In this dissertation, we consider two different strategies for improving the projection-based reduced order model (ROM) accuracy: (I) adding closure terms to the standard ROM; (II) using Lagrangian data to improve the ROM basis. Following strategy (I), we propose a new data-driven reduced order model (ROM) framework that centers around the hierarchical structure of the variational multiscale (VMS) methodology and utilizes data to increase the ROM accuracy at a modest computational cost. The VMS methodology is a natural fit for the hierarchical structure of the ROM basis: In the first step, we use the ROM projection to separate the scales into three categories: (i) resolved large scales, (ii) resolved small scales, and (iii) unresolved scales. In the second step, we explicitly identify the VMS-ROM closure terms, i.e., the terms representing the interactions among the three types of scales. In the third step, we use available data to model the VMS-ROM closure terms. Thus, instead of phenomenological models used in VMS for standard numerical discretizations (e.g., eddy viscosity models), we utilize available data to construct new structural VMS-ROM closure models. Specifically, we build ROM operators (vectors, matrices, and tensors) that are closest to the true ROM closure terms evaluated with the available data. We test the new data-driven VMS-ROM in the numerical simulation of four test cases: (i) the 1D Burgers equation with viscosity coefficient $nu = 10^{-3}$; (ii) a 2D flow past a circular cylinder at Reynolds numbers $Re=100$, $Re=500$, and $Re=1000$; (iii) the quasi-geostrophic equations at Reynolds number $Re=450$ and Rossby number $Ro=0.0036$; and (iv) a 2D flow over a backward facing step at Reynolds number $Re=1000$. The numerical results show that the data-driven VMS-ROM is significantly more accurate than standard ROMs. Furthermore, we propose a new hybrid ROM framework for the numerical simulation of fluid flows. This hybrid framework incorporates two closure modeling strategies: (i) A structural closure modeling component that involves the recently proposed data-driven variational multiscale ROM approach, and (ii) A functional closure modeling component that introduces an artificial viscosity term. We also utilize physical constraints for the structural ROM operators in order to add robustness to the hybrid ROM. We perform a numerical investigation of the hybrid ROM for the three-dimensional turbulent channel flow at a Reynolds number $Re = 13,750$. In addition, we focus on the mathematical foundations of ROM closures. First, we extend the verifiability concept from large eddy simulation to the ROM setting. Specifically, we call a ROM closure model verifiable if a small ROM closure model error (i.e., a small difference between the true ROM closure and the modeled ROM closure) implies a small ROM error. Second, we prove that a data-driven ROM closure (i.e., the data-driven variational multiscale ROM) is verifiable. For strategy (II), we propose new Lagrangian inner products that we use together with Eulerian and Lagrangian data to construct new Lagrangian ROMs. We show that the new Lagrangian ROMs are orders of magnitude more accurate than the standard Eulerian ROMs, i.e., ROMs that use standard Eulerian inner product and data to construct the ROM basis. Specifically, for the quasi-geostrophic equations, we show that the new Lagrangian ROMs are more accurate than the standard Eulerian ROMs in approximating not only Lagrangian fields (e.g., the finite time Lyapunov exponent (FTLE)), but also Eulerian fields (e.g., the streamfunction). We emphasize that the new Lagrangian ROMs do not employ any closure modeling to model the effect of discarded modes (which is standard procedure for low-dimensional ROMs of complex nonlinear systems). Thus, the dramatic increase in the new Lagrangian ROMs' accuracy is entirely due to the novel Lagrangian inner products used to build the Lagrangian ROM basis. / Doctor of Philosophy / Reduced order models (ROMs) are popular in physical and engineering applications: for example, ROMs are widely used in aircraft designing as it can greatly reduce computational cost for the aircraft's aeroelastic predictions while retaining good accuracy. However, for high Reynolds number turbulent flows, such as blood flows in arteries, oil transport in pipelines, and ocean currents, the standard ROMs may yield inaccurate results. In this dissertation, to improve ROM's accuracy for turbulent flows, we investigate three different types of ROMs. In this dissertation, both numerical and theoretical results show that the proposed new ROMs yield more accurate results than the standard ROM and thus can be more useful.

