{ "@context": { "@base": "https://portfoliodb.hslu.ch/", "madek": "https://portfoliodb.hslu.ch/ns#", "madek_system": "https://portfoliodb.hslu.ch/vocabulary/madek_system:", "Keyword": "https://portfoliodb.hslu.ch/vocabulary/keyword/", "Role": "https://portfoliodb.hslu.ch/roles/", "rdf": "http://www.w3.org/1999/02/22-rdf-syntax-ns#", "rdfs": "http://www.w3.org/2000/01/rdf-schema#", "owl": "http://www.w3.org/2002/07/owl#", "madek_core": "https://portfoliodb.hslu.ch/vocabulary/madek_core:", "madek_institution": "https://portfoliodb.hslu.ch/vocabulary/institution:", "madek_creative_work": "https://portfoliodb.hslu.ch/vocabulary/creative_work:", "madek_rights": "https://portfoliodb.hslu.ch/vocabulary/rights:" }, "@graph": [ { "@id": "https://portfoliodb.hslu.ch/entries/b07fa7c0-68c2-45bf-9ba9-5ffdf12f1fd3", "@type": "madek:MediaEntry", "madek_creative_work:format": { "@value": "119 Seiten", "@type": "madek:Text" }, "madek_creative_work:language": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/c6e7b266-1f09-4c82-88de-8542130ed40d" }, "madek_creative_work:location": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/905b81cf-659f-46f9-9a7b-a0679f0522d2" }, "madek_rights:lizenzierung": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/f9e075c7-4438-40ae-a720-38ee90913d26" }, "madek_rights:classification": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/1cc7a5b5-85bd-405f-9fc4-af99c6f880fd" }, "madek_rights:web_statement": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/b6725459-379b-4b85-b8cf-273fbbeb9265" }, "madek_core:authors": { "@id": "https://portfoliodb.hslu.ch/people/215b84d8-ec03-402b-840d-10c1a03de580" }, "madek_institution:project_leader": { "@id": "https://portfoliodb.hslu.ch/people/8705486c-f42f-4981-9980-57a112c9d41b" }, "madek_core:portrayed_object_date": { "@value": "28.02.2021", "@type": "madek:TextDate" }, "madek_institution:project_type": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/50110184-dc50-4f29-b456-e81e5e640598" }, "madek_core:description": { "@value": "Data-driven Modellierungsmethoden erhalten einen immer höheren Stellenwert in der Wissenschaft und der\r\nIndustrie. Computational Fluid Dynamics erweist sich als ein exemplarisches Gebiet, um solche Methoden zu\r\nentwickeln und anzutreiben. Im Zentrum dieser Thesis steht der spezifische Prozess Field Inversion and\r\nMachine Learning (FIML), entwickelt von Parish und Duraisamy [1], Duraisamy et al. [2, 3] und Singh et al.\r\n[4]. FIML strebt an, bestehende RANS Turbulenzmodelle mithilfe von Messdaten oder hoch qualitativen\r\nSimulationsdaten zu optimieren.\r\nDie Feld Inversion wird mittels einer Zielfunktion formuliert, welche die Differenz zwischen dem Base\r\nModel und dem Truth Model ausdrückt. Zusätzlich wird die Zielfunktion durch einen Tikhonov Term\r\nregularisiert. Diese Funktion wird anschliessend mit einem Optimierungsalgorithmus und der diskreten\r\nAdjoint Methode iterativ minimiert, um ein Diskrepanzfeld zu erhalten, welches den Unterschied zwischen\r\ndem Base und Truth Model wieder gibt. Ein Machine Learning Algorithmus lernt anschliessend die\r\nBeziehung zwischen dem Diskrepanzfeld und selektierten Features des Base Models. Das trainierte Model\r\nwird in einem finalen Schritt an das Base Turbulenzmodel gekoppelt, welches somit das optimierte Model\r\ndarstellt.\r\nZiel dieser Arbeit ist es, den FIML Prozess innerhalb des Druck-gekoppelten CFD Solvers coupledNumerics\r\n[5] anzuwenden. Mit dem proof of concept wird der Prozess an einem U-Rohr validiert. Dies, indem das\r\nSpalart-Allmaras Turbulenzmodell (Base Model) optimiert wird, um die gleichen Resultate wie das Spalart-\r\nAllmaras Model mit Rotations- und Krümmungskorrektur (Truth Model) zu erzielen. In einem zweiten\r\nSchritt wird die Generalisierungsfähigkeit des FIML Verfahrens auf verschiedene Reynolds Zahlen und\r\nRohrgeometrien getestet. Als letztes wird der Prozess an wenig vorhandenen realen Messdaten angewendet.\r\nHierfür wird der zwei dimensionale Hump als Geometrie verwendet, welcher einen hohen Druckgradienten\r\nerzeugt.\r\nDer proof of concept zeigt, dass der FIML Prozess fähig ist, das Diskrepanzfeld zu extrahieren und\r\nanschliessend die inherente Beziehung zwischen diesem Feld und den Features in neuen Simulation\r\nwiederzugeben. Der Random Forest Algorithmus erzeugte konsistentere Resultate mit weniger Aufwand im\r\nVergleich zu einem simplen Neuralen Netzwerk. Die Genauigkeit des Machine Learning Models basiert auf mehreren Faktoren, reduziert aber in jedem Fall die Genauigkeit des Diskrepanzfeldes von der\r\nvorhergehenden Feld Inversion. Der Generalisierungstest zeigt, dass das optimierte Turbulenzmodell auf\r\nunterschiedlichen Reynolds Zahlen und Geometrien anwendbar ist, solange diese genügende Ähnlichkeiten\r\naufweisen. Die Verwendung von gemessenen Reibungskoeffizienten für die Feld Inversion zeigte, dass das\r\nResultat einen under- oder overfit zu den Truth Daten erzeugen kann.