A Data-Free Physics-Informed Neural Network Framework for Laminar Flow Simulation Over a Backward-Facing Step and a RAE۲۸۲۲ Airfoil

Accurate modeling of fluid flow has long been regarded as a fundamental challenge in computational fluid dynamics. Although conventional CFD techniques are widely used, their application in iterative design and optimization problems is often limited by high computational cost and strong dependence on fine Mesh generation. In this study, Physics-Informed Neural Networks (PINNs) are employed as a novel data-free numerical framework to solve the two-dimensional incompressible laminar Navier–Stokes equations. Three benchmark problems are investigated, including the backward-facing step flow and the flow around a RAE۲۸۲۲ airfoil at angles of attack of ۰° and ۱۰° under a Reynolds number of ۵۰۰. The neural network is trained solely based on the governing equations and physical boundary conditions, without using any experimental or numerical data. The predicted results are validated against high-fidelity CFD simulations performed using ANSYS Fluent. A close agreement between the velocity and pressure fields, as well as the flow structures, demonstrates the strong capability of the proposed framework in accurately reconstructing the flow physics. The findings indicate that physics-informed neural networks, even in a purely data-free configuration, exhibit significant potential for solving partial differential equations governing laminar flows and can serve as an efficient alternative or complementary tool to conventional CFD methods. Finally, extending the proposed methodology to turbulent flow regimes is suggested as a promising direction for future research.

Keywords:

Physics-Informed Neural Networks , Laminar Flow , Airfoil Flow , Artificial Intelligence