Machine Learning for Reduced-Order Modeling (Prof. Bernd R. Noack) – Part 3

This lecture was given by Prof. Bernd R. Noack, Harbin Institute of Technology, Shenzhen, China and TU Berlin, Germany in the framework of the von Karman Lecture Series on Machine Learning for Fluid Mechanics organized by the von Karman Institute and the Université libre de Bruxelles in February 2020. We describe data-driven reduced-order modeling (ROM) approaches using powerful methods of machine learning. The focus is gray-box models distilling coherent structure dynamics from snapshot data. We highlight two generally applicable methods: the POD Galerkin method and cluster-based models. The Galerkin method has deep roots in the mathematical investigation of the Navier-Stokes equations and comes with illuminating physical insights. Yet, the accurate data-driven variant requires a rich set of tuning tools and tends to have a narrow range of validity. As alternative, the recently developed cluster-based network model is inherently robust, can easily integrate numerous operating conditions and its construction can fully be automated. However, we loose connection to first principles. Of course, there are many different shades of gray which will briefly be reviewed. The lecture concludes with a tutorial of xROM, a freely available software for POD and cluster modeling for different types grids and large volume of data.

Видео Machine Learning for Reduced-Order Modeling (Prof. Bernd R. Noack) – Part 3 канала von Karman Institute for Fluid Dynamics