Une forte concurrence avec d'autres moyens de transport a poussé l’industrie ferroviaire à se réinventer et rechercher des performances toujours plus élevées. De nos jours, l’obtention de vitesses chaque fois plus élevées exige le développement de modèles numériques précis pour concevoir et prédire le comportement des voies ferrées sous les contraintes mécaniques imposées par le passage du convoi. Dans cette thèse, nous avons concentré l'étude sur la couche de ballast. Ce composant présente un comportement mécanique complexe, lié à la nature granulaire de ses composants, il peut être solide, liquide ou gazeux.Ce comportement dépend de l'état de contrainte et de l'historique de déformation du milieu.Deux classes de modèles numériques sont couramment utilisées pour prédire le comportement de ces systèmes : (1) les approches discrètes et (2) les approches continues. Pour ces premières, chaque grain du ballast est représenté par un corps rigide et interagit avec ses voisins parle biais de forces de contact non linéaires en utilisant, par exemple, la méthode de dynamique non régulière des contacts. En raison des limites de calcul, ce type de méthode ne peut résoudre que quelques mètres de longueur de ballast. Le couplage avec le sol sous la couche de ballast et avec les traverses reste également un problème non résolu dans la littérature. Pour les approches continues, le ballast est remplacé par un milieu continu homogénéisé, de façon à permettre l’utilisation de la méthode par éléments finis classique (EF). Cependant, ces modèles sont normalement utilisés avec des paramètres mécaniques homogènes, de sorte qu'ils ne représentent pas complètement l'hétérogénéité des déformations et des contraintes dans la couche de ballast.Nous étudions dans cette thèse une approche alternative, utilisant un modèle de continuum hétérogène stochastique, qui peut être résolu avec une méthode par éléments finis tout en conservant dans une large mesure l'hétérogénéité des champs de contrainte et de déformation.L'objectif de ce modèle continu est de représenter statistiquement l'hétérogénéité du champ de contraintes dans un modèle de milieu continu ainsi que dans un modèle granulaire discret. Pour ce faire, les propriétés mécaniques sont représentées à l'aide de champs aléatoires. La présente thèse est divisée en trois parties: (1) la construction du modèle et l'identification des paramètres du matériau continuum (densité marginale de premier ordre, moyenne, variance, modèle de corrélation) ;(2) la propagation des ondes dans une voie ferrée ballastée et (3) l’exploration préliminaire de deux ensembles de données expérimentales. La première partie définit le modèle du continuum à fluctuations aléatoires et identifie les paramètres de notre modèle de continuum sur de petits échantillons cylindriques de ballast discret. Des modèles continus équivalents aux échantillons discrets sont générés et résolus en utilisant la méthode EF, et le champ stochastique utilisé pour fournir les propriétés mécaniques. Un processus d'optimisation est utilisé pour trouver une variance normalisée pour le matériau hétérogène stochastique. La deuxième partie de ce travail se concentre sur la résolution des équations dynamiques sur un modèle à grande échelle d'une voie ferrée ballastée utilisant la méthode des éléments spectraux. L'influence de l'hétérogénéité est mise en évidence et étudiée. En conséquence,des courbes de dispersion sont obtenues. Enfin, la troisième partie présente deux jeux de données distincts de mesures expérimentales sur le matériau de ballast : (1) une boîte de ballast ; (2) un passage de train dans un segment de voie ferrée ballastée.Les courbes de mobilité ont été extraites de l'expérience sur les ballasts. Un problème inverse a été résolu afin d'estimer la vitesse de l'onde homogénéisée et la vitesse de l'onde locale dans le milieu. Les passages de trains enregistrés pour l'analyse de la vibration à moyenne fréquences. / The stronger competition with other means of transportation has increased the demand for performance in the railway industry. One way to achieve higher performance is using accurate numerical models to design/predict railways tracks behaviour. Two classes of numerical models are commonly used to predict the behaviour of these systems: (i) discrete approaches and (ii) continuum approaches. In the former, each grain of the ballast is represented by a rigid body and interacts with its neighbours through nonlinear contact forces using, for example, the nonsmooth contact dynamics method. Due to computational limits, this kind of method can only solve a few meters-length of ballast. The coupling with the soil under the ballast layer and with the sleepers also remains an open problem. In continuum approaches, the ballast is replaced by a homogenized continuum and the classical Finite Element (FE) Method (or similar) is used. However, they are normally used with homogeneous mechanical parameters, so that they do not represent fully the heterogeneity of the strains and stresses within the ballast layer. We investigate in this thesis an alternative approach using a stochastic heterogeneous continuum model, that can be solved with a FElike method while retaining to a large degree the heterogeneity of the stress and strain fields. The objective of this continuous model is to represent statistically the heterogeneity of the stress field in a continuum model as well as in a discrete granular model. To do this, the mechanical properties are represented using random fields. The present thesis is divided into three parts: (1) the construction of the model and the identification of the parameters of the continuum material (first-order marginal density, mean, variance, correlation model, and correlation length); (2) wave propagation in a ballasted railway track. (3) preliminary exploration of two experimental datasets. The first part sets the randomly-fluctuating continuum model and identifies the parameters of our continuum model on small cylindrical samples of discrete ballast. Continuum models equivalent to the discrete samples are generated and solved using the FE method, and the stochastic field used as mechanical properties. An optimization process is used to find a normalized variance for the stochastic heterogeneous material. The second part of this work concentrates on the solution of the dynamical equations on a large-scale model of a ballasted railway track using the Spectral Element Method. The influence of the heterogeneity is highlighted and studied. As a result, dispersion curves are obtained. Finally, the third part presents two distinct datasets of experimental measurements on ballast material: (1) a ballast box; (2) a train passage in a segment of ballasted railway track. Mobility curves were extracted from the ballast box experiment. An inverse problem was solved in order to estimate the homogenized wave velocity and local wave velocity in the medium. The trains pass-by recorded for the analysis of the vibration at medium frequencies.

