Computational Technology Resources - CCC

M.J. Kong¹, J.W. Jeoung¹, S.K. Yeom¹, D.H. Jung¹, T.H. Hong¹, K.B. Jeong² and Y.W. Hwang²

¹Yonsei University, Department of Architecture and Architectural Engineering, Seoul, South Korea
²Sejong University, Deep Learning Architecture Research Center, Department of Architectural Engineering, Seoul, South Korea

doi:10.4203/ccc.13.1.1

M.J. Kong, J.W. Jeoung, S.K. Yeom, D.H. Jung, T.H. Hong, K.B. Jeong, Y.W. Hwang, "Propagation-Gated Graph Neural Operators for Steady-State CFD Surrogate Modelling on Unstructured Meshes", in P. Iványi, J. Kruis, B.H.V. Topping, (Editors), "Proceedings of the Thirteenth International Conference on Engineering Computational Technology", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 13, Paper 1.1, 2026, doi:10.4203/ccc.13.1.1

Keywords: computational fluid dynamics, graph neural operator, surrogate model, flow field reconstruction, sparse measurement, propagation gate.

Abstract

This study proposes a propagation-gated graph neural operator (PG-GNO) for steady-state computational fluid dynamics (CFD) surrogate modeling on unstructured meshes. The approach is motivated by the observation that steady-state CFD solutions are typically obtained through iterative fixed-point solvers converging to a stationary flow field. Instead of directly predicting the solution in a single step, PG-GNO explicitly mimics this iterative convergence process through recurrent neural operator updates. Unstructured CFD meshes are represented as graphs, and a graph neural operator is employed to achieve discretization invariance, enabling flexible application across different mesh resolutions and sensor configurations. To reconstruct global flow fields from limited inlet and sensor information, a propagation gate mechanism is introduced to regulate information transmission across the graph. This mechanism allows inlet and sensor information to be gradually propagated throughout the computational domain while suppressing contributions from uninformed regions. Validation experiments conducted on untrained inlet boundary conditions and sensor placements demonstrate that the proposed model achieves prediction accuracy comparable to that of existing CFD surrogate models. Overall, the PG-GNO framework offers a practical and generalizable surrogate modeling approach for steady-state CFD problems involving unstructured meshes and realistic, sparse measurements.

download the full-text of this paper (PDF, 14 pages, 1224 Kb)

go to the next paper
return to the table of contents
return to the volume description

	Computational & Technology Resources an online resource for computational, engineering & technology publications
	not logged in - login
Front Page Browse CCC CCP CSETS CTR IJRT Other Authors Search Purchase Guide FAQ Contact us	Civil-Comp Conferences ISSN 2753-3239 CCC: 13 PROCEEDINGS OF THE THIRTEENTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY Edited by: P. Iványi, J. Kruis and B.H.V. Topping Paper 1.1 Propagation-Gated Graph Neural Operators for Steady-State CFD Surrogate Modelling on Unstructured Meshes M.J. Kong¹, J.W. Jeoung¹, S.K. Yeom¹, D.H. Jung¹, T.H. Hong¹, K.B. Jeong² and Y.W. Hwang² ¹Yonsei University, Department of Architecture and Architectural Engineering, Seoul, South Korea ²Sejong University, Deep Learning Architecture Research Center, Department of Architectural Engineering, Seoul, South Korea doi:10.4203/ccc.13.1.1 Full Bibliographic Reference for this paper M.J. Kong, J.W. Jeoung, S.K. Yeom, D.H. Jung, T.H. Hong, K.B. Jeong, Y.W. Hwang, "Propagation-Gated Graph Neural Operators for Steady-State CFD Surrogate Modelling on Unstructured Meshes", in P. Iványi, J. Kruis, B.H.V. Topping, (Editors), "Proceedings of the Thirteenth International Conference on Engineering Computational Technology", Civil-Comp Press, Edinburgh, UK, Online volume: CCC 13, Paper 1.1, 2026, doi:10.4203/ccc.13.1.1 Keywords: computational fluid dynamics, graph neural operator, surrogate model, flow field reconstruction, sparse measurement, propagation gate. Abstract This study proposes a propagation-gated graph neural operator (PG-GNO) for steady-state computational fluid dynamics (CFD) surrogate modeling on unstructured meshes. The approach is motivated by the observation that steady-state CFD solutions are typically obtained through iterative fixed-point solvers converging to a stationary flow field. Instead of directly predicting the solution in a single step, PG-GNO explicitly mimics this iterative convergence process through recurrent neural operator updates. Unstructured CFD meshes are represented as graphs, and a graph neural operator is employed to achieve discretization invariance, enabling flexible application across different mesh resolutions and sensor configurations. To reconstruct global flow fields from limited inlet and sensor information, a propagation gate mechanism is introduced to regulate information transmission across the graph. This mechanism allows inlet and sensor information to be gradually propagated throughout the computational domain while suppressing contributions from uninformed regions. Validation experiments conducted on untrained inlet boundary conditions and sensor placements demonstrate that the proposed model achieves prediction accuracy comparable to that of existing CFD surrogate models. Overall, the PG-GNO framework offers a practical and generalizable surrogate modeling approach for steady-state CFD problems involving unstructured meshes and realistic, sparse measurements. download the full-text of this paper (PDF, 14 pages, 1224 Kb) go to the next paper return to the table of contents return to the volume description
Back to top	©Civil-Comp Limited 2023 - terms & conditions