Surrogate-based Bayesian optimization of an F1 front wing

Project · Master's thesis

My Master's thesis. An end-to-end aerodynamic design and optimization pipeline for a parametrized F1 front wing and nose assembly compliant with the FIA 2026 regulations, integrating CAD automation, CFD, and data-driven optimization.

Parametric geometry

The geometry is fully parametrized in CATIA V5 and controlled via Python, enabling automated design updates through parameter-driven loops. A single design regenerates across 26 design variables, and each candidate is exported twice — once per aerodynamic mode.

Design space exploration

Design space exploration is implemented using Latin Hypercube Sampling, with high-fidelity CFD simulations coupled to Bayesian Optimization to identify optimal aerodynamic configurations. Surrogate machine learning models approximate the aerodynamic performance and accelerate convergence within the design space, so the expensive CFD runs are spent only on training rather than on search.

Two competing objectives

The problem is genuinely multi-objective: maximize downforce in the cornering mode while minimizing drag in the straight-line mode. The optimization therefore returns a Pareto front rather than a single design, and the final selection is made per circuit using the fraction of each lap spent with DRS open.

Experimental validation

Final designs are validated through high-resolution CFD and experimental testing, including 3D-printed models at 32.4% scale with actuated configurations (high-downforce and low-drag) tested in a wind tunnel to assess aerodynamic coefficients and correlation with simulation.

The methodology and results are broken out across several papers, listed below.

Papers from this work

• Automated CAD–CFD optimization pipeline
• Grid-convergence study of the front-wing CFD
• Surrogate modelling and NSGA-II optimization
• CFD–wind-tunnel correlation of a scale model