A new hybrid approach to model high Reynolds number wall-bounded turbulent flows is developed based on coupling the two-level simulation (TLS) approach in the inner region with conventional large eddy simulation (LES) away from the wall. This new approach is significantly different from previous near-wall approaches for LES. In this hybrid TLS-LES approach, a very fine small-scale (SS) mesh is embedded inside the coarse LES mesh in the near-wall region. The SS equations capture fine-scale temporal and spatial variations in all three cartesian directions for all three velocity components near the wall. The TLS-LES equations are derived based on defining a new scale separation operator. The TLS-LES equations in the transition region are obtained by blending the TLS large-scale and LES equations. A new incompressible parallel flow solver is developed that accurately and reliably predicts turbulent flows using TLS-LES. The solver uses a primitive variable formulation based on an artificial compressibility approach and a dual time stepping method. The advective terms are discretized using fourth-order energy conservative finite differences. The SS equations are also integrated in parallel, which reduces the overall cost of the TLS-LES approach. The TLS-LES approach is validated and investigated for canonical channel flows, channel flow with adverse pressure gradient and asymmetric plane diffuser flow. The results suggest that the TLS-LES approach yields very reasonable predictions of most of the crucial flow features in spite of using relatively coarse grids.

L'impact environnemental du trafic aérien fait maintenant l'objet d'une réglementation qui tend à se sévériser. Dans ce contexte, les industriels misent sur l'amélioration des technologies afin de réduire la consommation de carburant et l'émission de polluants. Ces phénomènes dépendent en grande partie des chemins réactionnels sous-jacents, qui peuvent s'avérer très complexes. La Simulation aux Grandes Échelles (SGE) est un outil intéressant afin d'étudier ces phénomènes pour un coût de calcul qui reste raisonnable. Cependant, les processus chimiques, s'ils sont considérés sans simplification, font intervenir des centaines d'espèces aux temps caractéristiques très différents au sein de processus non-linéaires qui induisent une forte raideur dans le système d'équations, et un coût de calcul prohibitif. Permettant de s'absoudre de ces problèmes tout en conservant une bonne capacité de prédiction des polluants, les Chimies Analytiquement Réduites (CAR) font l'objet d'une attention grandissante au sein de la communauté. Les CAR permettent de conserver la physique du problème considéré, en conservant les espèces et voies réactionnelles les plus importantes. Grâce à l'évolution toujours croissante des moyens de calculs, les CAR sont appliqués dans des configurations de plus en plus complexes. Les travaux de thèse ont principalement portés sur deux sujets. Premièrement, une étude poussée des techniques et outils permettant une réduction efficace et systématique de chimies détaillées. L'outil de réduction multiétapes YARC est retenu et exhaustivement employé dans la dérivation et la validation d'une série de CAR préservant la description de la structure de flamme. Ensuite, une investigation de la faisabilité et des bénéfices qu'apportent l'utilisation de CAR en LES, comparé a des approches plus classiques, sur des cas tests de complexité croissante. La première configuration étudiée est une chambre de combustion partiellement pré-mélangée brûlant de l'éthylène, étudiée expérimentalement au DLR. Différentes modélisations de la chimie sont considérées, dont un CAR développé spécifiquement pour ce cas test, et les résultats démontrent qu'une prise en compte des interactions flamme-écoulement est cruciale pour une prédiction juste de la structure de la flamme et des niveaux de suies. La seconde configuration est un brûleur diphasique, avec une injection directe pauvre, brûlant du Jet-A2. Dans cette étude, une approche novatrice pour la prise en compte de la complexité du fuel réel (HyChem) est considérée, permettant la dérivation d’un CAR. Les résultats sont excellents et valident la méthodologie tout en fournissant une analyse précieuse des interactions flamme-spray et de la formation de polluants (NOx) dans des flammes à la structure complexe. / Recent implementation of emission control regulations has resulted in a considerable demand from industry to improve the efficiency while minimizing the consumption and pollutant emissions of the next generation of aero-engine combustors. Those phenomena are shown to strongly depend upon the underlying complex chemical pathways and their interaction with turbulence. Large Eddy Simulation (LES) is an attractive tool to address those issues with high accuracy at a reasonable computing cost. However, the computation of accurate combustion chemistry remains a