Use of machine learning to optimise turbulence modelling

Industry relies heavily on turbulence models to make numerical predictions sufficiently affordable. However, in many situations the lack of predictive accuracy of the underlying models limits the impact that computational methods can have on technology development.

In this project, a novel machine learning approach will be applied to translate the physics contained in data into tangible turbulence models with improved accuracy. In particular, this will be done by fusing the machine learning process with an a-posteriori evaluation of the novel models into a single, integrated framework. This will ensure that the models are useable, robust, and can be implemented easily.

Supervisors:

The University of Melbourne: Richard Sandberg and Mohsen Talei.

RWTH Aachen: Jens Göbbert, Antonio Attili, and Heinz Pitsch.