Reproduce “AirfRANS: High Fidelity Computational Fluid Dynamics Dataset for Approximating Reynolds-Averaged Navier-Stokes Solutions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Florent Bonnet, Ahmed Jocelyn Mazari, Paola Cinnella, Patrick Gallinari
This work introduces AIRFRANS, a high-fidelity computational fluid dynamics dataset containing 1000 2D RANS simulations of airfoils to benchmark machine learning surrogate models. It demonstrates that graph-based models can effectively predict and rank lift coefficients, but consistently struggle with accurately predicting wall shear stress and the drag coefficient across various data regimes and extrapolation tasks.
Key insights
- Graph-based machine learning models effectively predict and rank lift coefficients across interpolation and extrapolation tasks due to their robust pressure field inference capabilities.
- Accurate prediction of wall shear stress and, consequently, the drag coefficient, remains a significant challenge for all benchmarked ML models.
- The use of specialized metrics like Spearman's rank correlation highlights the practical utility of ML models for design optimization, even when absolute field predictions are not perfect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.