challenge. Indeed, combustion proceeds through complex and highly non-linear processes that involve up to hundreds of different chemical compounds, which significantly increases the computational time and often induces stiffness in the resolved equations. As a mean to circumvent these drawbacks while retaining the necessary kinetics for the prediction of pollutants, Analytically Reduced Chemistry (ARC) has recently received high interest in the Computational Fluid Dynamics (CFD) community. ARC is a strategy for the description of combustion chemistry where only the most important species and reactions are retained, in a "physically-oriented way". ARC is on the verge of becoming affordable at a design stage, thanks to the continuously increasing available computational resources. The goal of the present work is twofold. A first objective is to test and validate efficient techniques and tools by which detailed chemistries are reduced to an LES-compliant format. To do so, the multi-step reduction tool YARC is selected and employed to derive and validate a series of ARC specifically designed to retrieve correct flame structures. A second objective is to investigate the overall feasibility and benefits of using ARC, combined to the Thickened Flame model (DTFLES), in performing LES of configurations of increasing complexity. The first configuration is a sooting swirl-stabilized non-premixed aero-engine combustor experimentally studied at DLR, burning ethylene. LES of this configuration is performed with the AVBP solver, in which ARC has been implemented. By comparison with global chemistry and tabulated chemistry, results highlight the importance of accurately capturing the flow-flame interactions for a good prediction of pollutants and soot. The second configuration is a swirled twophase flow burner featuring a lean direct injection system and burning Jet-A2. A novel methodology to real fuel modeling (HyChem approach) is employed, which allows subsequent ARC derivation. The excellent results in comparison with measurements constitute an additional validation of the methodology, and provide valuable qualitative and quantitative insights on the flame-spray interactions and on the pollutant formation (NOx) mechanisms in complex flame configurations.

Flow induced self-oscillations cause acoustic pressure oscillations of large amplitude in pipe flows. If Reynolds number is treated as a parameter, these floinduced oscillations occur only at discrete and critical values of Reynolds number. However, for a small range of Reynolds numbers around such a critical value, such periodic oscillations may appear intermittently. If intermittency, which is a precursor to these self-oscillations, can be detected, prediction of an impending instability may be possible. In experiments done by Vineeth and Sujith (Int. J. Aeroacoustics, 2016) on flow in a duct orifice arrangement, where flow enters through the duct inlet, and leaves into the atmosphere through the orifice exit, “whistling” was observed at a Reynolds number of 4200 (based on the orifice thickness and flow speed within the orifice), where large amplitude pressure oscillations were observed. At slightly lower Reynolds numbers, bursts of relatively smaller amplitudes of pressure oscillations were observed to appear intermittently. For a similar configuration, Large Eddy Simulations (LES) have been carried out with explicit filtering as a sub­ grid scale model here. Both whistling and intermittency are observed in the simulations. As air flows from the duct into the orifice, it turns sharply around the corner at the duct­ orifice interface. Due to this sharp turn, flow separation occurs, and hence, a shear layer is formed at the mouth of the orifice. The mechanism of whistling is found to be this shear layer within the orifice flapping about and hitting the trailing edge of the orifice periodically, thus causing the shear layer to break and roll up into a vortex. At Reynolds numbers where intermittency is observed, the shear layer is found to very mildly come in contact with the edges of the orifice walls, thus disturbing it. In the simulations, time series data of pressure are recorded at various probe locations. In a given time series, if scale invariance behaviour exists, it can be quantified by measuring the Hurst exponent. The numerical value of the Hurst exponent is an index of “long range or short range dependence” in a time series. Hurst exponent is measured in the time series data obtained. It is found to drop to zero as the flow approaches the state of a self-sustained oscillation, since the growth rates of all the long term as well as short term trends in the time series vanish. A loss of multifractality in the time series is also observed as the flow approaches whistling. As a part of the this thesis, new, split high resolution schemes of high order are designed following the Hixon Turmel Proposal.

