Automotive Aerodynamic Drag Optimization Using CFD + AI

A complete simulation-driven design loop for the aerodynamics of a passenger car that runs on a student laptop: a parametric car side profile with eight shape parameters, an own lattice-Boltzmann CFD solver (validated against the Schaefer-Turek benchmark) that computes the drag of each design, 100 CFD cases, five AI surrogate models that predict the drag in milliseconds instead of minutes, Bayesian optimisation that chooses which designs to simulate next, and a web application for engineers. The optimised profile has 21.4 % less drag than the baseline in a long verification CFD run (Cd 1.578 -> 1.241, 2D centre-line flow at Re 1500), with the side area (cabin and boot) kept at least 92 % of the baseline. Cd uses the same reference height for every design, so this is also the reduction of the drag force.

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

Mechanical Engineering Project M05 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.