0602981 - ÚT 2025 eng A - Abstract
Kovárnová, Anna - Isoz, Martin
On model order reduction of parameter-dependent particle laden flows.
[COMPUTATIONAL MECHANICS 2024 conference with international participation /39./. Hotel Srní, 04.11.2024-06.11.2024]
Method of presentation: Přednáška
Institutional support: RVO:61388998
Keywords : model order reduction * shifted proper orthogonal decomposition * CFD-DEM
OECD category: Applied mathematics

Fully resolved CFD-DEM simulations of particle-laden flow (PLFs) are computationally intensive, which strongly limits their usability for optimizations or parametric studies. Still, the computational costs can be decreased by methods of model order reduction (MOR) that replace the original model with a low-dimensional reduced order model (ROM) in a controlled manner and increase its applicability. In our work, we employ the shifted proper orthogonal decomposition (sPOD), a method tailored for MOR of transport-dominated systems, of which PLFs are an example, and combine it with a regression using artificial neural networks to obtain ROM continuous in time and parameter. The approach is an a posteriori one, i.e., PLFs for several parameter values are simulated using our OpenHFDIB-DEM solver (see the talk of M. Isoz) and the obtained solutions are used for the preparation of the ROM, which is then able to quickly predict solutions for previously unseen parameter values.
Permanent Link: https://hdl.handle.net/11104/0361344