Neural networks allow detecting physical quantities that go beyond lab measurements.

What is it about?

In fluid dynamics research, one repeatedly encounters unknown quantities that are difficult to measure or simulate. This is especially true when flows are approximated on the basis of time averages. In this case, the influence of turbulent fluctuations must be estimated or modeled. In this study, a new method from the field of scientific machine learning is applied to problems in fluid mechanics, which makes it possible to identify so-called hidden variables by including physical equations in the learning process of neural networks. We apply the method to various fluid mechanics problems and show the great potential of the method for the study of complex phenomena in fluid mechanics.

Featured Image

Photo by Dan Cristian Pădureț on Unsplash

Photo by Dan Cristian Pădureț on Unsplash

Why is it important?

Machine learning methods have become more powerful in recent years due to the increasing availability of computing power. As a result, they are being used more and more frequently in various areas of science. In contrast to classic application areas of machine learning, such as image recognition, comparatively little data is available in fluid dynamics research. To take advantage of the power of these data-driven methods here, it is crucial to incorporate physics-based knowledge into the process. By applying Physics-Informed Neural Networks to mean field data assimilation, we show how data-driven ML methods and physical conservation equations can be combined for cutting-edge research in fluid dynamics.

Perspectives