When an electrically conducting fluid moves through a magnetic field, fluid mechanics and electromagnetism are coupled.<p>This interaction is the object of magnetohydrodynamics, a discipline which covers a wide range of applications, from electromagnetic processing to plasma- and astro-physics.<p><p>In this dissertation, the attention is restricted to turbulent liquid metal flows, typically encountered in steel and aluminium industries. Velocity measurements in such flows are extremely challenging because liquid metals are opaque, hot and often corrosive. Therefore, non-intrusive measurement devices are essential. One of them is the Lorentz force flowmeter. Its working principle is based on the generation of a force acting on a charge, which moves in a magnetic field. Recent studies have demonstrated that this technique can measure efficiently the mean velocity of a liquid metal. In the existing devices, however, the measurement depends on the electrical conductivity of the fluid. <p><p>In this work, a novel version of this technique is developed in order to obtain measurements that are independent of the electrical conductivity. This is particularly appealing for metallurgical applications, where the conductivity often fluctuates in time and space. The study is entirely numerical and uses a flexible computational method, suitable for industrial flows. In this framework, the cost of numerical simulations increases drastically with the level of turbulence and the geometry complexity. Therefore, the simulations are commonly unresolved. Large eddy simulations are then very promising, since they introduce a subgrid model to mimic the dynamics of the unresolved turbulent eddies. <p><p>The first part of this dissertation focuses on the quality and reliability of unresolved numerical simulations. The attention is drawn on the ambiguity that may arise when interpretating the results. Owing to coarse resolutions, numerical errors affect the performances of the discrete model, which in turn looses its physical meaning. In this work, a novel implementation of the turbulent strain rate appearing in the models is proposed. As opposed to its usual discretisation, the present strain rate is in accordance with the discrete equations of motion. Two types of flow are considered: decaying turbulence located far from boundaries, and turbulent flows between two parallel and infinite walls. Particular attention is given to the balance of resolved kinetic energy, in order to assess the role of the model.<p><p>The second part of this dissertation deals with a novel version of Lorentz force flowmeters, consisting in one or two coils placed around a circular pipe. The forces acting on each coil are recorded in time as the liquid metal flows through the pipe. It is highlighted that the auto- or cross-correlation of these forces can be used to determine the flowrate. The reliability of the flowmeter is first investigated with a synthetic velocity profile associated to a single vortex ring, which is convected at a constant speed. This configuration is similar to the movement of a solid rod and enables a simple analysis of the flowmeter. Then, the flowmeter is applied to a realistic three-dimensional turbulent flow. In both cases, the influence of the geometrical parameters of the coils is systematically assessed. / Doctorat en Sciences / info:eu-repo/semantics/nonPublished

A l’heure actuelle, seuls des facteurs locaux, stationnels, sont considérés pour le calcul des risques liés au vent alors que le vent qui aborde un peuplement forestier est affecté par les surfaces sur lesquelles il vient de passer ; les lisières en particulier jouent un rôle important sur l’écoulement atmosphérique, en contribuant à générer de la turbulence. Dans un paysage fragmenté, constitué d’une mosaïque de surfaces de différentes hauteurs et rugosités, la multiplicité des lisières est ainsi susceptible d’avoir des effets cumulatifs perceptibles à l’échelle régionale, qui pourraient contribuer de manière significative à la fragilité des massifs face à des tempêtes. Certains niveaux de fragmentation semblent susceptibles de conduire à un accroissement des risques en cas de vent violent. Bien que la région de lisière a été étudiée de manière approfondie dans le passé en raison de leur importance pour la détermination des vitesses de vent, des niveaux de turbulence et des échanges entre l’atmosphère et la canopée, il n’y a aucune étude de l’impact de lisières multiples ou de la fragmentation des forêts sur les caractéristiques de la couche limite à l’échelle du paysage. Quelques rares études laissent penser que la fragmentation du paysage pourrait moduler de manière significative la structure turbulente de la couche limite atmosphérique mais ces études concernent des réseaux de brisevents plutôt qu’un ensemble de parcelles forestières. On cherche par conséquent à caractériser les champs de vent et de turbulence pour ces différentes configurations. Pour ce faire, une expérimentation en soufflerie à été réalisée, visant à caractériser l’écoulement sur des maquettes de paysage présentant cinq degrés de fragmentation (L = ~ 5, ~ 10, ~15, ~20, ~30h, où L est la distance entre deux patchs de forêts régulièrement espacés et h est la hauteur de la canopée). Un cas homogène a également été simulé et sert ici de référence. Pour le modèle de canopée choisi, ces expérimentations montrent que l’énergie cinétique turbulente présente dans la basse atmosphère ne passe pas par un maximum pour une valeur de l’espacement intermédiaire comme il était supposé à l’origine. Le cas homogène est la configuration la plus rugueuse. Pour de grands espacements l’influence d’une parcelle ne se fait guère sentir sur la suivante et