January 05, 2021 10:57 am
The modeling of turbulent flows is considered a challenge in science. To date, the complexity of the flow behavior is not fully understood. The effects of turbulent currents must be taken into account in many areas – for example in the construction of airplanes, cars, heart valves or weather forecasts.
Researcher at the Swiss Federal Institute of Technology Zurich (ETH) have now developed so-called reinforcement learning (RL) algorithms and combined these with physical knowledge to model turbulence. 25 years ago, the ETH did pioneering work in the coupling of artificial intelligence and turbulent flows, says Petros Koumoutsakos, professor at the Laboratory for Computational Science and Engineering at ETH Zurich Message. However, back then there were no computers that were powerful enough to test the ideas. This has now been achieved with the Piz Daint supercomputer from the Swiss National Supercomputing Center (CSCS) in the canton of Ticino. This enabled the researchers to test their ideas and successfully develop them further.
Koumoutsakos and his team believe that their method will not only be important in building cars and weather forecasting. “For the most demanding problems in science and technology, we can only solve the ‘large scales’ and model the ‘fine’ ones,” says Koumoutsakos. “The newly developed methodology offers a new and powerful way to automate multi-scale modeling and advance science through sensible use of Artificial Intelligence.”
–