Simulations and modeling of large-scale systems are vital to understanding real world phenomena. However, even advanced numerical models can only approximate the true physics. The discrepancy between model results and nature can be attributed to different sources of uncertainty including the parameters of the model, input data, or some missing physics that is not included in the model due to a lack of knowledge or high computational costs. Uncertainty reduction approaches seek to improve the model accuracy by decreasing the overall uncertainties in models. Aiming to contribute to this area, this study explores uncertainty quantification and reduction approaches for complex physical problems. This study proposes several novel probabilistic and statistical approaches for identifying the sources of uncertainty, modeling the errors, and reducing uncertainty to improve the model predictions for large-scale simulations. We explore different computational models. The first class of models studied herein are inherently stochastic, and numerical approximations suffer from stability and accuracy issues. The second class of models are partial differential equations, which capture the laws of mathematical physics; however, they only approximate a more complex reality, and have uncertainties due to missing dynamics which is not captured by the models. The third class are low-fidelity models, which are fast approximations of very expensive high-fidelity models. The reduced-order models have uncertainty due to loss of information in the dimension reduction process. We also consider uncertainty analysis in the data assimilation framework, specifically for ensemble based methods where the effect of sampling errors is alleviated by localization. Finally, we study the uncertainty in numerical weather prediction models coming from approximate descriptions of physical processes. / Ph. D. / Computational models are used to understand the behavior of the natural phenomenon. Models are used to approximate the evolution of the true phenomenon or reality in time. We obtain more accurate forecast for the future by combining the model approximation together with the observation from reality. Weather forecast models, oceanography, geoscience, etc. are some examples of the forecasting models. However, models can only approximate the true reality to some extent and model approximation of reality is not perfect due to several sources of error or uncertainty. The noise in measurements or in observations from nature, the uncertainty in some model components, some missing components in models, the interaction between different components of the model, all cause model forecast to be different from reality. The aim of this study is to explore the techniques and approaches of modeling the error and uncertainty of computational models, provide solution and remedies to reduce the error of model forecast and ultimately improve the model forecast. Taking the discrepancy or error between model forecast and reality in time and mining that error provide valuable information about the origin of uncertainty in models as well as the hidden dynamics that is not considered in the model. Statistical and machine learning based solutions are proposed in this study to identify the source of uncertainty, capturing the uncertainty and using that information to reduce the error and enhancing the model forecast. We studied the error modeling, error or uncertainty quantification and reduction techniques in several frameworks from chemical models to weather forecast models. In each of the models, we tried to provide proper solution to detect the origin of uncertainty, model the error and reduce the uncertainty to improve the model forecast.

O objetivo deste estudo é apresentar uma revisão sobre técnicas de redução da ordem de modelos dinâmicos lineares e invariantes no tempo. Com a implementação de tais técnicas, o autor mostra que é possível reproduzir as principais características da resposta de um modelo de alta ordem através de um modelo de ordem reduzida com menor número de graus de liberdade. Uma metodologia para redução da ordem de modelos de sistemas dinâmicos foi apresentada. Os processos envolvidos nesta metodologia foram descritos, os quais são: técnicas para realizar projeções do sistema em diferentes bases, selecionar os graus de liberdade que são bons candidatos a eliminação, eliminar graus de liberdade do modelo completo e implementar correções na resposta do modelo reduzido. Foram apresentadas maneiras de quantificar as similitudes entre as respostas dos modelos completo e reduzido através de métricas de representatividade. Para implementar e estudar as técnicas de redução apresentadas, o autor elaborou dois modelos para estudo de caso: um modelo para estudo da dinâmica vertical de um veículo de passeio e outro modelo para estudo da dinâmica longitudinal de um trem. Diferentes técnicas de redução foram implementadas a partir dos dois modelos para estudo de caso e os resultados foram comparados através das métricas de representatividade. O critério proposto pelo autor para quantificar desempenho de um modelo reduzido foi utilizado para determinar de maneira objetiva o modelo reduzido mais adequado para cada aplicação. Como contribuição neste trabalho, o autor propôs uma definição de desempenho de um modelo reduzido e um método para quantificar o mesmo, além de duas novas métricas para mensurar a capacidade do modelo reduzido em reproduzir os máximos sobre-sinais e tempos de acomodação do modelo completo. / The objective of this work is to present a revision about model order reduction techniques applied to linear, time invariant dynamic systems. With the implementation of these techniques, the author shows that it is possible to reproduce the main characteristics of the response of a high order dynamic system using a reduced order model with fewer degrees of freedom. A model order reduction methodology was presented. The processes which are involved in this methodology were described, which are: techniques for projection onto different basis, selection of the most suitable degrees of freedom to be reduced, elimination of degrees of freedom from the high order model, implementation of corrections at the reduced model response. The author showed ways to quantify the similarities between the responses of the complete and reduced models using representativeness metrics. In order to implement and study the presented model order reduction techniques, the author developed two case study models: one model to study the vertical dynamics of a passenger car and another model to study the longitudinal dynamics of a train. Different model order reduction techniques were implemented and its results were compared using representativeness metrics and the performance of the reduced models. The criteria proposed by the author to quantify the performance of a reduced order model was used to objectively determine the most suitable reduced order model for each application. The author proposed, as contribution at this work, a definition of the reduced order model performance, a method to quantify its performance and two new metrics to measure the capacity of the reduced model to reproduce the overshoots and settling times of the complete model.

