Data-driven physical modelling
Data-driven physical modelling combines physics-based principles with machine learning techniques, enabling researchers to model complex behaviours, optimise designs, and accelerate scientific discovery across diverse fields such as materials science, engineering biology, neuroscience, and power systems engineering.
Our research in data-driven physical modelling integrates advanced machine learning with fundamental physics to accurately model and predict complex systems. With computational resources such as Isambard-AI, we have the capability to handle vast datasets and perform large-scale high-fidelity simulations. Our collaborative and interdisciplinary approach to research drives innovation in different fields across the university.
Our research covers topics such as:
- Data-driven reduced order modelling
- Surrogate modelling
- Neural differential equations
- Data-driven nonlinear dynamics
- Physics-informed neural networks
Applications of our research include:
- Decoding of stimuli for neural dynamics
- Acceleration of high-fidelity simulations for fusion energy
- Data-driven fluid dynamics for human physiology
- Surrogate models for power systems engineering
- Reinforcement learning for control of bio-engineered cells
Associated group members:
- David Barton
- Alberto Gambaruto
- Matthew Hennessy
- Tony Mulholland
- Robert Szalai
- Stuart Thomson