Numerical Analysis and Scientific Computing Seminar | Rajeev Jaiman, Bridging Scientific Computation with Machine Learning for Fluid-Structure Interaction

In this talk, I will highlight our recent efforts to accelerate high-fidelity (full-order) simulations by integrating them with data-driven models for multiphysics/multidomain analysis, with a particular emphasis on fluid-structure interaction (FSI). The main objectives are: (i) to illustrate the capabilities of our in-house multiphase FSI framework, utilizing both Eulerian-Lagrangian and fully Eulerian approaches, and (ii) to develop and apply efficient reduced-order and deep learning models to capture the physics of fluid-structure systems. I will present a systematic assessment of our methods and tools across progressively complex scenarios, along with demonstrations involving a full-scale flying bat, cavitating flexible propeller blades, and an ice-going ship in open water. A series of canonical test cases will be covered to elucidate the integration of full-order datasets with graph neural network techniques for predicting unsteady flow and fluid-structure interaction.