Ce mémoire de thèse concerne l'interpolation sur les variétés de Grassmann et ses applications à la réduction de modèles en mécanique et plus généralement aux systèmes d'équations aux dérivées partielles d'évolution. Après une description de la méthode POD, nous introduisons les fondements théoriques en géométrie des variétés de Grassmann, qui seront utilisés dans le reste de la thèse. Ce chapitre donne à ce mémoire à la fois une rigueur mathématique au niveau des algorithmes mis au point, leur domaine de validité ainsi qu'une estimation de l'erreur en distance grassmannienne, mais également un caractère auto-contenu "self-contained" du manuscrit. Ensuite, on présente la méthode d'interpolation sur les variétés de Grassmann introduite par David Amsallem et Charbel Farhat. Cette méthode sera le point de départ des méthodes d'interpolation que nous développerons dans les chapitres suivants. La méthode de Amsallem-Farhat consiste à choisir un point d'interpolation de référence, envoyer l'ensemble des points d'interpolation sur l'espace tangent en ce point de référence via l'application logarithme géodésique, effectuer une interpolation classique sur cet espace tangent, puis revenir à la variété de Grassmann via l'application exponentielle géodésique. On met en évidence par des essais numériques l'influence du point de référence sur la qualité des résultats. Dans notre premier travail, nous présentons une version grassmannienne d'un algorithme connu dans la littérature sous le nom de Pondération par Distance Inverse (IDW). Dans cette méthode, l'interpolé en un point donné est considéré comme le barycentre des points d'interpolation où les coefficients de pondération utilisés sont inversement "proportionnels" à la distance entre le point considéré et les points d'interpolation. Dans notre méthode, notée IDW-G, la distance géodésique sur la variété de Grassmann remplace la distance euclidienne dans le cadre standard des espaces euclidiens. L'avantage de notre algorithme, dont on a montré la convergence sous certaines conditions assez générales, est qu'il ne requiert pas de point de référence contrairement à la méthode de Amsallem-Farhat. Pour remédier au caractère itératif (point fixe) de notre première méthode, nous proposons une version directe via la notion de barycentre généralisé. Notons enfin que notre algorithme IDW-G dépend nécessairement du choix des coefficients de pondération utilisés. Dans notre second travail, nous proposons une méthode qui permet un choix optimal des coefficients de pondération, tenant compte de l'auto-corrélation spatiale de l'ensemble des points d'interpolation. Ainsi, chaque coefficient de pondération dépend de tous les points d'interpolation et non pas seulement de la distance entre le point considéré et un point d'interpolation. Il s'agit d'une version grassmannienne de la méthode de Krigeage, très utilisée en géostatique. La méthode de Krigeage grassmannienne utilise également le point de référence. Dans notre dernier travail, nous proposons une version grassmannienne de l'algorithme de Neville qui permet de calculer le polynôme d'interpolation de Lagrange de manière récursive via l'interpolation linéaire entre deux points. La généralisation de cet algorithme sur une variété grassmannienne est basée sur l'extension de l'interpolation entre deux points (géodésique/droite) que l'on sait faire de manière explicite. Cet algorithme ne requiert pas le choix d'un point de référence, il est facile d'implémentation et très rapide. De plus, les résultats numériques obtenus sont remarquables et nettement meilleurs que tous les algorithmes décrits dans ce mémoire. / This dissertation deals with interpolation on Grassmann manifolds and its applications to reduced order methods in mechanics and more generally for systems of evolution partial differential systems. After a description of the POD method, we introduce the theoretical tools of grassmannian geometry which will be used in the rest of the thesis. This chapter gives this dissertation a mathematical rigor in the performed algorithms, their validity domain, the error estimate with respect to the grassmannian distance on one hand and also a self-contained character to the manuscript. The interpolation on Grassmann manifolds method introduced by David Amsallem and Charbel Farhat is afterward presented. This method is the starting point of the interpolation methods that we will develop in this thesis. The method of Amsallem-Farhat consists in chosing a reference interpolation point, mapping forward all interpolation points on the tangent space of this reference point via the geodesic logarithm, performing a classical interpolation on this tangent space and mapping backward the interpolated point to the Grassmann manifold by the geodesic exponential function. We carry out the influence of the reference point on the quality of the results through numerical simulations. In our first work, we present a grassmannian version of the well-known Inverse Distance Weighting (IDW) algorithm. In this method, the interpolation on a point can be considered as the barycenter of the interpolation points where the used weights are inversely proportional to the distance between the considered point and the given interpolation points. In our method, denoted by IDW-G, the geodesic distance on the Grassmann manifold