A microscale study of turbulent flow in the porous medium and at the porous/fluid interface: combining LES, DNS, and Neural Network approaches

The dynamics of microscale turbulence transport in porous media (at the scale smaller than the pore size) is not understood even for simple porous matrix geometries. This understanding requires connecting turbulence transport in porous media to the microscale flow physics. This project will elucidate the flow physics of turbulence inside a porous medium. Preliminary results show that microscale turbulence in porous media constitutes a new physical phenomenon. The scientific outcomes of the project will have significant socio-economic impacts by enabling an improved systemic modeling of porous media flows. Immediate applications include combating COVID-19 through the design of more effective filter layers in masks. There are also long-term applications in energy storage and conversion. The project will also contribute to education and training of students. The investigator plans to engage undergraduate and high school students in the development of computational fluid dynamics code and neural network models. The exposure to lab work and academic research will allow the undergraduate and high school students improve their computational skills and help cultivate their research interests. Finally, the research results will be incorporated into the investigator's graduate class on advanced convection heat transfer.

The results from this project are vital for modeling turbulent flow associated with engineering porous media. The flow field will be phase-averaged to obtain the true turbulence statistics decomposed into non-stationary mean and fluctuation components. The proposed research will also combine traditional direct numerical simulation and large-eddy simulation with neural networks to interpret and model the flow physics of microscale turbulence. Neural networks will be used because they are superior to traditional methods for processing the intricate, inhomogeneous structure of the flow field. Supervised classification will be used to visualize 3D turbulent structures which are classified according to their turbulence kinetic energy and anisotropy. A supervised autoencoder will be used to develop the first macroscale model that takes the contribution of the inhomogeneous microscale flow field into consideration. By implementing the proposed methodology with rigorous parameter variation, the observations about the microscale flow physics will lead to understanding the main features of microscale turbulence.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.