A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations
The computational cost of computational fluid dynamics (CFD) simulation is relatively high due to its computational complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient time advancement methods, correction methods for discrete contr...
Gespeichert in:
Veröffentlicht in: | Physics of fluids (1994) 2024-03, Vol.36 (3) |
---|---|
Hauptverfasser: | , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | |
---|---|
container_issue | 3 |
container_start_page | |
container_title | Physics of fluids (1994) |
container_volume | 36 |
creator | Zhang, Xiaoyuan Sun, Guopeng Zhang, Peng Wang, Yueqing Zhang, Jian Deng, Liang Lin, Jie Chen, Jianqiang |
description | The computational cost of computational fluid dynamics (CFD) simulation is relatively high due to its computational complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient time advancement methods, correction methods for discrete control equations, multigrid methods, reasonable initial field setting methods, and parallel methods. Among these methods, the initial field setting method can provide significant performance improvements, but there is little work on it. Existing CFD industrial software typically uses inflow conditions for the initial flow field or applies empirical methods, which can cause instability in the CFD calculation process and make convergence difficult. With the rapid development of deep learning, researchers are increasingly attempting to replace CFD simulations with deep neural networks and have achieved significant performance improvements. However, these methods still face some challenges. First, they can only predict the computational flow field on regular grids. They cannot directly make predictions for irregular grids such as multi-block grids and unstructured grids, so the final flow field can only be obtained through interpolation and similar methods. Second, although these methods have been claimed to provide high accuracy, there is still a significant gap in performance with CFD and they cannot yet be applied to real scenarios. To address these issues, we propose a Residual Graph Convolutional Network for Initial Flow Field Setting (RGCN-IFS) in CFD simulations. This method converts the grid into a graph structure and uses an improved graph neural network to predict the flow field. In this way, we can predict the flow field on any type of grid. More importantly, this method does not directly replace CFD simulations, but it rather serves in an auxiliary role, providing appropriate initial flow fields for the CFD calculations, improving the convergence efficiency while ensuring calculation accuracy, and directly bridging the accuracy gap between intelligent surrogate models and CFD simulations. |
doi_str_mv | 10.1063/5.0195824 |
format | Article |
fullrecord | <record><control><sourceid>proquest_scita</sourceid><recordid>TN_cdi_scitation_primary_10_1063_5_0195824</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2979202253</sourcerecordid><originalsourceid>FETCH-LOGICAL-c287t-3024b3172ed58095a99776e59f24c877dae82253b6515e95208482f0b8691cf43</originalsourceid><addsrcrecordid>eNp9kMtOwzAQRS0EEqWw4A8ssQIpxXbi17KqeEmV2MA6chO7uCRx6gdV_56EdM1qRnPPHc1cAG4xWmDE8ke6QFhSQYozMMNIyIwzxs7HnqOMsRxfgqsQdgihXBI2A_sl9DrYOqkGbr3qv2Dluh_XpGhdN8w6HQ_Of0PjPAw6Rtttoe1stINmGneAxuqmHkaDr-1TVCefaZKtYX3sVGurAINtU_OnhWtwYVQT9M2pzsHn89PH6jVbv7-8rZbrrCKCxyxHpNjkmBNdU4EkVVJyzjSVhhSV4LxWWhBC8w2jmGpJCRKFIAZtBJO4MkU-B3fT3t67fdIhljuX_HBbKInkkqDRPVD3E1V5F4LXpuy9bZU_lhiVY6IlLU-JDuzDxIbKTo_-A_8C-7V2Vg</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2979202253</pqid></control><display><type>article</type><title>A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations</title><source>AIP Journals Complete</source><creator>Zhang, Xiaoyuan ; Sun, Guopeng ; Zhang, Peng ; Wang, Yueqing ; Zhang, Jian ; Deng, Liang ; Lin, Jie ; Chen, Jianqiang</creator><creatorcontrib>Zhang, Xiaoyuan ; Sun, Guopeng ; Zhang, Peng ; Wang, Yueqing ; Zhang, Jian ; Deng, Liang ; Lin, Jie ; Chen, Jianqiang</creatorcontrib><description>The computational cost of computational fluid dynamics (CFD) simulation is relatively high due to its computational complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient time advancement methods, correction methods for discrete control equations, multigrid methods, reasonable initial field setting methods, and parallel methods. Among these