Machine learning has revolutionized various fields in the past decade, due to the increasing availability of data and processing power. Reynolds-averaged Navier-Stokes (RANS) simulations are the most widely applied engineering simulation methods in many critical applications such as aerospace engineering, automotive engineering, nuclear engineering, chemical engineering, and more.
The reason for widespread use of RANS is that it is computationally affordable, and simulations can be completed in a short period of time. However, the speed and convenience of RANS often comes at the cost of accuracy, as many simplifications need to be made to reduce simulation time. For the past several decades, the turbulence modelling community has focused on improving the accuracy of these simplifications and models. Recently, using machine learning to learn subtle relationships in the underlying physics and improve models has generated interest in the turbulence modelling field.
This project aims to develop new data-driven turbulence models and methods, ready for immediate use in industrial applications. Due to the relative youth of this field and the machine learning field in general, several questions need to be addressed before these models are ready for use in improving the accuracy of industrial simulations. The first critical issue is the availability of suitable data to train machine learning models. This project has already made progress in addressing this issue, by publishing a curated dataset for machine learning in turbulence.
The second critical area of this research project is improving the stability of these models, by incorporating stability enhancements into the model structure itself. Finally, this project also focuses on the ability of the machine learnt turbulence model to generalize to unseen flows. Ultimately, this project represents innovation in a critical tool used by various industries to design safer, more efficient, and cost-effective systems.