ML4PhySim : Machine Learning for Physical Simulations Challenge (The airfoil design)
Mouadh Yagoubi, Milad Leyli-Abadi, David Danan, Jean-Patrick Brunet, Jocelyn Ahmed Mazari, Florent Bonnet, Asma Farjallah, Marc Schoenauer, Patrick Gallinari
TL;DR
ML4PhySim tackles the challenge of deploying ML surrogates for CFD-based airfoil design by introducing a unified, multi-criteria evaluation through the LIPS framework on Codabench. It leverages the AirfRANS dataset with OpenFOAM ground truth to jointly assess accuracy, computational speed, OOD generalization, and physics compliance, with a transparent, reproducible scoring formula $Score=\alpha_{ML}\cdot Score_{ML}+\alpha_{OOD}\cdot Score_{OOD}+\alpha_{PH}\cdot Score_{Physics}$. The paper also integrates NVIDIA Modulus as the surrogate-modeling framework and provides practical examples (OpenFOAM baseline and FC network) to demonstrate scoring mechanics, including detailed 0/1/2 point thresholds and speed-up normalization. By offering an end-to-end starting kit and online evaluation, the work aims to standardize benchmarks for PDE-based physical problems and encourage collaboration between ML and physical sciences communities toward faster, reliable CFD surrogate models.
Abstract
The use of machine learning (ML) techniques to solve complex physical problems has been considered recently as a promising approach. However, the evaluation of such learned physical models remains an important issue for industrial use. The aim of this competition is to encourage the development of new ML techniques to solve physical problems using a unified evaluation framework proposed recently, called Learning Industrial Physical Simulations (LIPS). We propose learning a task representing a well-known physical use case: the airfoil design simulation, using a dataset called AirfRANS. The global score calculated for each submitted solution is based on three main categories of criteria covering different aspects, namely: ML-related, Out-Of-Distribution, and physical compliance criteria. To the best of our knowledge, this is the first competition addressing the use of ML-based surrogate approaches to improve the trade-off computational cost/accuracy of physical simulation.The competition is hosted by the Codabench platform with online training and evaluation of all submitted solutions.