replaces the euclidean distance in the standard framework of euclidean spaces. The advantage of our algorithm that we show the convergence undersome general assumptions, does not require a reference point unlike the method of Amsallem-Farhat. Moreover, to carry out this, we finally proposed a direct method, thanks to the notion of generalized barycenter instead of an earlier iterative method. However, our IDW-G algorithm depends on the choice of the used weighting coefficients. The second work deals with an optimal choice of the weighting coefficients, which take into account of the spatial autocorrelation of all interpolation points. Thus, each weighting coefficient depends of all interpolation points an not only on the distance between the considered point and the interpolation point. It is a grassmannian version of the Kriging method, widely used in Geographic Information System (GIS). Our grassmannian Kriging method require also the choice of a reference point. In our last work, we develop a grassmannian version of Neville's method which allow the computation of the Lagrange interpolation polynomial in a recursive way via the linear interpolation of two points. The generalization of this algorithm to grassmannian manifolds is based on the extension of interpolation of two points (geodesic/straightline) that we can do explicitly. This algorithm does not require the choice of a reference point, it is easy to implement and very quick. Furthermore, the obtained numerical results are notable and better than all the algorithms described in this dissertation.

La résolution des problèmes de contrôle optimal nécessite des temps de calcul et des capacités de stockage très élevés. Pour s’affranchir de ces contraintes, il est possible d’utiliser les méthodes de réduction de modèles comme la POD (Proper Orthogonal Decomposition). L’inconvénient de cette approche est que la base POD n’est valable que pour des paramètres situés dans un voisinage proche des paramètres pour lesquels elle a été construite. Par conséquent, en contrôle optimal, cette base peut ne pas être représentative de tous les paramètres qui seront proposés par l’algorithme de contrôle. Pour s’affranchir de cet handicap, une méthodologie de contrôle optimal utilisant des modèles réduits adaptatifs a été proposée dans ce manuscrit. Les bases réduites adaptées sont obtenues à l’aide de la méthode d’interpolation ITSGM (Interpolation on Tangent Subspace of Grassman Manifold) ou de la méthode d’enrichissement PGD (Proper Generalized Decomposition). La robustesse de cette approche en termes de précision et de temps de calcul a été démontrée pour le contrôle optimal (basé sur les équations adjointes) des équations 2D de réaction-diffusion et de Burgers. L’approche basée sur l’interpolation ITSGM a également été appliquée avec succès pour contrôler l’écoulement autour d’un cylindre 2D. Deux méthodes de réduction non intrusives, ne nécessitant pas la connaissance des équations du modèle étudié, ont également été proposées. Ces méthodes appelées NIMR (Non Intrusive Model Reduction) et HNIMR (Hyper Non Intrusive Model Reduction) ont été couplées à un algorithme génétique pour résoudre rapidement un problème de contrôle optimal. Le problème du contrôle optimal de l’écoulement autour d’un cylindre 2D a été étudié et les résultats ont montré l’efficacité de cette approche. En effet, l’algorithme génétique couplé avec la méthode HNIMR a permis d’obtenir les solutions avec une bonne précision en moins de 40 secondes. / The numerical resolution of adjoint based optimal control problems requires high computational time and storage capacities. In order to get over these high requirement, it is possible to use model reduction techniques such as POD (Proper Orthogonal Decomposition). The disadvantage of this approach is that the POD basis is valid only for parameters located in a small neighborhood to the parameters for which it was built. Therefore, this basis may not be representative for all parameters in the optimizer’s path eventually suggested by the optimal control loop. To overcome this issue, a reduced optimal control methodology using adaptive reduced order models obtained by the ITSGM (Interpolation on a Tangent Subspace of the Grassman Manifold) method or by the PGD (Proper Generalized Decomposition) method, has been proposed in this work. The robustness of this approach in terms of precision and computation time has been demonstrated for the optimal control (based on adjoint equations) of the 2D reaction-diffusion and Burgers equations. The interpolation method ITSGM has also been validated in the control of flow around a 2D cylinder. In the context of non intrusive model reduction, two non intrusive reduction methods, which do not require knowledge of the equations of the studied model, have also been proposed. These methods called NIMR (Non-Intrusive Model Reduction) and HNIMR (Hyper Non-Intrusive Model Reduction) were developed and then coupled to a genetic algorithm in order to solve an optimal control problem in quasi-real time. The problem of optimal control of the flow around a 2D cylinder has been studied and the results have shown the effectiveness of this approach. Indeed, the genetic algorithm coupled with the HNIMR method allowed to obtain the solutions with a good accuracy in less than 40 seconds.

