Machine-learning based turbulence prediction and control
讲座名称：
Machine-learning based turbulence prediction and control
讲座时间：
2018-10-15
讲座人：
Prof.Haecheon Choi
形式：
校区：
兴庆校区
实践学分：
讲座内容：
时 间： 2018年10月15日下午14:30-16:00
地 点： 航天航空学院教一楼第三会议室
报告人：Prof.Haecheon Choi
报告题目: Machine-learning based turbulence prediction and control
摘要：An accurate but efficient prediction of turbulence is one of the ultimate goals in fluid mechanics society. To predict turbulence, it is important to know the correlations among two or more flow variables at different spatial locations. Understanding these correlations from the Navier-Stokes equations is quite difficult and sometimes impossible. In this talk, we apply machine learning (ML) tools (from linear regression to deep learning) to predict turbulence for two different problems: one is to model subgrid-scale (SGS) stresses for large eddy simulation, and the other is to predict near-wall turbulence based on wall variable sensing. For SGS modeling, we first train the ML model for homogeneous isotropic turbulence to predict SGS stresses from resolved flow variables. Results show that the ML model outperforms traditional models such as the Smagorinsky model for the predictions of SGS dissipation, energy spectrum, and vortical structures. We also train ML models for turbulent channel flow to predict near-wall flow based on wall variable sensing. The near-wall flows predicted are very similar to real ones. How to visualize (or understand) the ‘black box’ in the process of ML is also discussed in this talk. Finally, turbulence control based on wall variable sensing is performed for skin friction reduction in channel flow, providing a significant amount of drag reduction.
相关视频