PUSHING THE LIMIT FOR LARGE EDDY SIMULATION THROUGH A KINETIC ENERGY AND ENTROPY PRESERVING FR METHOD, AND MESH ADAPTATION BASED ON MACHINE LEARNING
Abstract
High-order methods have been demonstrated for large eddy simulation (LES) for flow problems with Reynolds numbers on the order of several million. However, the cost of LES has limited the range of applications. We plan to perform fundamental researches on the numerical algorithm and mesh adaptation based on machine-learning to further improve the robustness and efficiency of LES. More specifically, we plan to tackle the following two research challenges: 1) Develop and test a mesh adaptation criterion based on machine learning for LES. The output-based adjoint approach has been demonstrated very effective for steady flow problems. Its extension to unsteady flow problems is very expensive, and is rarely used. The idea which we will explore is to use the steady-state adjoint-based adaptation criterion to train an artificial neural network, and apply the trained network for unsteady flow simulations. We will employ local data as input to minimize the computational cost. This approach will be evaluated by comparing with other approaches based on the unsteady residual, or entropy. 2) Develop and demonstrate a high-order flux reconstruction (FR) method which is kinetic energy and entropy preserving (KEEP). Many LES on coarse meshes suffer from aliasing instabilities arising from the non-linearity of the governing equations. Ad-hoc treatments have been used to stabilize such simulations, e.g., through over-integration, extra solution filtering, or limiters. These treatments are either expensive (in the case of over-integration), or degrade the solution accuracy. We will explore the KEEP principles to introduce physics-based approaches to improve the stability and robustness. Since the KEEP method does not have numerical dissipation, an explicit sub-grid-scale stress model will be employed to perform LES, and compare with implicit LES.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2021
- Source ID
- FA95502010315
Entities
People
- Zhi Wang
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Kansas