Motivated by the end of Moore’s law, graphics processing units (GPUs) are replacing general-purpose processors as the main source of computational power in emerging supercomputing architectures. A challenge in systems with GPU accelerators is the cost of transferring data between the host memory and the GPU device memory. On supercomputers, the standard for communication between compute nodes is called Message Passing Interface (MPI). Recently, many MPI implementations support using GPU device memory directly as communication buffers, known as GPU-aware MPI. One of the most computationally demanding applications on supercomputers is high-fidelity simulations of turbulent fluid flow. Improved performance in high-fidelity fluid simulations can enable cases that are intractable today, such as a complete aircraft in flight. In this thesis, we compare the MPI performance with host memory and GPU device memory, and demonstrate how GPU-aware MPI can be used to accelerate high-fidelity incompressible fluid simulations in the spectral element code Neko. On a test system with NVIDIA A100 GPUs, we find that MPI performance is similar using host memory and device memory, except for intra-node messages in the range of 1-64 KB which is significantly slower using device memory, and above 1 MB which is faster using device memory. We also find that the performance of high-fidelity simulations in Neko can be improved by up to 2.59 times by using GPU-aware MPI in the gather–scatter operation, which avoids several transfers between host and device memory. / Motiverat av slutet av Moores lag så har grafikprocessorer (GPU:er) börjat ersätta konventionella processorer som den huvudsakliga källan till beräkningingskraft i superdatorer. En utmaning i system med GPU-acceleratorer är kostnaden att överföra data mellan värdminnet och acceleratorminnet. På superdatorer är Message Passing Interface (MPI) en standard för kommunikation mellan beräkningsnoder. Nyligen stödjer många MPI-implementationer direkt användning av acceleratorminne som kommunikationsbuffertar, vilket kallas GPU-aware MPI. En av de mest beräkningsintensiva applikationerna på superdatorer är nogranna datorsimuleringar av turbulenta flöden. Förbättrad prestanda i nogranna flödesberäkningar kan möjliggöra fall som idag är omöjliga, till exempel ett helt flygplan i luften. I detta examensarbete jämför vi MPI-prestandan med värdminne och acceleratorminne, och demonstrerar hur GPU-aware MPI kan användas för att accelerera nogranna datorsimuleringar av inkompressibla flöden i spektralelementkoden Neko. På ett testsystem med NVIDIA A100 GPU:er finner vi att MPI-prestandan är liknande med värdminne och acceleratorminne. Detta gäller dock inte för meddelanden inom samma beräkningsnod i intervallet 1-64 KB vilka är betydligt långsammare med acceleratorminne, och över 1 MB vilka är betydligt snabbare med acceleratorminne. Vi finner också att prestandan av nogranna datorsimuleringar i Neko kan förbättras upp till 2,59 gånger genom användning av GPU-aware MPI i den så kallade gather– scatter-operationen, vilket undviker flera överföringar mellan värdminne och acceleratorminne.

In computational fluid dynamics, we often need to solve complex problems with high precision and efficiency. We propose a three-pronged approach to attain this goal. First, we use the discontinuous Galerkin spectral element method (DG-SEM) for its high accuracy. Second, we use graphics processing units (GPUs) to perform our computations to exploit available parallel computing power. Third, we implement a parallel adaptive mesh refinement (AMR) algorithm to efficiently use our computing power where it is most needed. We present a GPU DG-SEM solver with AMR and dynamic load balancing for the 2D wave equation. The DG-SEM is a higher-order method that splits a domain into elements and represents the solution within these elements as a truncated series of orthogonal polynomials. This approach combines the geometric flexibility of finite-element methods with the exponential convergence of spectral methods. GPUs provide a massively parallel architecture, achieving a higher throughput than traditional CPUs. They are relatively new as a platform in the scientific community, therefore most algorithms need to be adapted to that new architecture. We perform most of our computations in parallel on multiple GPUs. AMR selectively refines elements in the domain where the error is estimated to be higher than a prescribed tolerance, via two mechanisms: p-refinement increases the polynomial order within elements, and h-refinement splits elements into several smaller ones. This provides a higher accuracy in important flow regions and increases capabilities of modeling complex flows, while saving computing power in other parts of the domain. We use the mortar element method to retain the exponential convergence of high-order methods at the non-conforming interfaces created by AMR. We implement a parallel dynamic load balancing algorithm to even out the load imbalance caused by solving problems in parallel over multiple GPUs with AMR. We implement a space-filling curve-based repartitioning algorithm which ensures good locality and small interfaces. While the intense calculations of the high order approach suit the GPU architecture, programming of the highly dynamic adaptive algorithm on GPUs is the most challenging aspect of this work. The resulting solver is tested on up to 64 GPUs on HPC platforms, where it shows good strong and weak scaling characteristics. Several example problems of increasing complexity are performed, showing a reduction in computation time of up to 3× on GPUs vs CPUs, depending on the loading of the GPUs and other user-defined choices of parameters. AMR is shown to improve computation times by an order of magnitude or more.

The spectral-element method (SEM) is used to study wall-bounded turbulent flowsin moderately complex geometries. The first part of the thesis is devoted to simulations of canonical flow cases, such as temporal K-type transitionand turbulent channel flow, to investigate general resolution requirements and computational efficiency of the numerical code nek5000. Large-eddy simulation (LES) is further performed of a plane asymmetric diffuser flow with an opening angle of 8.5 degrees, featuring turbulent flow separation. Good agreement with numerical studies of Herbst (2007) is obtained, and it is concluded that the use of a high-order method is advantageous for flows featuring pressure-induced separation. Moreover, it is shown, both a priori on simpler model problems and a posteriori using the full Navier--Stokes equations, that the numerical instability associated with SEM at high Reynolds numbers is cured either by employing over-integration (dealiasing) or a filter-based stabilisation, thus rendering simulations of moderate to high Reynolds number flows possible. The second part of the thesis is devoted to the first direct numerical simulation (DNS) of a truly three-dimensional, turbulent and separated diffuser flow at Re = 10 000 (based on bulk velocity and inflow-duct height), experimentally investigated by Cherry et al. (2008). The massively parallel capabilities of the spectral-element method are exploited by running the simulations on up to 32 768 processors. Very good agreement with experimental mean flow data is obtained and it is thus shown that well-resolved simulations of complex turbulent flows with high accuracy are possible at realistic Reynolds numberseven in complicated geometries. An explanation for the discovered asymmetry of the mean separated flow is provided and itis demonstrated that a large-scale quasi-periodic motion is present in the diffuser. In addition, a new diagnostic measure, based on the maximum vorticity stretching component in every spatial point, is designed and tested in a number of turbulent and transitional flows. Finally, Koopman mode decomposition is performed of a minimal channel flow and compared to classical proper orthogonal decomposition (POD). / QC 20111206

