SURF: A Generalization Benchmark for GNNs Predicting Fluid Dynamics

Stefan Künzli, Florian Grötschla, Joël Mathys, Roger Wattenhofer
Proceedings of the Second Learning on Graphs Conference, PMLR 231:13:1-13:23, 2024.

Abstract

Simulating fluid dynamics is crucial for the design and development process, ranging from simple valves to complex turbomachinery. Accurately solving the underlying physical equations is computationally expensive. Therefore, learning-based solvers that model interactions on meshes have gained interest due to their promising speed-ups. However, it is unknown to what extent these models truly understand the underlying physical principles and can generalize rather than interpolate. Generalization is a key requirement for a general-purpose fluid simulator, which should adapt to different topologies, resolutions, or thermodynamic ranges. We propose SURF, a benchmark designed to test the generalization of learned graph-based fluid simulators. SURF comprises individual datasets and provides specific performance and generalization metrics for evaluating and comparing different models. We empirically demonstrate the applicability of SURF by thoroughly investigating the two state-of-the-art graph-based models, yielding new insights into their generalization. SURF is available under https://github.com/s-kuenzli/surf-fluidsimulation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v231-kunzli24a, title = {SURF: A Generalization Benchmark for GNNs Predicting Fluid Dynamics}, author = {K{\"u}nzli, Stefan and Gr{\"o}tschla, Florian and Mathys, Jo{\"e}l and Wattenhofer, Roger}, booktitle = {Proceedings of the Second Learning on Graphs Conference}, pages = {13:1--13:23}, year = {2024}, editor = {Villar, Soledad and Chamberlain, Benjamin}, volume = {231}, series = {Proceedings of Machine Learning Research}, month = {27--30 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v231/kunzli24a/kunzli24a.pdf}, url = {https://proceedings.mlr.press/v231/kunzli24a.html}, abstract = {Simulating fluid dynamics is crucial for the design and development process, ranging from simple valves to complex turbomachinery. Accurately solving the underlying physical equations is computationally expensive. Therefore, learning-based solvers that model interactions on meshes have gained interest due to their promising speed-ups. However, it is unknown to what extent these models truly understand the underlying physical principles and can generalize rather than interpolate. Generalization is a key requirement for a general-purpose fluid simulator, which should adapt to different topologies, resolutions, or thermodynamic ranges. We propose SURF, a benchmark designed to test the generalization of learned graph-based fluid simulators. SURF comprises individual datasets and provides specific performance and generalization metrics for evaluating and comparing different models. We empirically demonstrate the applicability of SURF by thoroughly investigating the two state-of-the-art graph-based models, yielding new insights into their generalization. SURF is available under https://github.com/s-kuenzli/surf-fluidsimulation.} }
Endnote
%0 Conference Paper %T SURF: A Generalization Benchmark for GNNs Predicting Fluid Dynamics %A Stefan Künzli %A Florian Grötschla %A Joël Mathys %A Roger Wattenhofer %B Proceedings of the Second Learning on Graphs Conference %C Proceedings of Machine Learning Research %D 2024 %E Soledad Villar %E Benjamin Chamberlain %F pmlr-v231-kunzli24a %I PMLR %P 13:1--13:23 %U https://proceedings.mlr.press/v231/kunzli24a.html %V 231 %X Simulating fluid dynamics is crucial for the design and development process, ranging from simple valves to complex turbomachinery. Accurately solving the underlying physical equations is computationally expensive. Therefore, learning-based solvers that model interactions on meshes have gained interest due to their promising speed-ups. However, it is unknown to what extent these models truly understand the underlying physical principles and can generalize rather than interpolate. Generalization is a key requirement for a general-purpose fluid simulator, which should adapt to different topologies, resolutions, or thermodynamic ranges. We propose SURF, a benchmark designed to test the generalization of learned graph-based fluid simulators. SURF comprises individual datasets and provides specific performance and generalization metrics for evaluating and comparing different models. We empirically demonstrate the applicability of SURF by thoroughly investigating the two state-of-the-art graph-based models, yielding new insights into their generalization. SURF is available under https://github.com/s-kuenzli/surf-fluidsimulation.
APA
Künzli, S., Grötschla, F., Mathys, J. & Wattenhofer, R.. (2024). SURF: A Generalization Benchmark for GNNs Predicting Fluid Dynamics. Proceedings of the Second Learning on Graphs Conference, in Proceedings of Machine Learning Research 231:13:1-13:23 Available from https://proceedings.mlr.press/v231/kunzli24a.html.

Related Material