lorsqu’ils sont faibles l’écoulement ne "ressent" guère les clairières et présente des caractéristiques semblables au cas homogène. Nous avons également évalué un modèle atmosphérique de type "simulation des grandes échelles" à l’aide des données présentées précédemment. Le modèle est capable de reproduire les grandes caractéristiques de la turbulence telles que les vitesses de vent horizontales et verticales, l’énergie cinétique turbulente, les contraintes de Reynolds et les coefficients d’asymétrie horizontale et verticale en tous points du domaine. Cela nous a permis de confirmer la validité des calculs numériques et de simuler l’écoulement sur une plus large gamme de paysages fragmentés. Les résultats démontrent l’importance de l’indice foliaire pour le calcul de la rugosité effective sur une succession de patchs de forêt. / At present only the characteristics of a forest stand and its immediate environment are taken into account in calculating forest wind risk. However, it is known that the wind is strongly affected by the surfaces over which it has previously flowed. Forest edges in particular play an important role in determining the characteristics of the atmospheric flow by generating increased turbulence, triggering the formation of coherent tree scale structures. In a fragmented landscape, consisting of surfaces of different heights and roughness, the multiplicity of edges may have cumulative effects at the regional scale leading to increased forest damage during storms. Flow changes in the atmospheric boundary-layer across surface roughness changes have received extensive study in the past because of their importance in determining velocities, turbulence levels and exchange between the atmosphere and biosphere or ground. There have also been a number of studies across single forest edges both in the field, wind-tunnels and computer models. However, there have been no studies of flow across multiple forest edges or the impact of forest fragmentation on the characteristics of the boundary-layer. The only studies on multiple surface changes have been wind-tunnel examination of the flow though and across multiple wind-breaks. In this thesis we show results from a series of wind tunnel experiments on a range of levels of forest fragmentation. Five gap spacings (L = ~ 5, ~ 10, ~15, ~20, ~30h, where L is the length of the gap and h is the canopy height) were investigated using 3D laser doppler velocimetry in order to assess the effects of fragmented landscapes on mean and turbulent wind characteristics. The fragmentation was two-dimensional with the transition between forest and gaps only being along the wind direction and the forest and gaps were continuous perpendicular to the wind direction. The wind speeds and turbulence characteristics are compared against measurements from a single forest edge in the wind tunnel, which acts as a reference. No enhancement of turbulence formation at a particular level of fragmentation was observed but there was a consistent pattern of wind speed and turbulence back from the first edge of each simulation with the horizontal velocity at tree top increasing and the turbulent kinetic energy decreasing as gap size increased. We also compare mean wind speeds (U and W) and turbulence characteristics (variance in u, v, and w; skewness in U, V, and W; Reynold’s stress, and TKE) at all points in the experimental measurement domain of the wind tunnel with Large Eddy Simulation (LES) results, which allows us to confirm the validity of the LES calculations and to conduct a wider range of experiments than was possible in the wind-tunnel. The results demonstrate the importance of the frontal area index or roughness density of elements (in this case trees) in determining the nature of the flow and the effective roughness of the overall surface. They also show that as the gaps between forest blocks increases the flow transitions (at a gap size between 10 to 15 tree heights) from flow comparable to that over a continuous forest to flow across a set of isolated forest blocks.

Current climate models often predict fractional cloud cover on the basis of a diagnostic probability density function (PDF) describing the subgrid-scale variability of the total water specific humidity, qt, favouring schemes with limited complexity. Standard shapes are uniform or triangular PDFs the width of which is assumed to scale with the gridbox mean qt or the grid-box mean saturation specific humidity, qs. In this study, the qt variability is analysed from large-eddy simulations for two stratocumulus, two shallow cumulus, and one deep convective cases. We find that in most cases, triangles are a better approximation to the simulated PDFs than uniform distributions. In two of the 24 slices examined, the actual distributions were so strongly skewed that the simple symmetric shapes could not capture the PDF at all. The distribution width for either shape scales acceptably well with both the mean value of qt and qs, the former being a slightly better choice. The qt variance is underestimated by the fitted PDFs, but overestimated by the existing parameterisations. While the cloud fraction is in general relatively well diagnosed from fitted or parameterised uniform or triangular PDFs, it fails to capture cases with small partial cloudiness, and in 10 – 30% of the cases misdiagnoses clouds in clear skies or vice-versa. The results suggest choosing a parameterisation with a triangular shape, where the distribution width would scale with the grid-box mean qt using a scaling factor of 0.076. This, however, is subject to the caveat that the reference simulations examined here were partly for rather small domains and driven by idealised boundary conditions.

