Turbulence model augmented physics-informed neural networks for mean-flow reconstruction

Experimental measurements and numerical simulations of turbulent flows are characterized by a tradeoff between accuracy and resolution. In this study, we combine accurate sparse pointwise mean velocity measurements with the Reynolds-averaged Navier-Stokes (RANS) equations using data assimilation met...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Physical review fluids 2024-03, Vol.9 (3), Article 034605
Hauptverfasser: Patel, Yusuf, Mons, Vincent, Marquet, Olivier, Rigas, Georgios
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 Physical review fluids
container_volume 9
creator Patel, Yusuf
Mons, Vincent
Marquet, Olivier
Rigas, Georgios
description Experimental measurements and numerical simulations of turbulent flows are characterized by a tradeoff between accuracy and resolution. In this study, we combine accurate sparse pointwise mean velocity measurements with the Reynolds-averaged Navier-Stokes (RANS) equations using data assimilation methods. Importantly, we bridge the gap between data assimilation (DA) using physics-informed neural networks (PINNs) and variational methods based on classical spatial discretization of the flow equations, by comparing both approaches on the same turbulent flow case. First, by constraining the PINN with sparse data and the underdetermined RANS equations without closure, we show that the mean flow is reconstructed to a higher accuracy than a RANS solver using the Spalart-Allmaras (SA) turbulence model. Second, we propose the SA turbulence model augmented PINN (PINN-DA-SA), which outperforms the former approach by up to 73% reduction in mean velocity reconstruction error with coarse measurements. The additional SA physics constraints improve flow reconstructions in regions with high velocity and pressure gradients and separation. Third, we compare the PINN-DA-SA approach to a variational data assimilation using the same sparse velocity measurements and physics constraints. The PINN-DA-SA achieves lower reconstruction error across a range of data resolutions. This is attributed to discretization errors in the variational methodology that are avoided by PINNs. We demonstrate the method using high-fidelity measurements from direct numerical simulation of the turbulent periodic hill at Re = 5600 .
doi_str_mv 10.1103/PhysRevFluids.9.034605
format Article
fullrecord <record><control><sourceid>hal_cross</sourceid><recordid>TN_cdi_hal_primary_oai_HAL_hal_04567559v1</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>oai_HAL_hal_04567559v1</sourcerecordid><originalsourceid>FETCH-LOGICAL-c337t-f8bf66cd28ae9b885c271d771b568b3ade904da4053c15b684f0e6f2ff00f1ac3</originalsourceid><addsrcrecordid>eNpVkF1LwzAUhoMoOOb-gvTWi86TpkmbyzGcEwaKTNhdSPPhqmkzknZj_34dE9Gr9_DwnvfiQegewxRjII9v22N8N_uF62sdp3wKJGdAr9AoyxlPOYfN9Z_7Fk1i_AIAzEhR8HKENus-VL0zrTJJ47Vxiew_G9N2Rie7YbtWMa1b60MzgNb0QbohuoMP3zEZcNIY2abW-UMSjPJt7EKvutq3d-jGShfN5CfH6GPxtJ4v09Xr88t8tkoVIUWX2rKyjCmdldLwqiypygqsiwJXlJUVkdpwyLXMgRKFacXK3IJhNrMWwGKpyBg9XHa30oldqBsZjsLLWixnK3FmkFNWUMr3eOiyS1cFH2Mw9vcBgzjrFP90Ci4uOskJ9_VvHQ</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Turbulence model augmented physics-informed neural networks for mean-flow reconstruction</title><source>American Physical Society Journals</source><creator>Patel, Yusuf ; Mons, Vincent ; Marquet, Olivier ; Rigas, Georgios</creator><creatorcontrib>Patel, Yusuf ; Mons, Vincent ; Marquet, Olivier ; Rigas, Georgios</creatorcontrib><description>Experimental measurements and numerical simulations of turbulent flows are characterized by a tradeoff between accuracy and resolution. In this study, we combine accurate sparse pointwise mean velocity measurements with the Reynolds-averaged Navier-Stokes (RANS) equations using data assimilation methods. Importantly, we bridge the gap between data assimilation (DA) using physics-informed neural networks (PINNs) and variational methods based on classical spatial discretization of the flow equations, by comparing both approaches on the same turbulent flow case. First, by constraining the PINN with sparse data and the underdetermined RANS equations without closure, we show that the mean flow is reconstructed to a higher accuracy than a RANS solver using the Spalart-Allmaras (SA) turbulence model. Second, we propose the SA