Catalog
Machine Operating
State Change ID
Analytics Based
Maintenance Management
Workscope Optimization
DS Pilots and proof points
Objective: Deliver practical Data Science solutions quickly and predictably
powered by #IndustrialInternet
Maximize the value and performance of Assets
APM
Modules
Machine & Equipment
Health
Reliability
Response
Maintenance
Optimization
APM
Capabilities
CONNECTIVITY PREDICTIVE DIAGNOSTICS PERFORMANCE BENCHMARK
ANOMALY DETECTION NOTIFICATION MANAGEMENT ASSET MAINTENANCE STRATEGY/SCENARIOS
ASSET CONDITION MONITOR CASE MANAGEMENT FINANCIALLY OPTIMIZED
ASSET STRATEGY
DATA WORK BENCH RESPONSE MANAGEMENT WORK SCOPING & PRIORITIZATION
ANALYTICS WORK BENCH KNOWLEDGE MANAGEMENT INVENTORY OPRTIMIZATION
Shift Verifier
Trend Verifier
Temporal anomaly detector
Multi-sensor pattern detector
Regime detector
Asset Performance is
optimized with a wide array of algorithms
powered by #IndustrialInternet Data completeness, breadth, quality
Da
ta S
cie
nce
Co
mp
lexi
ty
Basic Reporting
Advanced Reporting
Anomaly Detection
Predictive analytics
Prescriptive analytics
Operational optimization
GE Confidential
The Asset Optimization Journey …
powered by #IndustrialInternet
Digital Twin: Driving Customer Outcomes
Build your digital industry
Sep 30, 2015
powered by #IndustrialInternet
Presented By
Mark Grabb
Achalesh Pandey
GE Global Research
powered by #IndustrialInternet
Big Data Physics Analytics Digital Twin
+
powered by #IndustrialInternet
Go backwards to understand what happened
Go forward to understand what to do
Can the Digital Twin Time Travel?
powered by #IndustrialInternet
Customers Want Outcomes
Asset Performance Mgmt. Operations Optimization Business Optimization
Single Asset Group of Assets Complex Operations
Reduce operating cost
Increase output
Reduce unplanned down time
Maintenance optimization
Increased revenue
Cost reduction
Analytics Based Maintenance Plant Level Optimization Trip Optimizer, Movement Planner
Life Models: Per Asset, Per Part and Per Failure Mode
Heterogeneous & Big Data
Accurate Performance Models
Complex Interactions
Market Conditions
Interactions at scale
powered by #IndustrialInternet
Digital Twin
1. The model is applied per asset
2. The model must be used to create
demonstrable business value.
3. The model must be adaptable.
4. The model must be used in a
continuous update capacity.
5. The model must be scalable.
Design & Test
Operation
Inspection &
Maintenance
Market
ESN #906260
Model #906260
≈
Necessary Attributes
powered by #IndustrialInternet
A Scalable Framework for Industrial Analytics
Online Sensor data
Performance Models
Data Lake and
Predix Cloud
Inspection data, e.g., Borescope
Operational State data, e.g., Temp etc.
Asset Performance Management Operations Optimization
Business Optimization
Life Models
Asset Life data, e.g., cycles to failure
Field and Service Data and Actions
Data Layer Model Layer Application Layer
Engineering models that continuously increase insights in each asset to deliver specific business outcomes.
powered by #IndustrialInternet
Customer Experience – Value Stream Co-Creation
Prioritize inspections/maintenance and
proactively manage and optimize fleet on
per engine basis
Saving tens of $MM in
unnecessary service overhauls
(per customer)
Digital Twin Business Outcomes Improved Availability
Single Component Distress Model
Cumulative Cycles ->
Cu
mu
lati
ve
Dis
tre
ss -
>
powered by #IndustrialInternet
Magnetic
Sensors
Digital Twin for Crack Simulation
24x7 Continuous Monitoring
Compressor Blade Health Monitoring
Digital Twin Business Outcomes Improved Reliability
Challenges: • No direct measurement • Supersonic tip speeds • Operational variations & S/N ratio
powered by #IndustrialInternet
Baseline
Business Outcome:
• Degradation monitoring for diagnostics and
forecasting for performance optimization
High fidelity Digital Twins Asset specific, adaptable,
scalable & continuous learning
LP compressor
issue diagnosed
Operating Time
Po
we
r Lo
ss
Digital Twin Business Outcomes Improved Performance & Reliability
Component Level Degradation Digital Twin System
Gas Turbine
Digital Twin
Adaptation
Machine Learning
Forecasting
powered by #IndustrialInternet
Digital Twin: Using Artificial Intelligence Improved Model Fidelity
Automated Inspection Data Quality & Imputation
Automated anomaly detection & classification Deep Learning, physics knowledge & machine learning
to clean & impute the data (multiple strategies)
Deep Learning changes Inspection Paradigm
Data Preprocessing
Classification
Multi-modal data fusion
Measurement & characterization
Incoming image data Inspection insight Inspection Agent
Anomaly detection
Localization
powered by #IndustrialInternet
Digital Twin Modeling Framework
Physical + Digital at Scale
Digital Twin Modeling
powered by #IndustrialInternet
Key Enabling Technologies for Digital Twin
PREDIX PLATFORM 4
Innovation, Speed, and Scale
Inspection Capabilities
Automated Data Pre-processing
T 4 9
T 3
cycles
Inputs
Domain Data 1
Dynamic Performance Estimation
Cumulative Damage
Physical + Digital Engineering Models 2
Knowledge Extraction
Model Generation & Automation
Industrial Analytics 3
In-situ / on-wing inspection
Manual inspection to Automation
Quantitative inspection
powered by #IndustrialInternet
Conclusion Digital Twin: What it is…
• Asset specific with continuous updates
• Deep physics models and advanced analytics
• Scalable & automated modeling framework
Digital Twin: What it does…
• Quantifies asset health in physical terms
• Provides continuously improving insights
• Predicts asset health, leveraging data & domain
• Enables a spectrum of customer outcomes
General Electric Company reserves the right to make changes in specifications and features, or discontinue the product or service described at any time, without notice or obligation. These
materials do not constitute a representation, warranty or documentation regarding the product or service featured. Illustrations are provided for informational purposes, and your configuration
may differ.
This information does not constitute legal, financial, coding, or regulatory advice in connection with your use of the product or service. Please consult your professional advisors for any such
advice.
No part of this document may be distributed, reproduced or posted without the express written permission of General Electric Company.
GE, Predix and the GE Monogram are trademarks of General Electric Company.
©2015 General Electric Company – All rights reserved.