In the field of mechanics, we will investigate the use of Data Science to predict turbulent flows, with applications, for example, in the area of transport. We have both theoretical knowledge about turbulence as the friction produced by the thin boundary layers forming along the solid surfaces on the vehicles, and TBs of experimental data on how flow behaves. This gives us a solid foundation for bringing recent algorithms developed in the area of machine learning to model flow in controlled (simulated) scenarios. The application of established Deep Learning methods to flow estimation will raise its unique challenges that requires development of new theory and/or practice.

In this project, we propose to develop and use generic Deep Learning techniques that are able to model physical (simulated and/or measured) dynamics. In particular, we would like to predict turbulent flows in various situations, using different types of datasets as training data. The ultimate application is to reproduce datasets from direct numerical simulations (DNSs) of turbulent boundary layers at high Re. Additional applications to fluid flow will involve the development of models for the near-wall region of turbulent boundary layers, which will allow to perform very large-scale flow simulations at a significantly reduced cost. Within this project, we aim to bring the recent algorithms developed in the area of Machine Learning and develop them further to model physical dynamisms such as the flow in controlled (simulated) scenarios. The Deep Learning technologies to be studied and tailored include multi-layer perceptrons, convolutional neural networks, as well as recurrent models.