Dans le cadre de la lutte contre les nuisances sonores et vibratoires, cette thèse porte sur la modélisation numérique des structures amorties par dispositifs piézoélectriques shuntés. La première partie du travail concerne la modélisation par éléments finis de structures en vibrations avec des pastilles piézoélectriques shuntées. Dans un premier temps, une formulation éléments finis originale, qui utilise des variables électriques globales (différence de potentiel et charge dans chaque pastille piézoélectrique), est analysée et validée. Dans un second temps, différentes stratégies de réduction de modèle basées sur la méthode de projection modale sont proposées pour résoudre le problème électromécanique discrétisé par éléments finis à moindre coût. La convergence de ces modèles d’ordre réduits est ensuite analysée pour les cas de shunts résistif et résonant. La deuxième partie du travail est consacrée à l’optimisation du système électromécanique, dans le but de maximiser l’amortissement apporté par les dispositifs piézoélectriques shuntés. Pour cela, une procédure d’optimisation topologique, basée sur la méthode SIMP (Solid Isotropic Material with Penalization method), est développée pour déterminer les géométries et les emplacements optimaux des pastilles piézoélectriques. Cette procédure permet de maximiser le coefficient de couplage électromécanique modal entre les éléments piézoélectriques et la structure hôte, ceci de façon indépendante du choix des composants du circuit électrique. Les avantages de l’approche proposée sont mis en avant à travers un exemple de validation et un cas d'application industrielle. Enfin, la dernière partie du travail propose une approche numérique pour modéliser et optimiser la réduction du rayonnement acoustique de plaques minces dans le domaine des basses fréquences avec des éléments piézoélectriques shuntés. Cette approche est valable pour n’importe quelle plaque mince bafflée et non trouée, indépendamment des conditions aux limites. Un exemple d’application concernant l’atténuation du rayonnent acoustique d’une plaque avec renforts est présenté et analysé. / Passive structural vibration and noise reduction by means of shunted piezoelectric patches is addressed in this thesis. The first part of the work concerns the finite element modeling of shunted piezoelectric systems. Firstly, an original finite element formulation, with only a couple of electric variables per piezoelectric patch (the global charge/ voltage), is analyzed and validated. Secondly, several reduced order models based on a normal mode expansion are proposed to solve the electromechanical problem. The convergence of these reduced order models is then analyzed for a resistive and a resonant shunt circuits. In the second part of the work, the concept of topology optimization, based on the Solid Isotropic Material with Penalization method (SIMP), is employed to optimize, in terms of damping efficiency, the geometry of piezoelectric patches as well as their placement on the host elastic structure. The proposed optimization procedure consists of distributing the piezoelectric material in such a way as to maximize the modal electromechanical coupling factor of the mechanical vibration mode to which the shunt is tuned, independently of the choice of electric circuit components. Numerical examples validate and demonstrate the potential of the proposed approach for the design of piezoelectric shunt devices. Finally, the last part of the work concerns the numerical modeling of noise and vibration reduction of thin structures in the low frequency range by using shunted piezoelectric elements. An efficient approach that can be applied to any thin continuous plates in an infinite baffle, independently of the boundary conditions, is proposed. An application example of a thin plate with reinforcements is presented and analyzed.

