AI summary by TwinMind
How “Digital Twins” Could Help Us Predict the Future | Karen Willcox | TED
TED
15m 36s
Introduction to Digital Twins Through Personal Devices
- Speaker asks audience about wearable health devices (Fitbit, Apple Watch) and smartphones
- These devices represent computing revolution over last decade
- Four key elements of this revolution:
- Data collection: personalized health/movement/habit data (not generic population data)
- Models: powerful mathematical/statistical models
- Machine learning models (classify running/walking/biking/sleeping)
- Physics-based models (cardiac function, circadian rhythm equations)
- Data assimilation: combining data and models mathematically
- Continuous model updates as new data collected
- Models evolve with user changes (aging, health changes)
- Personalizes models to individual
- Prediction: personalized recommendations tailored to individual's dynamically evolving state
Digital Twins in Engineering Systems
- Similar revolution occurring in engineering systems
- Increasing sensor data: smaller, lighter, cheaper, more powerful sensors
- Physics-based models: represent governing laws of nature
- Example: Speaker's unmanned aircraft research
- Finite element models predict structural response
- Answer questions: will structure hold on takeoff? what if wing damaged?
- Digital twin definition: personalized, dynamically evolving model of physical system
- Aircraft digital twin:
- Collects data from onboard sensors and inspections
- Assimilates data into models
- Personalized to specific aircraft (not generic model)
- Captures variability between individual aircraft
- Changes as aircraft ages, degrades, gets damaged/repaired
- Follows aircraft through entire lifecycle
Fleet Management Applications
- Airline/drone fleet operators could have digital twin for every vehicle
- Decision-making capabilities:
- Maintenance scheduling based on individual aircraft state
- Optimal flight operations for specific aircraft health
- Consider mission needs and environmental conditions
- Optimal fleet management
Historical Context and Apollo Program
- Term "digital twin" coined in 2010 NASA report
- Concept much older - Apollo program (1960s-70s) first practical use
- Apollo 13 mission example:
- Spacecraft malfunction and damage in space
- NASA fed real aircraft data into ground simulator
- Data assimilation updated simulator to match damaged spacecraft conditions
- Used simulator predictions to guide decisions
- Successfully brought astronauts home safely
Current Applications Across Domains
- Engineered world:
- Bridges/civil infrastructure: structural health monitoring, predictive maintenance
- Buildings: energy efficiency
- Wind farms: increase efficiency, reduce downtime
- Natural world:
- Forests, farms, ice sheets, coastal regions, oil reservoirs
- Potential digital twin of planet Earth
- Medical world:
- Medical assessment and diagnosis
- Personalized treatment
- In silico drug testing
Technical Challenges
- Current limitations: cannot yet create digital twins of very complex systems
- Still beyond reach:
- Entire aircraft digital twins
- Cancer patient digital twins
- Planet Earth digital twin
- Scale challenges:
- Microscopic damage affects vehicle-level performance
- Molecular/cellular changes impact human system level
- Computational models resolving all scales are intractable
- Data limitations despite "big data" era:
- Data sparse in space and time
- Data noisy and indirect
- Cannot directly measure desired parameters
- Engineering: limited to surface sensors, must infer internal wing condition
- Medicine: limited to external observations, must infer internal organ state
- Future sensing improvements insufficient alone:
- Better sensors only tell current state
- Need predictive capability for future scenarios
- Always need models for prediction
Solution Approach: Computational Science
- Hope lies in predictive physics-based models
- Encode governing laws of nature
- Examples: cancer tumor growth/radiotherapy response, Antarctic ice sheet flow
- Combine with:
- Machine learning
- Scalable data simulation methods
- Optimization and decision making
- High performance computing
- Odin Institute at UT Austin:
- Interdisciplinary computational science focus
- Faculty from 24 departments
- Tackles challenging problems
Research Examples at UT Austin
- Space Systems:
- Dawn of new space era, Central Texas involvement
- Digital twins for launch vehicles, satellites, space object tracking
- Collaboration: Renato Zanetti, Srinivas Better (Cockrell School)
- Murray Baja: space domain awareness digital twins
- Geosciences/Environment:
- Omar Ghattas: high-resolution Antarctic ice sheet physics models
- Combined with observational data for ice core drilling decisions, climate decision making
- Clint Dawson: Gulf Coast digital twin
- Storm surge modeling for hurricanes, critical decision support
- Medicine:
- Personalized medicine applications
- Michael Sachs (Odin Institute Willison Center): patient-specific heart care
- Tom Yankeelov, David Hormuth: cancer patient digital twins
- Collaborations with Dell Medical School, biomedical engineering
Future Vision
- Safer, more efficient engineering systems
- Better understanding of natural world
- Improved medical outcomes for individuals
