We develop data assimilation and scientific machine learning methods that combine physical models with sparse, noisy sensor data, so that predictions of turbulent and chaotic systems can be corrected in real time.
Computers & Fluids · 107250 · 2026
A parameterised variational autoencoder with a transformer latent propagator, retrained on sparse data to follow a flow into regimes it never saw.
ICCS 2026 Workshops, Lecture Notes in Computer Science · pp. 181–188 · 2026
An interpretable vision transformer that rebuilds a full flow field from a 90%-masked, noisy input — and shows you where to put the sensors.
Computer Methods in Applied Mechanics and Engineering · 450, 118600 · 2026
A convolutional autoencoder and an echo state network, kept on track by an ensemble Kalman filter, forecast chaos in real time from sparse sensors.
Proceedings of the Royal Society A · 481(2327), 20250476 · 2025
Estimation and control at once: assimilation supplies the state that a reinforcement learning agent needs but never directly observes.
Journal of Fluid Mechanics · 1001, A49 · 2024
A digital twin synchronised with a real annular combustor, inferring the state, the parameters and the model bias as the experiment runs.
We are looking for PhD students, UROP and Erasmus+ students interested in data assimilation and scientific machine learning for fluids.
Openings and how to apply