<p dir="ltr">Grid convergence and scaling analyses have not been done rigorously for practical large-eddy simulations (LES). The challenge arises from the fact that there are two grid-related length scales: grid size and LES filter width. It causes the numerical and model errors in LES to be inherently coupled, making the convergence of either error difficult to analyze. This study works to overcome the challenge by developing scaling laws that can be used to guide the convergence analysis of errors in LES. Three different convergence cases are considered, and their respective scaling laws are developed by varying the ratio between grid size and filter width. A pool fire is adopted as a test case for the convergence analysis of LES. Qualitative and quantitative assessments of the LES results are made first to ensure reliable numerical solutions. In the subsequent scaling analysis, it is found that the results are consistent with their respective scaling laws. The results provide strong support to the developed scaling laws. The work is significant as it proposes a rigorous way to guide convergence analysis of LES errors. In a world where LES already has a wide range of applicability and is still becoming more prominent, it is imperative to have a thorough understanding of how it works including its convergence and scaling laws with respect to the change of grid size and filter width.</p>

More and larger wind farms are planned offshore. As the most suitable build sites are limited wind farms will be constructed near to each other in so called wind farm clusters. Behind the wind turbines in these farms there is a disrupted flow of air called a wake that is characterized by reduced wind speed and increased turbulence. These individual turbine wakes combine to form a farm wake that can travel a long distance. In wind farm clusters farm to farm interaction will occur, i.e. the long distance wake from one wind farm will impact the wind conditions for other farms in the surrounding area. The thesis contains numerical studies of these long distance wakes. In this study Large Eddy Simulations (LES) using an Actuator Disc method (ACD) are used. A prescribed boundary layer is used where the wind shear is introduced using body forces. The turbulence, based on the Mann model, is introduced as fluctuating body forces upstream of the farm. A neutral atmosphere is assumed. The applied method has earlier been used for studies of wake effects inside farms but not for the longer distances needed for farm to farm interaction. Numerical studies are performed to get better knowledge about the use of this model for long distance wakes. The first study compares the simulation results with measurements behind an existing farm. Three parameter studies are thereafter setup to analyze how to best use the model. The first parameter study examines numerical and physical parameters in the model. The second one looks at the extension of the domain and turbulence as well as the characteristics of the flow far downstream. The third one gathers information on the downstream development of turbulence with different combinations of wind shear and turbulence level. The impact of placing the turbines at different distances from the turbulence plane is also studied. Finally a second study of an existing wind farm is performed and compared with a mesoscale model. The model is shown to be relevant also for studies of long distance wakes. Combining LES with a mesoscale model can be of interest.

