Join Zoom Meeting

https://hku.zoom.us/j/97114343355?pwd=N77jGVVAItwvItyMbHX1FevUe67kpU.1

Meeting ID: 971 1434 3355

Password: 875169

Abstract:

Accurately predicting complex turbulent flows with pressure gradients and separation is essential for optimizing engineering designs and ensuring reliability. Wall-modeled large-eddy simulation (WMLES) offers a practical and computationally feasible approach by resolving dynamically significant turbulence scales away from the wall

while employing a reduced-order model to represent near-wall effects. However, achieving robustness and accuracy across different flow regimes in complex scenarios remains a key challenge. Many existing wall-modeling techniques are developed based on a limited range of canonical turbulent flows, restricting their applicability. To address this limitation, the development of more advanced techniques capable of handling a broader range of flow conditions has become a key focus in the wall-modeling community.

This presentation discusses our contributions to improving WMLES in two key areas: wall modeling and subgrid-scale (SGS) modeling. First, I will introduce a framework for developing a wall model capable of capturing pressure-gradient effects using multi-agent reinforcement learning. In this framework, model training is conducted in situ with WMLES grid resolutions without relying on high-resolution velocity fields. Using this approach, a wall model is trained on low-Reynolds-number periodic-hill flow and validated on higher-Reynolds-number flows over periodic hills and the Boeing Gaussian bump. The developed model successfully captures the acceleration and deceleration of the flows under pressure gradients, outperforming traditional equilibrium models in predicting skin friction. Second, I will present an a posteriori analysis of the role of SGS anisotropy in WMLES of separated turbulent flows. Simulations of flow over a Gaussian-shaped bump reveal that eddy-viscosity-based SGS models often produce nonmonotonic predictions of the separation bubble size on the leeward side of the bump under grid refinement, whereas models incorporating anisotropic stress terms yield consistent results. Through analysis, the windward side of the bump is identified as a critical region influencing downstream flow separation. Within this region, anisotropic SGS stress directly modifies the resolved Reynolds stress, significantly affecting the formation of the downstream separation bubble. These findings highlight the importance of incorporating SGS anisotropy effects in WMLES to improve robustness and accuracy in complex flow predictions.

Biography

Dr. Di Zhou is a Postdoctoral Scholar Research Associate in the Graduate Aerospace Laboratories at the California Institute of Technology. He received his Ph.D. from the University of Notre Dame and his M.Sc. from Beihang University in China. His research focuses on the modeling and analysis of turbulence and other complex fluid flow phenomena using both physics-based and data-driven approaches. His broader interests include aeroacoustics and high-fidelity flow simulation techniques.

ALL INTERESTED ARE WELCOME