Multi-parameter regularization arising in optimal control of fluid flows

DSpace Repositorium (Manakin basiert)


Dateien:

Zitierfähiger Link (URI): http://hdl.handle.net/10900/59004
http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-590044
http://dx.doi.org/10.15496/publikation-428
Dokumentart: Dissertation
Erscheinungsdatum: 2015-01-27
Sprache: Englisch
Fakultät: 7 Mathematisch-Naturwissenschaftliche Fakultät
Fachbereich: Mathematik
Gutachter: Prohl, Andreas (Prof. Dr.)
Tag der mündl. Prüfung: 2015-01-26
DDC-Klassifikation: 510 - Mathematik
Schlagworte: Optimierung , Finite-Elemente-Methode , Differentialgleichung
Freie Schlagwörter: Gamma-Konvergenz
Lagrange Multiplikatoren
Verfahren des steilsten Abstiegs
Perimeterfunktional
Navier-Stokes Gleichung
Dünne Filme Gleichung
Optimalsteuerung
optimal control
Navier-Stokes equation
thin film equation
Gamma convergence
Lagrange multipliers
Method of steepest descent
perimeter functional
Lizenz: http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=de http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=en
Gedruckte Kopie bestellen: Print-on-Demand
Zur Langanzeige

Inhaltszusammenfassung:

Ziel dieser Arbeit ist es, optimale Steuerungsprobleme mit Gleichungen, die auf dem Gebiet der Fluiddynamik auftauchen, zu studieren. Die vorliegende Doktorarbeit ist in zwei wesentliche Teile aufgespalten. Jeder Teil behandelt eine wichtige partielle Differentialgleichung, die von Interesse in weitreichenden Anwendungen sind und noch immer aktuell erforscht werden: Die dichteabhängige Navier-Stokes Gleichung und die dünne Filme Gleichung. Diese optimalen Steuerungsprobleme sind vielseitig motiviert: Zunächst sind die Gleichungen aufgrund starker nichtlinearer Effekte, die im Rahmen der Optimierung zusätzlich zu Kopplungseffekten führen, mathematisch interessant. Weiter ist es nicht unmittelbar klar, ob sich Eigenschaften (wie etwa die Konvergenz numerischer Approximationen) innerhalb der Optimalsteuerungsproblems vererben. Die Erkenntnisse in der Literatur über Optimalsteuerungsprobleme bezüglich nichtlinearen partiellen Differentialgleichungen sind rar, während der Kenntnisstand über jene Optimalsteuerungsprobleme, die sich mit den benannten Gleichungen beschäftigten, noch viel unvollständiger ist: Es sind bisher nur sehr wenige Arbeiten darüber bekannt und der Inhalt der vorliegenden Doktorarbeit ist ein großer Beitrag zu diesem Thema. Schließlich gibt es für beide Probleme industrielle Anwendungen, welche die Steuerung von Flüssigkeitsströmungen erforderlich machen (diese werden innerhalb der vorliegenden Arbeit auch thematisiert), wie etwa die Kontrolle von Grenzflächen in der Aluminiumproduktion oder die Kontrolle von dünnen Flüssigkeitsschicht auf Siliziumwafern. In beiden Teilen dieser Arbeit ist die Verwendung von Regularisierungsparametern unerlässlich, um analytische Probleme zu überwinden. Die Kopplung dieser Parameter wird spezifiziert und (im zweiten Teil) ist ein Grenzproblem für den Fall gelöst, dass die Parameter gegen Null konvergieren. Im ersten Teil der vorliegenden Doktorarbeit betrachten wir ein Optimierungsproblem, um die Grenzfläche eines zweidimensionalen Zwei-Phasen Problems zu steuern. Das zu minimierende Funktional besteht dabei aus zwei Teilen: Dem Abstand zu einem gegebenen gewünschten Dichteprofil (gemessen in der $L^2$-Norm) sowie der Länge der Grenzfläche. Wir zeigen Existenz einer optimalen Steuerung und leiten notwendige Optimalitätsbedingungen erster Ordnung für eine zugehörige Phasenfeld-Approximation her. Wir schlagen ein vorbehaltlos stabiles volldiskretes Schema vor, welches auf einer Finite Elemente Diskretisierung niedriger Ordnung beruht, und wir zeigen für verschwindende Diskretisierungsparameter die Konvergenz zugehöriger optimaler Steuerungen gegen Lösungen der kontinuierlichen Optimalitätsbedingungen. Der zweite Teil besteht aus einem Optimalsteuerungsproblem bezüglich der dünnen Filme Gleichung, welche aus der Navier-Stokes Gleichung hergeleitet wird. Der dünnen Filme Gleichung fehlt für allgemeine Kontrolle die Wohlgestelltheit aufgrund möglicher Degeneriertheit; Zustandsbeschränkungen werden benutzt, um dieses Problem in den Griff zu umgehen und Wohlgestelltheit des Optimalsteuerungsproblems sicherzustellen sowie notwendige Optimalitätsbedingungen erster Ordnung rigoros herzuleiten. Ein Mehrparameteransatz zur Regularisierung, der beide Probleme -- den möglicherweise degenerierten Term in der Gleichung und die Zustandsbeschränkungen -- anspricht, wird betrachtet, und für diesen Ansatz wird Konvergenz für verschwindende Regularisierungsparameter gezeigt, der beide Effekte entkoppelt. Beide Teile werden werden von entsprechenden numerischen Experimenten abgeschlossen, welche die Modelle prüfen, Parameter und deren Skalierungen miteinander vergleichen (solche, die für die Regularisierung nötig sind, aber auch welche, die in den numerischen Algorithmen vorkommen) und Beispiel der industriellen Anwendungen auf akademischen Niveau beinhalten.