La régulation des polluants a mené à la création de nouveaux systèmes de combustion. Le carburant étant stocké sous forme liquide, sa transformation jusqu’à sa combustion est complexe. La capacité de la Simulation aux grandes échelles à simuler des écoulements turbulents réactifs a été montrée sur des cas académiques comme sur des configurations industrielles, tout en prenant en compte les phénomènes multiphysiques intervenant dans ces configurations, mais les études sur la structure de flamme diphasique sont encore trop peu nombreuses. La présence de deux solveurs pour la simulation d’une phase liquide étant disponible dans le code AVBP, leur utilisation permet une comparaison et une compréhension des phénomènes en jeu combinant dispersion, évaporation, et combustion. La première partie de l’étude relate la validation du modèle d’injection FIM-UR. Ce modèle est capable de reconstruire les profils de vitesses et de granulométrie à l’injecteur sans avoir à simuler les phénomènes d’atomisation primaire et secondaire. Une validation en régime turbulent avait déjà été réalisée, et on propose ici de valider le modèle dans un cas laminaire. Des comparaisons entre simulations monodisperses et polydisperse et des expériences sont effectuées. La simulation monodisperse Lagrangienne donne une bonne structure globale mais la simulation polydisperse Lagrangienne permet de retrouver le comportement au centre du cône avec la présence des petites gouttes et à la périphérie du cône par la présence des grosses gouttes. De plus, des améliorations sont apportées au modèle pour le formalisme Eulérien et montrent de bons résultats. La partie suivante s’intéresse à caractériser un spray polydisperse par une distribution monodisperse. En effet, au cas où une approche polydisperse n’est pas possible, le choix du diamètre moyen à prendre pour une simulation monodisperse est délicat. On propose donc d’analyser le comportement d’un spray polydisperse en le comparant à ceux de sprays monodisperses. Deux configurations académiques sont choisies : des cas de Turbulence Homogène Isotrope chargée en particules pour étudier la dynamique, et des calculs d’évaporation 0D. Trois paramètres sont étudiés pour la dynamique : la concentration préférentielle (ou ségrégation), la traînée moyenne et la traînée réduite moyenne. Cette dernière et la ségrégation de la distribution polydisperse semblent affectées par les tailles de goutte les plus faibles, et la concentration préférentielle apparait alors comme la moyenne des ségrégations des classes qui la composent pondérées par l’inverse du nombre de Stokes associé à chacune de ces classes. La traînée moyenne de la simulation polydisperse possède un comportement proche des diamètres moyens D10 et D20. Ces analyses nous poussent donc à choisir le D10 pour caractériser la dynamique d’un spray polydisperse. Les calculs d’évaporation 0D ne permettent pas dans un premier temps de caractériser efficacement la masse évaporée d’un spray polydisperse par celle d’un spray monodisperse équivalent, mais la définition de nouveaux diamètres issus de la littérature des lits fluidisés comme le D50% le permet, ce qui le place autour du D32. On propose donc de caractériser l’évaporation d’un spray polydisperse par ce diamètre. Enfin, la dernière partie étudie la structure de flamme diphasique dans la chambre MERCATO, à l’aide du formalisme Lagrangien, monodisperse et polydisperse, mais aussi en utilisant le formalisme Eulérien. La validation du modèle FIM-UR du premier chapitre et ses améliorations sont utilisées pour représenter les conditions d’injection liquide. En plus d’un calcul polydisperse, deux simulations monodisperses Lagrangiennes sont réalisées en prenant les diamètres moyens D10 et D32, suite à la partie précédente. Des comparaisons qualitatives et des validations sont réalisées, en comparant des profils de vitesses gazeuses axiale et fluctuante et vitesse axiale liquide issus de l’expérience. / Regulations on pollutants have led to the creation of new combustion systems. Giving that fuel is stored in a liquid form, its evolution until combustion is complex. The ability of Large Eddy Simulation has been demonstrated on academic cases, as well as on industrial configurations, by taking into account the multi-physics phenomena, but there is a lack of studies about two-phase flow flame structures. Two solvers for the simulation of two-phase flows are available in the AVBP code, hence both simulations are performed to compare and increase understanding of the phenomena involved such as dispersion, evaporation and combustion. The first part of the study focuses on the validation of the FIM-UR injection model. This model is able to build velocity and droplet profiles at the injector, without simulating primary and secondary break up. A validation in a turbulent case has already been done, and this study validates the model in a laminar case. Comparisons between monodisperse and polydisperse simulations, and experiments are performed. The monodisperse Lagrangian simulation shows good results but the polydisperse simulation is able to represent profiles in the center of the cone by small droplets and at the peripheral part of the cone, by big ones. Moreover, improvements in the Eulerian model exhibit good results. The next section tries to evaluate the impact of polydispersion. Indeed, when a polydisperse approach is not available, choosing the mean diameter can be tricky. A comparison between the behavior of polydisperse spray and monodisperse sprays ones is realised. Two academic cases are studied: Homogeneous Isotropic Turbulence with particles to analyze the