\r\n\r\nData-driven modelling has gained momentum in science and engineering and computational fluid dynamics\r\nis an exemplary field to explore this approach. Central to this thesis is the specific framework of Field\r\nInversion and Machine Learning (FIML) developed by Parish and Duraisamy [1], Duraisamy et al. [2, 3] and\r\nSingh et al. [4] to improve existing RANS turbulence models using measurement or high-fidelity simulation\r\ndata.\r\nKey of the paradigm is the field inversion, where an objective function is formulated in terms of the\r\ndifference between the base model and the truth model. This thesis uses a formulation with Tikhonov\r\nregularization. The objective function is then iteratively minimized using an optimization algorithm with a\r\ndiscrete adjoint method to extract the spatial discrepancy of the model. A machine learning algorithm is\r\nemployed to learn the relationship between this discrepancy field and selected features of the base model.\r\nThe machine learning model is linked to the base model, capable of conducting simulations resembling the\r\ntruth model.\r\nThe goal of this thesis is to apply the proposed FIML framework within the pressure-coupled solver\r\ncoupledNumerics [5]. The proof of concept aims to validate the FIML procedure: The Spalart-Allmaras\r\nturbulence model (base model) is optimized to equal the Spalart-Allmaras rotation and curvature correction\r\nmodel (truth model) using a U-turn pipe as geometry. In a second step, the generalization capabilities of the\r\nFIML procedure are verified by training and predicting on varying Reynolds numbers and pipe geometries. In\r\na final step, the FIML paradigm is applied to an adverse pressure gradient case, using the 2-dimensional wall\r\nmounted hump geometry and sparse experimental values only as truth model.\r\nThe proof of concept shows that the FIML paradigm is capable of inferring the discrepancy field via an\r\ninverse problem using full-field data and that machine learning tools are able to recreate this field for new\r\nsimulations. The random forest algorithm showed more consistent results with less effort than a simple neural\r\nnetwork implementation. The accuracy of the machine learning model depends on many factors such as the\r\nselected hyperparameters and features but can never perfectly reproduce the field inversion results,\r\ndiminishing therefore the overall accuracy of FIML. The generalization test proofs that the paradigm is\r\ncapable of being applied to different flows and geometries, as long as they are similar. Using the friction\r\ncoefficient as sparse data to inform the objective function for the hump, it is found that difficulties arise to\r\nobtain a result which neither under- nor overfits the truth model data during the field inversion.", "@type": "madek:Text" }, "madek_core:copyright_notice": { "@value": "Kiener Anna, Hochschule Luzern - Departement Technik & Architektur", "@type": "madek:Text" }, "madek_institution:project-expert": { "@id": "https://portfoliodb.hslu.ch/people/a6fde3ee-1bba-4e63-95b7-b085d9e3c704" }, "madek_institution:field_of_department": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/e02dca77-c61e-48af-be55-0bbb8a6d916a" }, "madek_institution:field_of_study": { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/64d4a247-1d84-4fcd-a089-a3bbf7447c95" }, "madek_core:title": { "@value": "Improving Turbulence Models in RANS Simulations with Adjoint Method Field Inversion and Machine Learning Approaches", "@type": "madek:Text" } }, { "@type": "madek:MetaKey", "@id": "madek_creative_work:format", "rdfs:label": [ { "@language": "de", "@value": "Format" }, { "@language": "en", "@value": "Format" } ], "rdfs:comment": [ { "@language": "de", "@value": "z.B. 237 x 170 cm" }, { "@language": "en", "@value": "E.g. 237 x 170 cm" } ] }, { "@type": "madek:MetaKey", "@id": "madek_creative_work:language", "rdfs:label": [ { "@language": "de", "@value": "Sprache" }, { "@language": "en", "@value": "Language" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_creative_work:location", "rdfs:label": [ { "@language": "de", "@value": "Ort" }, { "@language": "en", "@value": "Place" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_rights:lizenzierung", "rdfs:label": [ { "@language": "de", "@value": "Lizenzierung" }, { "@language": "en", "@value": "Licence" } ], "rdfs:comment": [ { "@language": "de", "@value": "Wird bei einer Veröffentlichung durch die HSLU bestätigt. Verwendung der Arbeit: Namensnennung – Nicht-kommerziell – Keine Bearbeitung" }, { "@language": "en", "@value": "If publication on website only. Confirmation by HSLU. CC-BY-NC-ND." } ] }, { "@type": "madek:MetaKey", "@id": "madek_rights:classification", "rdfs:label": [ { "@language": "de", "@value": "Klassifikation" }, { "@language": "en", "@value": "Classification" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_rights:web_statement", "rdfs:label": [ { "@language": "de", "@value": "Hinweis zu Rechtsgrundlagen" }, { "@language": "en", "@value": "Notes on legal basis" } ], "rdfs:comment": [ { "@language": "de", "@value": "Im Merkblatt unter Hilfe finden Sie den Link zur Studienordnung der HSLU." }, { "@language": "en", "@value": "See Academic Ordinance Governing." } ] }, { "@type": "madek:MetaKey", "@id": "madek_core:authors", "rdfs:label": [ { "@language": "de", "@value": "Autor/in" }, { "@language": "en", "@value": "Author" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_institution:project_leader", "rdfs:label": [ { "@language": "de", "@value": "Betreuer/Betreuerin" }, { "@language": "en", "@value": "Supervisor" } ], "rdfs:comment": [ { "@language": "de", "@value": "Es sind alle Betreuer/Betreuerinnnen aufzuführen." }, { "@language": "en", "@value": "Show all supervisors" } ] }, { "@type": "madek:MetaKey", "@id": "madek_core:portrayed_object_date", "rdfs:label": [ { "@language": "de", "@value": "Datierung" }, { "@language": "en", "@value": "Date" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_institution:project_type", "rdfs:label": [ { "@language": "de", "@value": "Art der Arbeit" }, { "@language": "en", "@value": "Type of assignment" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_core:description", "rdfs:label": [ { "@language": "de", "@value": "Beschreibung" }, { "@language": "en", "@value": "Description" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_core:copyright_notice", "rdfs:label": [ { "@language": "de", "@value": "Urheberrechtshinweis" }, { "@language": "en", "@value": "Copyright Notice" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_institution:project-expert", "rdfs:label": [ { "@language": "de", "@value": "Experte/Expertin" }, { "@language": "en", "@value": "Expert" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_institution:field_of_department", "rdfs:label": [ { "@language": "de", "@value": "Departement" }, { "@language": "en", "@value": "School" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_institution:field_of_study", "rdfs:label": [ { "@language": "de", "@value": "Studiengang/Studienrichtung" }, { "@language": "en", "@value": "Programme/Specialisation" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@type": "madek:MetaKey", "@id": "madek_core:title", "rdfs:label": [ { "@language": "de", "@value": "Titel" }, { "@language": "en", "@value": "Title" } ], "rdfs:comment": [ { "@language": "de", "@value": null }, { "@language": "en", "@value": null } ] }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/c6e7b266-1f09-4c82-88de-8542130ed40d", "@type": "madek:Keyword", "rdfs:label": "Englisch" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/905b81cf-659f-46f9-9a7b-a0679f0522d2", "@type": "madek:Keyword", "rdfs:label": "Luzern" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/f9e075c7-4438-40ae-a720-38ee90913d26", "@type": "madek:Keyword", "rdfs:label": "Creative Commons Lizenz: CC-BY-NC-ND" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/1cc7a5b5-85bd-405f-9fc4-af99c6f880fd", "@type": "madek:Keyword", "rdfs:label": "öffentlich" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/b6725459-379b-4b85-b8cf-273fbbeb9265", "@type": "madek:Keyword", "rdfs:label": "Studienordnung für die Ausbildung an der Hochschule Luzern, FH Zentralschweiz" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/50110184-dc50-4f29-b456-e81e5e640598", "@type": "madek:Keyword", "rdfs:label": "Masterarbeit" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/e02dca77-c61e-48af-be55-0bbb8a6d916a", "@type": "madek:Keyword", "rdfs:label": "Technik & Architektur" }, { "@id": "https://portfoliodb.hslu.ch/vocabulary/keyword/64d4a247-1d84-4fcd-a089-a3bbf7447c95", "@type": "madek:Keyword", "rdfs:label": "Engineering" }, { "@type": "madek:Person", "@id": "https://portfoliodb.hslu.ch/people/215b84d8-ec03-402b-840d-10c1a03de580", "rdfs:label": "Anna Kiener" }, { "@type": "madek:Person", "@id": "https://portfoliodb.hslu.ch/people/8705486c-f42f-4981-9980-57a112c9d41b", "rdfs:label": "Luca Mangani (TA)" }, { "@type": "madek:Person", "@id": "https://portfoliodb.hslu.ch/people/a6fde3ee-1bba-4e63-95b7-b085d9e3c704", "rdfs:label": "Christof Gentner" }, { "@id": "madek:MetaKey", "@type": "rdf:Property" }, { "@id": "madek:Role", "@type": "rdf:Property" } ] }