In this thesis, simulations are performed to study the motion ofindividual cells in flow, focusing on the hydrodynamics of actively swimming cells likethe self-propelling microorganisms, and of passively advected objects like the red bloodcells. In particular, we develop numerical tools to address the locomotion ofmicroswimmers in viscoelastic fluids and complex geometries, as well as the motion ofdeformable capsules in micro-fluidic flows. For the active movement, the squirmer is used as our model microswimmer. The finiteelement method is employed to study the influence of the viscoelasticity of fluid on theperformance of locomotion. A boundary element method is implemented to study swimmingcells inside a tube. For the passive counterpart, the deformable capsule is chosen as the modelcell. An accelerated boundary integral method code is developed to solve thefluid-structure interaction, and a global spectral method is incorporated to handle theevolving cell surface and its corresponding membrane dynamics. We study the locomotion of a neutral squirmer with anemphasis on the change of swimming kinematics, energetics, and flowdisturbance from Newtonian to viscoelastic fluid. We also examine the dynamics of differentswimming gaits resulting in different patterns of polymer deformation, as well as theirinfluence on the swimming performance. We correlate the change of swimming speed withthe extensional viscosity and that of power consumption with the phase delay of viscoelasticfluids. Moreover, we utilise the boundary element method to simulate the swimming cells in astraight and torus-like bent tube, where the tube radius is a few times the cell radius. Weinvestigate the effect of tube confinement to the swimming speed and power consumption. Weanalyse the motions of squirmers with different gaits, which significantly affect thestability of the motion. Helical trajectories are produced for a neutralsquirmer swimming, in qualitative agreement with experimental observations, which can beexplained by hydrodynamic interactions alone. We perform simulations of a deformable capsule in micro-fluidic flows. We look atthe trajectory and deformation of a capsule through a channel/duct with a corner. Thevelocity of capsule displays an overshoot as passing around the corner, indicating apparentviscoelasticity induced by the interaction between the deformable membrane and viscousflow. A curved corner is found to deform the capsule less than the straight one. In addition, we propose a new cell sorting device based on the deformability of cells. Weintroduce carefully-designed geometric features into the flow to excite thehydrodynamic interactions between the cell and device. This interaction varies andclosely depends on the cell deformability, the resultant difference scatters the cellsonto different trajectories. Our high-fidelity computations show that the new strategy achievesa clear and robust separation of cells. We finally investigate the motion of capsule in awall-bounded oscillating shear flow, to understand the effect of physiological pulsation to thedeformation and lateral migration of cells. We observe the lateral migration velocity of a cellvaries non-monotonically with its deformability. / <p>QC 20140313</p>

Die Dissertation behandelt methodische und algorithmische Aspekte der Spektralelementemethode zur räumlichen Diskretisierung partieller Differentialgleichungen. Die Weiterentwicklung einer symmetriebasierten Faktorisierung ermöglicht effiziente Operatoren für Tetraederelemente. Auf Grundlage einer umfassenden Leistungsanalyse werden Engpässe in der Implementierung der Operatoren identifiziert und durch algorithmische Modifikationen der Methode eliminiert.