dynamics, and 0D evaporation cases. For the dynamics, preferential concentration, mean drag and reduced mean drag are studied. The latter and preferential concentration are affected by small droplets, and the preferential concentration of a polydisperse spray is equivalent to the average of preferential concentration of classes, extracted from the polydisperse distribution, weighted by the inverse of the Stokes number of each class. The mean drag behaves like the D10 and D20 mean drags. This analysis allows us to choose the D10 to characterize a polydisperse distribution for the dynamics. Zero-D evaporation simulations cannot characterize the polydisperse spray evaporated mass by the evaporated mass of monodisperses sprays. New definitions of diameters from fluidized bed literature enable the use of D50%, which is close to D32. We propose to use this diameter to characterize the evaporation of a polydisperse spray. Finally, the last section studies the structure of two-phase flames in the MERCATO bench, using the Lagrangian formalism, monodisperse and polydisperse but also using the Eulerian formalism. The validation of FIM-UR model and improvements from the first section are used to represent liquid injection conditions. A polydisperse simulation is realized and two monodisperse simulations are computed using mean diameters D10 and D32, thanks to the previous section. Qualitative comparisons and validations are realized, comparing gaseous velocity profiles and liquid velocity profiles. Good agreements are found and the mean diameter D32 seems to be close to the polydisperse spray. A comparison between mean flames is done with an Abel transform of the flame from the experiments. The flame has an "M shape", anchored by small recirculation zones out of the swirler, and by a point at the tip of the central recirculation zone. Then, the impact of droplet distributions is analyzed. Even if few bigger droplets from the polydisperse distribution are convected in the hot gases due to bigger particular time and evaporation time, two-phase flow flame structures are equivalent. Different combustion regimes appeared with premixed flames and pockets of fuel burning in the hot gases.

The phenomenon of spray in crossflow is of relevance in gas turbine combustor development. The current work focuses on spray in crossflow rather than liquid jet in crossflow from the standpoint of enhancing fuel dispersion and mixing. Specifically, the first part of the work involves study of spray structure, droplet sizing, and velocimetry for sprays of water and ethanol in a crossflow under ambient conditions. Laser-based diagnostic techniques such as Particle/Droplet Image Analysis (PDIA) and Particle Tracking Velocimetry (PTV) are utilized. Using spray structure images, trajectory equations are derived by multi-variable regression. It is found that the spray trajectory depends only on the two-phase momentum ratio and is independent of other flow parameters. A generalized correlation for the spray trajectory is proposed incorporating the liquid surface tension, which is found to be effective for our data, with water and ethanol, as well as data on Jet-A from the literature for a wide variety of operating conditions. An interesting phenomenon of spatial bifurcation of the spray is observed at low Gas-to-Liquid ratios (GLRs). The reason for this phenomenon is attributed to the co-existence of large and highly deformed ligaments along with much smaller droplets at low GLR conditions. The smaller droplets lose their vertical momentum rapidly leading to lower penetration, whereas the larger ligaments/droplets penetrate much more due to their larger momentum leading to a spatial separation of the two streams. The second part of the study focuses on evaporating sprays in preheated crossflow. Experiments are conducted using ethanol, decane, Jet-A1 fuel, and a two-component surrogate for Jet-A1 fuel. The crossflow air is heated up to 418 K and the effect of evaporation is studied on spray trajectory and droplet sizes. Measured droplet sizes and velocities at two successive locations are used to estimate droplet evaporation lifetimes. Evaporation constant for the d2 law derived from the droplet lifetimes represents the first-ever data for the above-mentioned liquids under forced convective conditions. This data can be used to validate multi-component droplet evaporation models. The last part of the study focuses on Large Eddy Simulations (LES) of the spray in crossflow. The near-nozzle spray structure is investigated experimentally to obtain droplet size and velocity distributions that are used as inputs to the computational model. For the spray in crossflow under ambient conditions, trajectory and droplet sizes at different locations are compared with experimental results. While the predicted trajectory is found to be in good agreement with data, the predicted droplet sizes are larger than the measured values. This is attributed to the implicit assumption in the secondary breakup model that the droplets are spherical, whereas the experimental data in the near-nozzle region clearly shows presence of mostly ligaments and non-spherical droplets, especially for the low GLR cases. A modified breakup model is found to lead to improved agreement in droplet sizes between predictions and measurements. Overall, the experiments and computations have provided significant insight into spray in crossflow phenomenon, and have yielded useful results in terms of validated spray trajectory correlations, droplet evaporation lifetimes under forced convective conditions, and a methodology for simulation of airblast sprays.