How “Digital Twins” Could Help Us Predict the Future | Karen Willcox | TED

Key Concepts:

• Data-driven revolution
• Personalized data
• Mathematical and statistical models
• Data assimilation
• Prediction
• Engineering systems
• Digital twin
• Predictive physics-based models
• Computational science

1. Introduction: The Data-Driven Revolution

• The speaker begins by highlighting the prevalence of health tracking devices (Fitbit, Apple Watch) and smartphones, emphasizing the ongoing revolution in computing.
• This revolution is characterized by the convergence of data, models, and their integration for personalized predictions.

2. Elements of the Data-Driven Revolution

• Data: Devices collect personalized data about health, movements, and habits, moving beyond generic population data.
• Models: Devices incorporate powerful mathematical and statistical models, including:
  • Machine learning models (e.g., classifying running, walking, biking, or sleeping).
  • Physics-based models (e.g., physiological models of cardiac function or circadian rhythm).
• Data Assimilation: The process of continually updating models with new data collected from the system. This personalizes the models and allows them to evolve with the individual.
• Prediction: Personalized models enable tailored predictions and recommendations based on an individual's dynamically evolving state.

3. Application to Engineering Systems: The Digital Twin

• The speaker draws a parallel between the data-driven revolution in personal choices and a similar revolution in engineering systems.
• Engineering systems are generating increasing amounts of data through smaller, cheaper, and more powerful sensors.
• Engineering also relies on physics-based models that represent the governing laws of nature, allowing for predictions about system behavior.
• Example: The speaker's research group uses unmanned aircraft with finite element models to predict structural response under different conditions (e.g., takeoff, damage).

4. Defining the Digital Twin

• A digital twin is defined as a personalized, dynamically evolving model of a physical system.
• Example: A digital twin of the speaker's aircraft would incorporate data from onboard sensors and inspections, assimilated into the models.
• The digital twin captures the variability specific to that aircraft and evolves as the aircraft ages, degrades, is damaged, or is repaired.

5. Benefits and Applications of Digital Twins in Engineering

• Digital twins enable optimized decision-making for managing fleets of vehicles (e.g., airlines, cargo delivery drones).
• Decisions can be made about maintenance, optimal flight paths, and operational parameters based on the evolving state of each individual aircraft.

6. Historical Context: The Apollo Program

• The term "digital twin" was coined in 2010, but the concept dates back to the Apollo program.
• NASA used a simulator in Houston to mirror the Apollo spacecraft in space.
• Case Study: Apollo 13: During the Apollo 13 mission, data from the damaged spacecraft was fed into the simulator, which was then used to run predictions and guide decisions that brought the astronauts home safely.

7. Expanding Applications of Digital Twins

• Digital twins are moving beyond aerospace engineering into various fields:
  • Civil infrastructure (bridges) for structural health monitoring and predictive maintenance.
  • Buildings for energy efficiency.
  • Wind farms to increase efficiency and reduce downtime.
  • Natural world (forests, farms, ice sheets, coastal regions, oil reservoirs, planet Earth).
  • Medical world for medical assessment, diagnosis, personalized treatment, and in silico drug testing.

8. Challenges in Creating Digital Twins of Complex Systems

• Creating digital twins of complex systems (e.g., entire aircraft, cancer patient, planet Earth) remains challenging due to:
  • Scales: The systems cross multiple scales (e.g., microscopic damage affecting vehicle-level flight). Computational models that resolve all scales are computationally intractable.
  • Data Sparsity: Data is often sparse in space and time, noisy, and indirect. Engineers and medical practitioners are limited to external observations to infer internal conditions.
  • Prediction: Even with improved sensing technology, models are needed to predict future behavior under different actions.

9. Addressing the Challenges: Predictive Physics-Based Models

• Hope lies in predictive physics-based models that encode the governing laws of nature.
• These models can predict tumor growth, response to radiotherapy, or ice sheet flow under different temperature scenarios.
• Combining these models with machine learning, scalable methods, data assimilation, optimization, decision-making, and high-performance computing is the focus of computational science.

10. Examples of Digital Twin Research at UT Austin

• Space Systems: Managing the health and operations of launch vehicles and satellites (Renato Zanetti, Srinivas Bettadpur), and tracking space objects and debris (Moriba Jah).
• Environment and Geosciences: Modeling the Antarctic ice sheet (Omar Ghattas) and storm surge modeling for hurricanes (Clint Dawson).
• Medicine: Patient-specific heart care (Michael Sacks) and digital twins for cancer patients (Tom Yankeelov, David Hellmuth).

11. Conclusion

• The speaker expresses excitement about the future of digital twins in enabling safer, more efficient engineering systems, a better understanding of the natural world, and improved medical outcomes.