Dans le cadre de la lutte contre les nuisances sonores et vibratoires, cette thèse porte sur la modélisation numérique des structures amorties par dispositifs piézoélectriques shuntés. La première partie du travail concerne la modélisation par éléments finis de structures en vibrations avec des pastilles piézoélectriques shuntées. Dans un premier temps, une formulation éléments finis originale, qui utilise des variables électriques globales (différence de potentiel et charge dans chaque pastille piézoélectrique), est analysée et validée. Dans un second temps, différentes stratégies de réduction de modèle basées sur la méthode de projection modale sont proposées pour résoudre le problème électromécanique discrétisé par éléments finis à moindre coût. La convergence de ces modèles d’ordre réduits est ensuite analysée pour les cas de shunts résistif et résonant. La deuxième partie du travail est consacrée à l’optimisation du système électromécanique, dans le but de maximiser l’amortissement apporté par les dispositifs piézoélectriques shuntés. Pour cela, une procédure d’optimisation topologique, basée sur la méthode SIMP (Solid Isotropic Material with Penalization method), est développée pour déterminer les géométries et les emplacements optimaux des pastilles piézoélectriques. Cette procédure permet de maximiser le coefficient de couplage électromécanique modal entre les éléments piézoélectriques et la structure hôte, ceci de façon indépendante du choix des composants du circuit électrique. Les avantages de l’approche proposée sont mis en avant à travers un exemple de validation et un cas d'application industrielle. Enfin, la dernière partie du travail propose une approche numérique pour modéliser et optimiser la réduction du rayonnement acoustique de plaques minces dans le domaine des basses fréquences avec des éléments piézoélectriques shuntés. Cette approche est valable pour n’importe quelle plaque mince bafflée et non trouée, indépendamment des conditions aux limites. Un exemple d’application concernant l’atténuation du rayonnent acoustique d’une plaque avec renforts est présenté et analysé. / Passive structural vibration and noise reduction by means of shunted piezoelectric patches is addressed in this thesis. The first part of the work concerns the finite element modeling of shunted piezoelectric systems. Firstly, an original finite element formulation, with only a couple of electric variables per piezoelectric patch (the global charge/ voltage), is analyzed and validated. Secondly, several reduced order models based on a normal mode expansion are proposed to solve the electromechanical problem. The convergence of these reduced order models is then analyzed for a resistive and a resonant shunt circuits. In the second part of the work, the concept of topology optimization, based on the Solid Isotropic Material with Penalization method (SIMP), is employed to optimize, in terms of damping efficiency, the geometry of piezoelectric patches as well as their placement on the host elastic structure. The proposed optimization procedure consists of distributing the piezoelectric material in such a way as to maximize the modal electromechanical coupling factor of the mechanical vibration mode to which the shunt is tuned, independently of the choice of electric circuit components. Numerical examples validate and demonstrate the potential of the proposed approach for the design of piezoelectric shunt devices. Finally, the last part of the work concerns the numerical modeling of noise and vibration reduction of thin structures in the low frequency range by using shunted piezoelectric elements. An efficient approach that can be applied to any thin continuous plates in an infinite baffle, independently of the boundary conditions, is proposed. An application example of a thin plate with reinforcements is presented and analyzed.

O objetivo deste estudo é apresentar uma revisão sobre técnicas de redução da ordem de modelos dinâmicos lineares e invariantes no tempo. Com a implementação de tais técnicas, o autor mostra que é possível reproduzir as principais características da resposta de um modelo de alta ordem através de um modelo de ordem reduzida com menor número de graus de liberdade. Uma metodologia para redução da ordem de modelos de sistemas dinâmicos foi apresentada. Os processos envolvidos nesta metodologia foram descritos, os quais são: técnicas para realizar projeções do sistema em diferentes bases, selecionar os graus de liberdade que são bons candidatos a eliminação, eliminar graus de liberdade do modelo completo e implementar correções na resposta do modelo reduzido. Foram apresentadas maneiras de quantificar as similitudes entre as respostas dos modelos completo e reduzido através de métricas de representatividade. Para implementar e estudar as técnicas de redução apresentadas, o autor elaborou dois modelos para estudo de caso: um modelo para estudo da dinâmica vertical de um veículo de passeio e outro modelo para estudo da dinâmica longitudinal de um trem. Diferentes técnicas de redução foram implementadas a partir dos dois modelos para estudo de caso e os resultados foram comparados através das métricas de representatividade. O critério proposto pelo autor para quantificar desempenho de um modelo reduzido foi utilizado para determinar de maneira objetiva o modelo reduzido mais adequado para cada aplicação. Como contribuição neste trabalho, o autor propôs uma definição de desempenho de um modelo reduzido e um método para quantificar o mesmo, além de duas novas métricas para mensurar a capacidade do modelo reduzido em reproduzir os máximos sobre-sinais e tempos de acomodação do modelo completo. / The objective of this work is to present a revision about model order reduction techniques applied to linear, time invariant dynamic systems. With the implementation of these techniques, the author shows that it is possible to reproduce the main characteristics of the response of a high order dynamic system using a reduced order model with fewer degrees of freedom. A model order reduction methodology was presented. The processes which are involved in this methodology were described, which are: techniques for projection onto different basis, selection of the most suitable degrees of freedom to be reduced, elimination of degrees of freedom from the high order model, implementation of corrections at the reduced model response. The author showed ways to quantify the similarities between the responses of the complete and reduced models using representativeness metrics. In order to implement and study the presented model order reduction techniques, the author developed two case study models: one model to study the vertical dynamics of a passenger car and another model to study the longitudinal dynamics of a train. Different model order reduction techniques were implemented and its results were compared using representativeness metrics and the performance of the reduced models. The criteria proposed by the author to quantify the performance of a reduced order model was used to objectively determine the most suitable reduced order model for each application. The author proposed, as contribution at this work, a definition of the reduced order model performance, a method to quantify its performance and two new metrics to measure the capacity of the reduced model to reproduce the overshoots and settling times of the complete model.

Ce mémoire de thèse est formé de quatre chapitres. Un premier chapitre présente les différentes notions et outils mathématiques utilisés dans le corps de la thèse ainsi qu’une description des résultats principaux que nous avons obtenus. Le second chapitre présente une généralisation d’un résultat obtenu par Rousselet-Chénais en 1990 qui décrit la sensibilité des sous-espaces propres d’opérateurs compacts auto-adjoints. Rousselet-Chénais se sont limités aux sous-espaces propres de dimension 1 et nous avons étendu leur résultat aux dimensions supérieures. Nous avons appliqué nos résultats à la Décomposition par Projection Orthogonale (POD) dans le cas de variation paramétrique, temporelle ou spatiale (Gappy-POD). Le troisième chapitre traite de l’estimation du flot optique avec des énergies quadratiques ou linéaires à l’infini. On montre des résultats mathématiques de convergence de la méthode de Décomposition Progressive Généralisée (PGD) dans le cas des énergies quadratiques. Notre démonstration est basée sur la décomposition de Brézis-Lieb via la convergence presque-partout de la suite gradient PGD. Une étude numérique détaillée est faite sur différents type d’images : sur les équations de transport de scalaire passif, dont le champ de déplacement est solution des équations de Navier-Stokes. Ces équations présentent un défi pour l’estimation du flot optique à cause du faible gradient dans plusieurs régions de l’image. Nous avons appliqué notre méthode aux séquences d’images IRM pour l’estimation du mouvement des organes abdominaux. La PGD a présenté une supériorité à la fois au niveau du temps de calcul (même en 2D) et au niveau de la représentation correcte des mouvements estimés. La diffusion locale des méthodes classiques (Horn & Schunck, par exemple) ralentit leur convergence contrairement à la PGD qui est une méthode plus globale par nature. Le dernier chapitre traite de l’application de la méthode PGD dans le cas d’équations elliptiques variationnelles dont l’énergie présente tous les défis aux méthodes variationnelles classiques : manque de convexité, manque de coercivité et manque du caractère borné de l’énergie. Nous démontrons des résultats de convergence, pour la topologie faible, des suites PGD (lorsqu’elles sont bien définies) vers deux solutions extrémales sur la variété de Nehari. Plusieurs questions mathématiques concernant la PGD restent ouvertes dans ce chapitre. Ces questions font partie de nos perspectives de recherche. / This thesis is formed of four chapters. The first one presents the mathematical notions and tools used in this thesis and gives a description of the main results obtained within. The second chapter presents our generalization of a result obtained by Rousselet-Chenais in 1990 which describes the sensitivity of eigensubspaces for self-adjoint compact operators. Rousselet-Chenais were limited to sensitivity for specific subspaces of dimension 1, we have extended their result to higher dimensions. We applied our results to the Proper Orthogonal Decomposition (POD) in the case of parametric, temporal and spatial variations (Gappy- POD). The third chapter discusses the optical flow estimate with quadratic or linear energies at infinity. Mathematical results of convergence are shown for the method Progressive Generalized Decomposition (PGD) in the case of quadratic energies. Our proof is based on the decomposition of Brézis-lieb via the convergence almost everywhere of the PGD sequence gradients. A detailed numerical study is made on different types of images : on the passive scalar transport equations, whose displacement fields are solutions of the Navier-Stokes equations. These equations present a challenge for optical flow estimates because of the presence of low gradient regions in the image. We applied our method to the MRI image sequences to estimate the movement of the abdominal organs. PGD presented a superiority in both computing time level (even in 2D) and accuracy representation of the estimated motion. The local diffusion of standard methods (Horn Schunck, for example) limits the convergence rate, in contrast to the PGD which is a more global approach by construction. The last chapter deals with the application of PGD method in the case of variational elliptic equations whose energy present all challenges to classical variational methods : lack of convexity, lack of coercivity and lack of boundedness. We prove convergence results for the weak topology, the PGD sequences converge (when they are well defined) to two extremal solutions on the Nehari manifold. Several mathematical questions about PGD remain open in this chapter. These questions are part of our research perspectives.