methods, the initial field setting method can provide significant performance improvements, but there is little work on it. Existing CFD industrial software typically uses inflow conditions for the initial flow field or applies empirical methods, which can cause instability in the CFD calculation process and make convergence difficult. With the rapid development of deep learning, researchers are increasingly attempting to replace CFD simulations with deep neural networks and have achieved significant performance improvements. However, these methods still face some challenges. First, they can only predict the computational flow field on regular grids. They cannot directly make predictions for irregular grids such as multi-block grids and unstructured grids, so the final flow field can only be obtained through interpolation and similar methods. Second, although these methods have been claimed to provide high accuracy, there is still a significant gap in performance with CFD and they cannot yet be applied to real scenarios. To address these issues, we propose a Residual Graph Convolutional Network for Initial Flow Field Setting (RGCN-IFS) in CFD simulations. This method converts the grid into a graph structure and uses an improved graph neural network to predict the flow field. In this way, we can predict the flow field on any type of grid. More importantly, this method does not directly replace CFD simulations, but it rather serves in an auxiliary role, providing appropriate initial flow fields for the CFD calculations, improving the convergence efficiency while ensuring calculation accuracy, and directly bridging the accuracy gap between intelligent surrogate models and CFD simulations.</description><identifier>ISSN: 1070-6631</identifier><identifier>EISSN: 1089-7666</identifier><identifier>DOI: 10.1063/5.0195824</identifier><identifier>CODEN: PHFLE6</identifier><language>eng</language><publisher>Melville: American Institute of Physics</publisher><subject>Accuracy ; Artificial neural networks ; Computational efficiency ; Computational fluid dynamics ; Computing costs ; Computing time ; Convergence ; Flow stability ; Fluid dynamics ; Graph neural networks ; Interpolation ; Machine learning ; Mathematical models ; Multiblock grids ; Multigrid methods ; Neural networks ; Simulation ; Unstructured grids (mathematics)</subject><ispartof>Physics of fluids (1994), 2024-03, Vol.36 (3)</ispartof><rights>Author(s)</rights><rights>2024 Author(s). Published under an exclusive license by AIP Publishing.</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><cites>FETCH-LOGICAL-c287t-3024b3172ed58095a99776e59f24c877dae82253b6515e95208482f0b8691cf43</cites><orcidid>0009-0007-3684-4079 ; 0000-0002-9301-2355 ; 0009-0008-2858-5465 ; 0000-0002-7676-7609 ; 0000-0002-8734-8346 ; 0000-0003-1444-4588</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>314,776,780,790,4497,27903,27904</link.rule.ids></links><search><creatorcontrib>Zhang, Xiaoyuan</creatorcontrib><creatorcontrib>Sun, Guopeng</creatorcontrib><creatorcontrib>Zhang, Peng</creatorcontrib><creatorcontrib>Wang, Yueqing</creatorcontrib><creatorcontrib>Zhang, Jian</creatorcontrib><creatorcontrib>Deng, Liang</creatorcontrib><creatorcontrib>Lin, Jie</creatorcontrib><creatorcontrib>Chen, Jianqiang</creatorcontrib><title>A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations</title><title>Physics of fluids (1994)</title><description>The computational cost of computational fluid dynamics (CFD) simulation is relatively high due to its computational complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient time advancement methods, correction methods for discrete control equations, multigrid methods, reasonable initial field setting methods, and parallel methods. Among these methods, the initial field setting method can provide significant performance improvements, but there is little work on it. Existing CFD industrial software typically uses inflow conditions for the initial flow field or applies empirical methods, which can cause instability in the CFD calculation process and make convergence difficult. With the rapid development of deep learning, researchers are increasingly attempting to replace CFD simulations with deep neural networks and have achieved significant performance improvements. However, these methods still face some challenges. First, they can only predict the computational flow field on regular grids. They cannot directly make predictions for irregular grids such as multi-block grids and unstructured grids, so the final flow field can only be obtained through interpolation and similar methods. Second, although these methods have been claimed to provide high accuracy, there is still a significant gap in performance with CFD and they cannot yet be applied to real scenarios. To address these issues, we propose a Residual Graph Convolutional Network for Initial Flow Field Setting (RGCN-IFS) in CFD simulations. This method converts the grid into a graph structure and uses an improved graph neural network to predict the flow field. In this way, we can predict the flow field on any type of grid. More importantly, this method does not directly replace CFD simulations, but it rather serves in an auxiliary role, providing appropriate initial flow fields for the CFD calculations, improving the convergence efficiency while ensuring calculation accuracy, and directly bridging the accuracy gap between intelligent surrogate models and CFD simulations.</description><subject>Accuracy</subject><subject>Artificial neural networks</subject><subject>Computational efficiency</subject><subject>Computational fluid dynamics</subject><subject>Computing costs</subject><subject>Computing time</subject><subject>Convergence</subject><subject>Flow stability</subject><subject>Fluid dynamics</subject><subject>Graph neural networks</subject><subject>Interpolation</subject><subject>Machine learning</subject><subject>Mathematical models</subject><subject>Multiblock grids</subject><subject>Multigrid methods</subject><subject>Neural networks</subject><subject>Simulation</subject><subject>Unstructured grids (mathematics)</subject><issn>1070-6631</issn><issn>1089-7666</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><recordid>eNp9kMtOwzAQRS0EEqWw4A8ssQIpxXbi17KqeEmV2MA6chO7uCRx6gdV_56EdM1qRnPPHc1cAG4xWmDE8ke6QFhSQYozMMNIyIwzxs7HnqOMsRxfgqsQdgihXBI2A_sl9DrYOqkGbr3qv2Dluh_XpGhdN8w6HQ_Of0PjPAw6Rtttoe1stINmGneAxuqmHkaDr-1TVCefaZKtYX3sVGurAINtU_OnhWtwYVQT9M2pzsHn89PH6jVbv7-8rZbrrCKCxyxHpNjkmBNdU4EkVVJyzjSVhhSV4LxWWhBC8w2jmGpJCRKFIAZtBJO4MkU-B3fT3t67fdIhljuX_HBbKInkkqDRPVD3E1V5F4LXpuy9bZU_lhiVY6IlLU-JDuzDxIbKTo_-A_8C-7V2Vg</recordid><startdate>202403</startdate><enddate>202403</enddate><creator>Zhang, Xiaoyuan</creator><creator>Sun, Guopeng</creator><creator>Zhang, Peng</creator><creator>Wang, Yueqing</creator><creator>Zhang, Jian</creator><creator>Deng, Liang</creator><creator>Lin, Jie</creator><creator>Chen, Jianqiang</creator><general>American Institute of Physics</general><scope>AAYXX</scope><scope>CITATION</scope><scope>8FD</scope><scope>H8D</scope><scope>L7M</scope><orcidid>https://orcid.org/0009-0007-3684-4079</orcidid><orcidid>https://orcid.org/0000-0002-9301-2355</orcidid><orcidid>https://orcid.org/0009-0008-2858-5465</orcidid><orcidid>https://orcid.org/0000-0002-7676-7609</orcidid><orcidid>https://orcid.org/0000-0002-8734-8346</orcidid><orcidid>https://orcid.org/0000-0003-1444-4588</orcidid></search><sort><creationdate>202403</creationdate><title>A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations</title><author>Zhang, Xiaoyuan ; Sun, Guopeng ; Zhang, Peng ; Wang, Yueqing ; Zhang, Jian ; Deng, Liang ; Lin, Jie ; Chen, Jianqiang</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c287t-3024b3172ed58095a99776e59f24c877dae82253b6515e95208482f0b8691cf43</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Accuracy</topic><topic>Artificial neural networks</topic><topic>Computational efficiency</topic><topic>Computational fluid dynamics</topic><topic>Computing costs</topic><topic>Computing time</topic><topic>Convergence</topic><topic>Flow stability</topic><topic>Fluid dynamics</topic><topic>Graph neural networks</topic><topic>Interpolation</topic><topic>Machine learning</topic><topic>Mathematical models</topic><topic>Multiblock grids</topic><topic>Multigrid methods</topic><topic>Neural networks</topic><topic>Simulation</topic><topic>Unstructured grids (mathematics)</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Zhang, Xiaoyuan</creatorcontrib><creatorcontrib>Sun, Guopeng</creatorcontrib><creatorcontrib>Zhang, Peng</creatorcontrib><creatorcontrib>Wang, Yueqing</creatorcontrib><creatorcontrib>Zhang, Jian</creatorcontrib><creatorcontrib>Deng, Liang</creatorcontrib><creatorcontrib>Lin, Jie</creatorcontrib><creatorcontrib>Chen, Jianqiang</creatorcontrib><collection>CrossRef</collection><collection>Technology Research Database</collection><collection>Aerospace Database</collection><collection>Advanced Technologies Database with Aerospace</collection><jtitle>Physics of fluids (1994)</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Zhang, Xiaoyuan</au><au>Sun, Guopeng</au><au>Zhang, Peng</au><au>Wang, Yueqing</au><au>Zhang, Jian</au><au>Deng, Liang</au><au>Lin, Jie</au><au>Chen, Jianqiang</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations</atitle><jtitle>Physics of fluids (1994)</jtitle><date>2024-03</date><risdate>2024</risdate><volume>36</volume><issue>3</issue><issn>1070-6631</issn><eissn>1089-7666</eissn><coden>PHFLE6</coden><abstract>The computational cost of computational fluid dynamics (CFD) simulation is relatively high due to its computational complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient time advancement methods, correction methods for discrete control equations, multigrid methods, reasonable initial field setting methods, and parallel methods. Among these methods, the initial field setting method can provide significant performance improvements, but there is little work on it. Existing CFD industrial software typically uses inflow conditions for the initial flow field or applies empirical methods, which can cause instability in the CFD calculation process and make convergence difficult. With the rapid development of deep learning, researchers are increasingly attempting to replace CFD simulations with deep neural networks and have achieved significant performance improvements. However, these methods still face some challenges. First, they can only predict the computational flow field on regular grids. They cannot directly make predictions for irregular grids such as multi-block grids and unstructured grids, so the final flow field can only be obtained through interpolation and similar methods. Second, although these methods have been claimed to provide high accuracy, there is still a significant gap in performance with CFD and they cannot yet be applied to real scenarios. To address these issues, we propose a Residual Graph Convolutional Network for Initial Flow Field Setting (RGCN-IFS) in CFD simulations. This method converts the grid into a graph structure and uses an improved graph neural network to predict the flow field. In this way, we can predict the flow field on any type of grid. More importantly, this method does not directly replace CFD simulations, but it rather serves in an auxiliary role, providing appropriate initial flow fields for the CFD calculations, improving the convergence efficiency while ensuring calculation accuracy, and directly bridging the accuracy gap between intelligent surrogate models and CFD simulations.</abstract><cop>Melville</cop><pub>American Institute of Physics</pub><doi>10.1063/5.0195824</doi><tpages>14</tpages><orcidid>https://orcid.org/0009-0007-3684-4079</orcidid><orcidid>https://orcid.org/0000-0002-9301-2355</orcidid><orcidid>https://orcid.org/0009-0008-2858-5465</orcidid><orcidid>https://orcid.org/0000-0002-7676-7609</orcidid><orcidid>https://orcid.org/0000-0002-8734-8346</orcidid><orcidid>https://orcid.org/0000-0003-1444-4588</orcidid><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | ISSN: 1070-6631 |
ispartof | Physics of fluids (1994), 2024-03, Vol.36 (3) |
issn | 1070-6631 1089-7666 |
language | eng |
recordid | cdi_scitation_primary_10_1063_5_0195824 |
source | AIP Journals Complete |
subjects | Accuracy Artificial neural networks Computational efficiency Computational fluid dynamics Computing costs Computing time Convergence Flow stability Fluid dynamics Graph neural networks Interpolation Machine learning Mathematical models Multiblock grids Multigrid methods Neural networks Simulation Unstructured grids (mathematics) |
title | A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-24T22%3A53%3A02IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_scita&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=A%20residual%20graph%20convolutional%20network%20for%20setting%20initial%20flow%20field%20in%20computational%20fluid%20dynamics%20simulations&rft.jtitle=Physics%20of%20fluids%20(1994)&rft.au=Zhang,%20Xiaoyuan&rft.date=2024-03&rft.volume=36&rft.issue=3&rft.issn=1070-6631&rft.eissn=1089-7666&rft.coden=PHFLE6&rft_id=info:doi/10.1063/5.0195824&rft_dat=%3Cproquest_scita%3E2979202253%3C/proquest_scita%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2979202253&rft_id=info:pmid/&rfr_iscdi=true |