Model order reduction for Eulerian-Lagrangian direct numerical simulation of particle-laden flows

0579626 - ÚT 2024 IT eng A - Abstrakt
Kovárnová, A. - Isoz, Martin
Model order reduction for Eulerian-Lagrangian direct numerical simulation of particle-laden flows.
18th OpenFOAM Workshop. Itálie, 2023 - (Guerrero, J.; Pralits, J.)
[OpenFOAM Workshop /18./. 11.07.2023-14.07.2023, Genoa]
Grant CEP: GA TA ČR(CZ) TM04000048
Institucionální podpora: RVO:61388998
Klíčová slova: model order reduction * artificial neural networks * shifted POD * CFD-DEM * OpenFOAM
Obor OECD: Applied mathematics
https://figshare.com/articles/book/18th_OpenFOAM_Workshop_2023_Book_of_unedited_abstracts/24081426

Model order reduction (MOR) is a class of methods that aim to reduce the dimension, and thus the computational demands, of a model, while also preserving its most important properties. Particle-laden flows pose two specific challenges to MOR. First, they tend to be strongly transport-dominated and, therefore, ill-suited for common modal-based techniques. Second, Eulerian-Lagrangian models are mathematically inconsistent and it is impossible to treat them via traditional projection. In this contribution, we utilize the shifted proper orthogonal decomposition (sPOD) by Reiss et al., a method created for MOR of transport-dominated systems, and combine it with interpolation via artificial neural networks (ANN) to obtain a time-continuous reduced order model without using any projection. The resulting method, shifted proper orthogonal decomposition with interpolation via artificial neural networks (sPODIANN) is data-driven and usable even for Eulerian-Lagrangian models.
Trvalý link: https://hdl.handle.net/11104/0350008