IRTG Modern Inverse Problems (MIP)
Lade Veranstaltungen

« Alle Veranstaltungen

  • Diese Veranstaltung hat bereits stattgefunden.

SSD Serie Seminar with Andre Weiner

7. Juni 2021 @ 16:00 - 17:00

Dr.-Ing. Andre Weiner – Computational fluid dynamics and machine learning with OpenFOAM and PyTorch

Institute of Fluid Mechanics, Flow Modeling and Control Group, TU Braunschweig,


Simulations and experiments of fluid flows produce vast amounts of heterogeneous data. Oftentimes, much of the data are not used to their full extent, but only a fraction is further processed and evaluated. A potential reason is that data-processing and knowledge extraction is challenging. Typical flow simulations can produce datasets ranging from several gigabytes to petabytes. Processing such amounts of data requires parallel programming, possibly for dedicated hardware. Another challenge is the increasing complexity of models derived from such data. Modeling one, two, or three-dimensional relationships is often guided by intuition. However, for higher dimensions, statistical reasoning is required. Machine learning is a field of research that aims to automate the creation of data-driven models, at least partially. Moreover, the availability of extremely powerful open-source machine learning frameworks like Scikit-Learn, PyTorch, or Tensorflow lowers the barrier on the implementation side. Combined with a powerful open-source fluid mechanics toolbox like OpenFOAM, data-driven models can create more accurate and faster simulations or harvest numerical data to a much greater extent.

This contribution walks the audience through selected applications combining machine learning and computational fluid dynamics. After a short introduction to OpenFOAM and PyTorch follow three practical examples presented in a mix of theory, visualizations, and programming. The first example demonstrates an efficient approach to compute highly accurate mass transfer from rising bubbles through regression. In the second example, we extract coherent structures from a turbulent flow using modal decomposition and create a reduced-order model. In the last example, we learn to control the flow past a cylinder by rotating the cylinder such that the forces acting on it are reduced.


7. Juni 2021
16:00 - 17:00


a link for the Zoom meeting room will be send in the newsletter one week before the seminar starts. If you want to participate, please send an email to office@aices.rwth-aachen.de to get the zoom link.