AeroDML

Graph neural networks for CFD.

Version 0.1

Distribution Statement A: Approved for public release; distribution is unlimited. PA# AFRL-2026-3662.

AeroDML, for Aerodynamic Deep Multifidelity Learning, predicts three-dimensional aerodynamic flow fields with graph neural networks trained on data from multiple fidelity levels. The repository includes RANS-GNN models, training workflows and data-generation tools. It accompanies the scientific machine learning work in [1], which was supported by the AFRL collaborative center.

Spanwise pressure slices of an optimized blended-wing-body geometry compare CFD, RANS-GNN prediction, and absolute error at five span stations. — Spanwise pressure slices of an optimized geometry compare CFD with the RANS-GNN prediction and its absolute error. [1]

References

Mark Sperry, John T. Hwang. Multifidelity Surrogate Modeling for 3D Aerodynamic Flow Field Prediction Using Graph Neural Networks. AIAA AVIATION 2026 Forum, 2026. Distribution Statement A: Approved for public release; distribution is unlimited. PA# AFRL-2026-2231

DOI PDF

@inproceedings{sperry2026multifidelity,
  author = {Mark Sperry and John T. Hwang},
  title = {Multifidelity Surrogate Modeling for 3D Aerodynamic Flow Field Prediction Using Graph Neural Networks},
  booktitle = {AIAA AVIATION 2026 Forum},
  year = {2026},
  doi = {10.2514/6.2026-4802},
  url = {https://doi.org/10.2514/6.2026-4802},
  note = {Metadata verification: Crossref title}
}

AeroDML

References

Related to AeroDML

Methods

Authors of related work