Abstract:

The objective of this thesis is to study optimal control problems subject to equations arising in the field of fluid dynamics. This thesis is split into two essential parts. Each of them deals with an important partial differential equation, that are of interest in various applications and are widely considered in current research: The density dependent Navier--Stokes equation and the thin-film equation. These optimal control problems are motivated in many ways: First, the equations are mathematically interesting due to strong nonlinear effects occurring additionally as coupling effects in the context of optimization. Also, it is not immediate that properties (such as convergence of numerical approximations) are inherited by the optimal control problem. The literature on optimal control subject to nonlinear partial differential equation is rare, while the knowledge on those problems subject to the mentioned equations is even more rare: Only very few works are known, and the content of this thesis is a big contribution to this topic. Finally, for both control problems, there are industrial applications requiring the optimal control of fluid flows (which will also be addressed within the this thesis) such as the control of the interface in aluminum production, or the control of thin liquid layer on a silicon wafer. In both parts, the use of regularization parameters is vital in order to overcome analytical issues. The coupling of these parameters is specified, and (in the second part) a limiting problem is solved for these parameters tending to zero. In the first part of this thesis, we consider an optimal control problem for the interface in a two-dimensional two-phase fluid problem. The minimization functional consists of two parts: The $L^2$-distance to a given density profile and the interfacial length. We show existence of an optimal control and derive necessary first order optimality conditions for a corresponding phase field approximation. An unconditionally stable fully discrete scheme which is based on low order finite element discretization is proposed, and convergence of corresponding iterates to solutions of the continuous optimality conditions for vanishing discretization parameters is shown. The second part consists of an optimal control problem subject to the thin-film equation which is deduced from the Navier--Stokes equation. The thin-film equation lacks well-posedness for general controls due to possible degeneracies; state constraints are used to circumvent this problematic issue, and ensure well-posedness of the optimal control problem as well as the rigorous derivation of necessary first order optimality conditions for the optimal control problem. A multi-parameter regularization addressing both, the possibly degenerate term in the equation and the state constraint, is considered, and convergence is shown for vanishing regularization parameters by decoupling both effects. Both parts are concluded by corresponding numerical experiments, validating the models, comparing parameters (of the regularization and of the numerical algorithms) and their scaling to each other, and including academic examples of industrial applications.

Das Dokument erscheint in: