Speaker
Description
Fluid-Structure Interaction (FSI) plays a crucial role in predicting the dynamic response of systems across diverse engineering applications, from aeroelasticity to biomechanics. While high-fidelity Computational Fluid Dynamics (CFD) accurately captures the complex physics of these moving-boundary problems, the computational cost of resolving the deforming domains at every time step is very expensive for analysis and design.
We address this limitation by presenting an efficient data driven Reduced Order Model (ROM) implemented within the ITHACA-FV open source library. Our approach achieves significant dimensionality reduction by constructing a low dimensional subspace using Proper Orthogonal Decomposition (POD) and projecting the governing fluid equations via a Galerkin method.
To handle the continuous mesh deformation governed by the Arbitrary Lagrangian Eulerian (ALE) formulation we implement a Radial Basis Function (RBF) interpolation strategy. We perform RBF interpolation to train the reduced fluid operators based on the mesh deformation parameters. During the online phase we then evaluate these trained operators directly for each new deformed mesh configuration. This approach completely bypasses the expensive online construction and matrix assemblies of those operators at every time step.Furthermore the fluid and structural domains are integrated via a robust strong coupling algorithm extending partitioned solver approaches [1]. The structural dynamics are resolved using the Newmark time integration scheme to maintain numerical stability and strict kinematic compatibility at the interface.
Applied to the benchmark case of laminar flow around a moving cylinder at Re 200, the proposed framework significantly reduces the computational cost. By cutting the simulation time from 1 hour 40 minutes for the Full Order Model to just 58 seconds for the online method, this work demonstrates a fast and accurate approach for resolving complex FSI problems.