Modeling and simulating fluid flows using physics-informed neural networks and computational fluid dynamics
Loading...
Date
2024-07-05
Authors
Λάσδας, Χρήστος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This MSc thesis deals with Physics-Informed Neural Networks (PINNs) and their applications in Fluid Dynamics. In a world where Machine Learning and Artificial Intelligence (AI) is rapidly gaining ground and making strides, the Computational Fluid Dynamics field has to follow along. In this thesis we will analyze the “black-box” of Neural Networks and we will describe how they operate, train and predict in detail. We will, not only dive into theoretical concepts behind Neural Networks and PINNs, but we will also thoroughly scrutinize the mathematics involved inside the networks’ algorithms. Furthermore, since we are dealing with fluid flows, we will derive the governing equations for fluid dynamics, the Navier-Stokes equations. Afterwards, we will model and simulate the Steady State flow inside a Channel, as well as for a Lid Driven Cavity. We will train PINNs for different scenarios and extract the interpolation and extrapolation results. The PINNs’ simulations and predictions will be verified and compared to conventional Computational Fluid Dynamics (CFD) solvers in order to check their validity. Transient simulations of the same cases will follow up. Last but not least, the more complex flows around a Cylinder and an Airfoil will be studied qualitatively. Finally, we conclude with the thesis’ discussion and possibilities for future studies.
Description
Keywords
Physics-informed neural networks (PINNs), Computational fluid dynamics (CFD), Fluid flows, Neural networks, Steady-state flows, Transient flows, Unsteady flows, Interpolation, Extrapolation, Channel flow, Pipe flow, Lid driven cavity, Flow around an airfoil, Flow around a cylinder, Ansys fluent