Final Year Project

Project Title

Turbodiff: Differentiable Fluid Dynamics Simulator for Airfoil Shape Optimization

Description

In this project, we engineered a custom Computational Fluid Dynamics (CFD) simulator with a novel shape optimization pipeline leveraging differentiable fluid dynamics to maximize airfoil efficiency. Traditionally, aerodynamic optimization requires computationally expensive iterative testing. Turbodiff solves this by applying deep learning-inspired back-propagation directly through the fluid simulation to perform gradient-based optimization of the airfoil geometry. The core simulator is built using JAX for fluid dynamics and incorporates Reynolds-Averaged Navier-Stokes (RANS) equations to efficiently model complex turbulent flows. Additionally, a secure web application enables users to simulate and optimize airfoils over the web and store/retrieve simulation outputs. Primarily, we aimed to provide an efficient, AI-integrated computational approach to aerodynamic design, allowing rapid optimization of airfoils for specific local atmospheric conditions.