turbulence model augmented PINN (PINN-DA-SA), which outperforms the former approach by up to 73% reduction in mean velocity reconstruction error with coarse measurements. The additional SA physics constraints improve flow reconstructions in regions with high velocity and pressure gradients and separation. Third, we compare the PINN-DA-SA approach to a variational data assimilation using the same sparse velocity measurements and physics constraints. The PINN-DA-SA achieves lower reconstruction error across a range of data resolutions. This is attributed to discretization errors in the variational methodology that are avoided by PINNs. We demonstrate the method using high-fidelity measurements from direct numerical simulation of the turbulent periodic hill at Re = 5600 .</description><identifier>ISSN: 2469-990X</identifier><identifier>EISSN: 2469-990X</identifier><identifier>DOI: 10.1103/PhysRevFluids.9.034605</identifier><language>eng</language><publisher>American Physical Society</publisher><subject>Engineering Sciences ; Physics</subject><ispartof>Physical review fluids, 2024-03, Vol.9 (3), Article 034605</ispartof><rights>Attribution</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c337t-f8bf66cd28ae9b885c271d771b568b3ade904da4053c15b684f0e6f2ff00f1ac3</citedby><cites>FETCH-LOGICAL-c337t-f8bf66cd28ae9b885c271d771b568b3ade904da4053c15b684f0e6f2ff00f1ac3</cites><orcidid>0000-0002-9702-4724 ; 0000-0001-6692-6437</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>230,314,780,784,885,2876,2877,27924,27925</link.rule.ids><backlink>$$Uhttps://hal.science/hal-04567559$$DView record in HAL$$Hfree_for_read</backlink></links><search><creatorcontrib>Patel, Yusuf</creatorcontrib><creatorcontrib>Mons, Vincent</creatorcontrib><creatorcontrib>Marquet, Olivier</creatorcontrib><creatorcontrib>Rigas, Georgios</creatorcontrib><title>Turbulence model augmented physics-informed neural networks for mean-flow reconstruction</title><title>Physical review fluids</title><description>Experimental measurements and numerical simulations of turbulent flows are characterized by a tradeoff between accuracy and resolution. In this study, we combine accurate sparse pointwise mean velocity measurements with the Reynolds-averaged Navier-Stokes (RANS) equations using data assimilation methods. Importantly, we bridge the gap between data assimilation (DA) using physics-informed neural networks (PINNs) and variational methods based on classical spatial discretization of the flow equations, by comparing both approaches on the same turbulent flow case. First, by constraining the PINN with sparse data and the underdetermined RANS equations without closure, we show that the mean flow is reconstructed to a higher accuracy than a RANS solver using the Spalart-Allmaras (SA) turbulence model. Second, we propose the SA turbulence model augmented PINN (PINN-DA-SA), which outperforms the former approach by up to 73% reduction in mean velocity reconstruction error with coarse measurements. The additional SA physics constraints improve flow reconstructions in regions with high velocity and pressure gradients and separation. Third, we compare the PINN-DA-SA approach to a variational data assimilation using the same sparse velocity measurements and physics constraints. The PINN-DA-SA achieves lower reconstruction error across a range of data resolutions. This is attributed to discretization errors in the variational methodology that are avoided by PINNs. We demonstrate the method using high-fidelity measurements from direct numerical simulation of the turbulent periodic hill at Re = 5600 .</description><subject>Engineering Sciences</subject><subject>Physics</subject><issn>2469-990X</issn><issn>2469-990X</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><recordid>eNpVkF1LwzAUhoMoOOb-gvTWi86TpkmbyzGcEwaKTNhdSPPhqmkzknZj_34dE9Gr9_DwnvfiQegewxRjII9v22N8N_uF62sdp3wKJGdAr9AoyxlPOYfN9Z_7Fk1i_AIAzEhR8HKENus-VL0zrTJJ47Vxiew_G9N2Rie7YbtWMa1b60MzgNb0QbohuoMP3zEZcNIY2abW-UMSjPJt7EKvutq3d-jGShfN5CfH6GPxtJ4v09Xr88t8tkoVIUWX2rKyjCmdldLwqiypygqsiwJXlJUVkdpwyLXMgRKFacXK3IJhNrMWwGKpyBg9XHa30oldqBsZjsLLWixnK3FmkFNWUMr3eOiyS1cFH2Mw9vcBgzjrFP90Ci4uOskJ9_VvHQ</recordid><startdate>20240311</startdate><enddate>20240311</enddate><creator>Patel, Yusuf</creator><creator>Mons, Vincent</creator><creator>Marquet, Olivier</creator><creator>Rigas, Georgios</creator><general>American