Digital twins are real-time digital replicas of physical systems that enable simulation, monitoring, and optimization. Unlike traditional models, they are dynamic (continuously updated), bi-directional (inform physical systems), and predictive (enable forecasting).
Technical challenges
To create a true digital twin, we need to combine:
- Accurate modeling
- Real-time data assimilation
- Control of the physical system using the digital twin
Combining these three aspects for complex, large scale systems with potentially chaotic behavior is the core challenge of digital twin research.
My research in Digital Twins
I use modern machine learning techniques to create better models of physical systems. With Computational Hypergraph Discovery (CHD), we can sort through large datasets with many variables to discover the underlying functional relationships and graphical structure. This allows us to build more accurate and interpretable models of complex systems. With model aggregation, we can combine multiple models to get better predictions. When designing the research, I specifically focused on aggregating models for large scale physical systems, such as Earth’s climate.