AI-Based Aerodynamic Optimization of an Airfoil Using MATLAB and ANSYS Fluent

An AI-driven airfoil design system that runs on a student laptop CPU. A 12-variable CST (Kulfan) shape around NACA 2412 is analysed with a viscous aerodynamic solver (vortex panel method with an integral boundary layer and displacement coupling), the low-cost substitute for ANSYS Fluent RANS runs in this build. The solver produces 14,400 polars for 2,400 airfoils. A deep ensemble of neural networks learns lift, drag and pitching moment, and a genetic algorithm searches the design space on the network. Three active-learning rounds re-check the best candidates with the solver and refine the network. The verified optimum raises the mean lift-to-drag ratio at 2, 4 and 6 degrees from 77.2 to 114.3 at Reynolds number 1 million, and by 14 % (Cl 0.7) to 30 % (Cl 0.9) at equal lift, while keeping 12 % thickness and the pitching-moment limit. The surrogate predicts L/D with a 1.7 % error, and 1000 predictions take 17 ms. MATLAB scripts, ANSYS Fluent journals, a meshing guide and CAD exports are included (not run on the build machine).

Category: Mechanical Engineering / Aerodynamics / Computational Fluid Dynamics / Machine Learning Surrogate Modelling / Design Optimisation

Mechanical Engineering Project M01 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.