Physical Society</general><scope>AAYXX</scope><scope>CITATION</scope><scope>1XC</scope><scope>VOOES</scope><orcidid>https://orcid.org/0000-0002-9702-4724</orcidid><orcidid>https://orcid.org/0000-0001-6692-6437</orcidid></search><sort><creationdate>20240311</creationdate><title>Turbulence model augmented physics-informed neural networks for mean-flow reconstruction</title><author>Patel, Yusuf ; Mons, Vincent ; Marquet, Olivier ; Rigas, Georgios</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c337t-f8bf66cd28ae9b885c271d771b568b3ade904da4053c15b684f0e6f2ff00f1ac3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Engineering Sciences</topic><topic>Physics</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Patel, Yusuf</creatorcontrib><creatorcontrib>Mons, Vincent</creatorcontrib><creatorcontrib>Marquet, Olivier</creatorcontrib><creatorcontrib>Rigas, Georgios</creatorcontrib><collection>CrossRef</collection><collection>Hyper Article en Ligne (HAL)</collection><collection>Hyper Article en Ligne (HAL) (Open Access)</collection><jtitle>Physical review fluids</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Patel, Yusuf</au><au>Mons, Vincent</au><au>Marquet, Olivier</au><au>Rigas, Georgios</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Turbulence model augmented physics-informed neural networks for mean-flow reconstruction</atitle><jtitle>Physical review fluids</jtitle><date>2024-03-11</date><risdate>2024</risdate><volume>9</volume><issue>3</issue><artnum>034605</artnum><issn>2469-990X</issn><eissn>2469-990X</eissn><abstract>Experimental measurements and numerical simulations of turbulent flows are characterized by a tradeoff between accuracy and resolution. In this study, we combine accurate sparse pointwise mean velocity measurements with the Reynolds-averaged Navier-Stokes (RANS) equations using data assimilation methods. Importantly, we bridge the gap between data assimilation (DA) using physics-informed neural networks (PINNs) and variational methods based on classical spatial discretization of the flow equations, by comparing both approaches on the same turbulent flow case. First, by constraining the PINN with sparse data and the underdetermined RANS equations without closure, we show that the mean flow is reconstructed to a higher accuracy than a RANS solver using the Spalart-Allmaras (SA) turbulence model. Second, we propose the SA turbulence model augmented PINN (PINN-DA-SA), which outperforms the former approach by up to 73% reduction in mean velocity reconstruction error with coarse measurements. The additional SA physics constraints improve flow reconstructions in regions with high velocity and pressure gradients and separation. Third, we compare the PINN-DA-SA approach to a variational data assimilation using the same sparse velocity measurements and physics constraints. The PINN-DA-SA achieves lower reconstruction error across a range of data resolutions. This is attributed to discretization errors in the variational methodology that are avoided by PINNs. We demonstrate the method using high-fidelity measurements from direct numerical simulation of the turbulent periodic hill at Re = 5600 .</abstract><pub>American Physical Society</pub><doi>10.1103/PhysRevFluids.9.034605</doi><orcidid>https://orcid.org/0000-0002-9702-4724</orcidid><orcidid>https://orcid.org/0000-0001-6692-6437</orcidid><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 2469-990X
ispartof Physical review fluids, 2024-03, Vol.9 (3), Article 034605
issn 2469-990X
2469-990X
language eng
recordid cdi_hal_primary_oai_HAL_hal_04567559v1
source American Physical Society Journals
subjects Engineering Sciences
Physics
title Turbulence model augmented physics-informed neural networks for mean-flow reconstruction
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T23%3A17%3A15IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-hal_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Turbulence%20model%20augmented%20physics-informed%20neural%20networks%20for%20mean-flow%20reconstruction&rft.jtitle=Physical%20review%20fluids&rft.au=Patel,%20Yusuf&rft.date=2024-03-11&rft.volume=9&rft.issue=3&rft.artnum=034605&rft.issn=2469-990X&rft.eissn=2469-990X&rft_id=info:doi/10.1103/PhysRevFluids.9.034605&rft_dat=%3Chal_cross%3Eoai_HAL_hal_04567559v1%3C/hal_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true