About the project
This project is dedicated to the creation of a comprehensive numerical framework, with the primary objective of comprehending and simulating unsteady boundary layers on dynamic geometries.
The research of alternative novel propulsion systems relies on the capability to produce reliable and predictive numerical simulations of such systems including details of the moving parts. In recent years, substantial advancements have been achieved in the fields of Large Eddy Simulation (LES) and Immersed Boundary (IB) modeling techniques, paving the way for accurate predictions of time-dependent flow patterns around complex and dynamic marine structures.
The novel “Immersed Large Eddy Simulation” (ILES) approach will include two main parts: the combination of IB into a CFD solver with a dynamically adaptive grid, and a deep learning model for the closure of the sub-grid-scales terms in LES.The first step is to create a tool for the high-fidelity numerical simulation of such phenomena.
The Wavelet Adaptive Multiresolution Representation (WAMR) method, developed by the project’s lead supervisor, uses the wavelet representation to generate a dynamically adaptive 3D grid. The second step is to use the database created to train generative adversarial networks (GANs) to overcome the limitations in modeling the interaction between moving walls and turbulence.
The project aims to:
- further develop WAMR to be efficiently used for massive numerical simulations (both DNS and LES) on High-performance computing (Tier-1) facilities
- investigate and model moving wall-turbulence interaction in the unique database realized with WAMR.
The main tasks of the project are:
- develop and implement an asynchronous time integrator for WAMR
- adapt the WAMR parallel algorithm to new computational resources (hybrid parallelization)
- exploit the use of GPU for some tasks such as wavelet transform
- enhancement of the scalability performance up to Exa-scale computing
- collection and production of databases (DNS) for wall-bounded turbulence (to also be used for machine learning training in parallel projects)
- investigation of interaction between moving walls and turbulence
- data-driven (GAN) modelling of the wall-bounded turbulence
- integration into WAMR of classical models as well as data-driven models
- a posteriori validation of the models (LES)
