Physics-Constrained Reservoir-Computing for Turbulence and Chaotic Learning
Explore the intersection of fluid mechanics modeling and artificial intelligence in this 56-minute talk by Luca Magri at the Alan Turing Institute. Delve into the complementary capabilities of physical principles and empirical approaches in predicting flow evolution. Discover three physics-constrained architectures: physics-informed echo state networks (PI-ESN), automatic-differentiated physics-informed echo state networks (API-ESN), and auto-encoder echo state networks (AE-ESN). Learn how these computational methodologies are applied to learning hidden variables, noise filtering, optimal design, and turbulence learning. Examine the application of these techniques to aerospace propulsion, with a focus on thermoacoustics and turbulence in Kolmogorov flow. Gain insights into how physics is embedded as soft and hard constraints in these innovative approaches to fluid mechanics modeling.