Final Year Project

Project Title

TurboDiff: A 2D Differentiable Fluid Simulator for Wind Turbine Airfoil Optimization

Description

To maximize wind turbine efficiency, traditional design relies on computationally expensive solvers or low-fidelity tools lacking gradient capabilities. We developed TurboDiff, a lightweight, 2D differentiable CFD (Computational Fluid Dynamics) simulator specifically for Aerodynamic Shape Optimization (ASO). By leveraging Automatic Differentiation (AD) and Numerical Optimization, TurboDiff solves incompressible Navier-Stokes equations using Cartesian grids and Signed Distance Functions. This high-performance approach bypasses complex adjoint derivations, instead utilizing automatic gradients to iteratively refine airfoil geometries parameterized by Class-Shape Transformation. The research produced three high-performance airfoil designs tailored for varying power scales. Ultimately, TurboDiff provides an open-source framework bridging differentiable physics and scalable engineering, accelerating high-efficiency wind energy deployment in unique climatic regions like Pakistan’s coastal corridors.