L’objectif de ce travail consiste à développer la méthode de résolution PGD (Proper Generalized Decomposition), qui est une méthode de réduction de modèle où la solution est recherchée sous forme séparée, à la résolution des équations de Navier-Stokes. Dans un premier temps, cette méthode est appliquée à la résolution d’équations modèles disposant d’une solution analytique. L’ équation de diffusion stationnaire 2D et 3D, l’équation de diffusion instationnaire 2D et les équations de Burgers et Stokes sont traitées. Nous montrons que dans tous ces cas la méthode PGD permet de retrouver les solutions analytiques avec une précision équivalente au modèle standard. Nous mettons également en évidence la supériorité de la PGD par rapport au modèle standard en terme de temps de calcul. En effet, dans tous ces cas, laPGD se montre beaucoup plus rapide que le solveur standard (plusieurs dizaine de fois). La résolution des équations de Navier-Stokes isothermes et anisothermes est ensuite effectuée par une discrétisation volumes finis sur un maillage décalé où le couplage vitesse-pression a été géré à l’aide d’un schéma de prédiction-correction. Dans ce cas une décomposition PGD sur les variables d’espaces uniquement a été choisie. Pour les écoulements incompressibles 2D stationnaire ou instationnaire, de type cavité entrainée et/ou différentiellement chauffé, les résultats obtenus par résolution PGD sont similaires à ceux du solveur standard avec un gain de temps significatif (la PGD est une dizaine de fois plus rapide que le solveur standard). Enfin ce travail introduit une première approche de la résolution des équations de transferts par méthode PGD en formulation spectrale. Sur les différents problèmes traités, à savoir l’équation de diffusion stationnaire, l’équation de Darcy et les équations de Navier-Sokes, la PGD a montré une précision aussi bonne que le solveur standard. Un gain de temps a été observé pour le cas de l’équation de Poisson, par contre, concernant le problème de Darcy ou les équations de Navier-Stokes les performances de la PGD en terme de temps de calcul peuvent encore être améliorées. / Motivated by solving the Navier-Stokes equations, this work presents the implementation and development of a reduced order model, the PGD (Proper Generalized Decomposition).Firstly, this method is applied to solving equations models with an analytical solution. The stationary diffusion equation 2D and 3D, 2D unsteady diffusion equation and Burgers equations and Stokes are processed. We show that in all these cases, the PGD method allows to find analytical solutions with a good accuracy compared to the standard model. We also demonstrate the superiority of the PGD relative to the standard model in terms of computing time. Indeed, in all these cases, PGD was much more rapid than the standard solver (several dozen times). The Navier-Stokes 2D and 3D thermal and isothermal isotherms are then processed by a finite volume discretization on a staggered grid where the velocity-pressure coupling was handled using a prediction-correction scheme. In this case a decomposition of the space variables only was chosen. The results in 2D for Reynolds numbers equal to 100, 1000and 10, 000 are similar to those of the solver standard with a significant time saving (PGD isten times faster than the solver standard). Finally, this work introduces a first approach tosolving the Navier-Stokes equations with a spectral method coupled with the PGD. Different cases were dealed, the stationary diffusion equation, the Darcy equation and the Navier-Sokesequations. PGD showed a good accuracy compared with the standard solver. Saving time was observed for the case of the Poisson equation, on the other hand, about Darcy’s problem or Navier-Stokes’ equations, performance of the PGD in terms of computing time may yet be improved.