Cette thèse mêle observations, simulations et développement d'outils numériques haute performance dans le domaine de l’acoustique sous-marine, et notamment pour l’étude des ondes T. Après une revue de la littérature sur les ondes T, nous avons analysé des données réelles enregistrées en Italie. Afin de modéliser le phénomène nous avons développé un solveur éléments spectraux axisymétriques dans le domaine temporel, que nous présentons et validons. Nous présentons également une étude paramétrique de l'influence de la pente du plancher océanique dans un scénario typique de génération/conversion d'une onde T. L'énergie et la durée de ces ondes s’avère être particulièrement sensible à l'environnement. En particulier nous avons vu que les pentes et les caractéristiques du fond marin jouaient un rôle capital. Nos études confirment qu’aux distances régionales le profil de vitesse dans l'océan s'avère n'être qu'un paramètre de deuxième ordre. Pour en évaluer l’impact nous avons développé une procédure pour le calcul de cartes de perte de transmission et de dispersion à partir de simulations numériques en forme d'onde complète dans le domaine temporel. Dans un second temps nous montrons qu'un bateau commercial de taille moyenne peut créer par diffraction des ondes T d'une d'amplitude conséquente et de faible dispersion. Ce mode de génération d'onde T, encore non documenté, doit être particulièrement fréquent dans les zones où le trafic maritime est important et pourrait expliquer certaines ondes T abyssales encore incomprises. Pour finir, nous présentons des outils numériques pour calculer le champ acoustique créé par une source en mouvement / This thesis combines observations, simulations and development of high performance numerical tools in the field of underwater acoustics, and in particular for the study of T-waves. After a literature review on T-waves, we analysed real data recorded in Italy. In order to model the phenomenon we have developed an axisymmetric spectral element solver in the time domain, which we present and validate. We also present a parametric study of the influence of seafloor slope in a typical scenario of generation / conversion of a T-wave. The energy and duration of these waves is particularly sensitive to the environment. In particular, we have seen that the slopes and characteristics of the seabed are of crucial importance. Our studies confirm that at regional distances the ocean speed profile is only a second order parameter. To evaluate its impact we have developed a procedure for the calculation of transmission and dispersion loss maps from full waveform numerical simulations in the time domain. In a second step we show that a medium-sized commercial boat can create T-waves of a significant amplitude and of low dispersion by diffraction. This T-wave generation mode, still undocumented, must be particularly frequent in areas where maritime traffic is dense and could explain some abyssal T-waves still misunderstood. Finally, we present numerical tools for calculating the acoustic field created by a moving source

Tato diplomová práce prezentuje metodu spektrálních prvků. Tato metoda je použita k řešení stacionárního 2-D laminárního proudění Newtonovské nestlačitelné tekutiny. Proudění je popsáno stacionarní Navier-Stokesovou rovnicí. Dohromady s okrajovou pod- mínkou tvoří Navier-Stokesův problém. Na slabou formulaci této úlohy je aplikována metoda spektrálních prvků. Touto discretizací se získá soustava nelineárních rovnic. K obrdžení lineární soustavy je použita Newtonova iterační metoda. Podorobný algorit- mus tvoří jádro Navier-Stokeseva solveru, který je naprogramován v Matlabu. Na závěr jsou pomocí tohoto solveru řešeny dva příklady: proudění v kavitě a obtékání válce. Přík- lady jsou řešeny pro různé Reynoldsovy čísla. První od 1 do 1000 a druhý od 1 do 100.

High-frequency wave phenomena is observed in many physical settings, most notably in acoustics, electromagnetics, and elasticity. In all of these fields, numerical simulation and modeling of the forward propagation problem is important to the design and analysis of many systems; a few examples which rely on these computations are the development of metamaterial technologies and geophysical prospecting for natural resources. There are two modes of modeling the forward problem: the frequency domain and the time domain. As the title states, this work is concerned with the former regime. The difficulties of solving the high-frequency wave propagation problem accurately lies in the large number of degrees of freedom required. Conventional wisdom in the computational electromagnetics commmunity suggests that about 10 degrees of freedom per wavelength be used in each coordinate direction to resolve each oscillation. If K is the width of the domain in wavelengths, the number of unknowns N grows at least by O(K^2) for surface discretizations and O(K^3) for volume discretizations in 3D. The memory requirements and asymptotic complexity estimates of direct algorithms such as the multifrontal method are too costly for such problems. Thus, iterative solvers must be used. In this dissertation, I will present fast algorithms which, in conjunction with GMRES, allow the solution of the forward problem in O(N) or O(N log N) time. / text

Ziel der vorliegenden Arbeit ist es, Rühr- und Mischungsvorgänge in Flüssigmetallströmungen zu untersuchen, die mittels rotierender und wandernder Magnetfelder bzw. deren Kombination induziert werden. Im Mittelpunkt steht dabei die Charakterisierung der dreidimensionalen Strömungsstrukturen innerhalb zylindrischer Geometrien bei der Verwendung überkritischer Magnetfelder. Neben der Untersuchung der Strömungseigenschaften stellen die physikalische Modellierung der angreifenden Kräfte, die geeignete Wahl und Validierung eines effizienten numerischen Lösungsverfahrens und dessen Erweiterung für die Durchführung von Large Eddy Simulationen wesentliche Eckpfeiler dieser Arbeit dar.