Multiphase flows are a common phenomenon. Rains, sediment transport in rivers, snow and dust storms, mud slides and avalanches are examples of multiphase flows occurring in nature. Blood flow is an example of multiphase flow in the human body, which is of vital importance for survival. Multiphase flows occur widely in industrial applications from hydrocarbon extrac-tion to fuel combustion in engines, from spray painting to spray drying, evaporators, pumps and pneumatic conveying. Predicting multiphase flows is of vital importance to understand natural phenomenon and to design and improve industrial processes. Separated flows and dispersed flows are two types of multiphase flows, which occur together in many industrial applications. Physical features of these two classes are different and the transition from one to another involves complex flow physics. Experimental studies of multiphase flows are not easy, as most real world phenomenon cannot be scaled down to laboratory models. Even for those phenomenon that can be demonstrated at lab-oratory scale, rescaling to real world applications requires mathematical models. There are many challenges in experimental measurements of multiphase flows as well. Measurement techniques well suited for single phase flows have constraints when measuring multiphase phenomenon. Un-certainty in experimental measurements poses considerable difficulties in validating numerical models developed for predicting these flows. Owing to the computational effort required, direct simulation of multiphase flows, even for small scale real world applications is out of present scope. Numerical methods have been developed for dealing with each class of flow separately, that in-volves use of models for phenomenon that is computationally demanding. Reynolds Averaged Navier-Stokes (RANS) methods for predicting multiphase flows place strong requirements on turbulence models, as information about fluctuating quantities in the field, that have significant effects on dispersed phase, is not available. Large Eddy Simulation (LES) gives better predictions than RANS as the instantaneous field data is available and large scale unsteadiness that effects the dispersed phase can be captured. Recent LES studies of multiphase flows showed that the sub-grid-scale (SGS) model used for the continuous phase has an effect on the evolution of the dispersed phase. In this work, LES of multiphase flows is performed using Explicit Filtering Large Eddy Sim-ulation method. In this method, spatial derivatives are computed using higher order compact schemes that have spectral-like resolution. SGS modeling is provided by the use of a filter with smoothly falling transfer function. This method is mathematically consistent and converges to a DNS as the grid is refined. It has been successfully applied to combustion and aero-acoustics and this work is the first application of the method to multiphase flows. Study of dispersed multiphase flows was carried out in this work. Modeling of the dispersed phase is kept simple since the in-tention was to evaluate the capability of explicit filtering LES method in predicting multiphase flows. Continuous phase is solved using a compressible formulation with explicit filtering method. Spatial derivatives are computed using fourth and sixth order compact schemes that use derivative splitting method proposed by Hixon & Turkel (2000a) and second order Runge-Kutta (RK2) time stepping. The grid is stretched as needed. Non-reflecting boundary conditions due to Poinsot & Lele (1992) are used to avoid acoustic reflections from boundaries. Buffer zones (Bogey & Bailly (2002)) are employed at outflow and lateral boundaries to damp vortical structures. The code developed for continuous phase is evaluated by studying round jets at Re =36,000 and comparing with experimental measurements of Hussein et al. (1994) and Panchapakesan & Lumley (1993). Simulations showed excellent agreement with experimental results. Rate of decay of axial velocity and the evolution of turbulence intensities on the centerline matched very well with measurements. Radial profiles of mean and fluctuating components of velocities exhibit self-similarity. A set of studies were then performed using this code to assess the effect of numerical scheme, grid refinement & stretching and simulation times on the predictions. Results from these simulations showed good agreements with experiments and established the code for use in multiphase flows under various simulation conditions. To assess the prediction of multiphase flows using this LES method, an evaporating spray ex-periment by Chen et al. (2006) was simulated. The experiment uses a nebuliser for generating a finely atomized spray of acetone, which avoids complex breakdown phenomenon associated with air blast atomizers and provides well defined boundary conditions for model evaluation. The neb-uliser sits upstream in a pipe carrying air and droplets travel along with air for a distance of 10 diameters before exiting into a wind tunnel with co-flowing air. Droplet breakdown, if any, takes place inside the pipe and the spray is finely atomized by the time it reaches pipe exit. One of the experimental cases at Re =31,600, with a mass loading of 1.1% and a jet velocity of 56 m/s is simulated. Particle size has a χsquared distribution with a Sauter mean diameter of 18µm. In the self-similar region, decay of centerline velocity and turbulence intensities matched well with ex-perimental results. Continuous phase exhibits self-similar behavior. A series of simulations were then performed to match the initial region of the spray by altering the inflow conditions in the sim-ulation. Simulation that matched the breakdown location of the experiment revealed the presence of a relaxation zone with a higher initial spreading rate, followed by a lower asymptotic spreading rate. Studies were performed to understand the effect of various phenomenon like evaporation and droplet size on this behavior. A study of breakdown region of particle-laden jets was performed to understand the presence of relaxation zone post breakdown. Flow conditions were similar to evaporating spray experiment except that particles do not evaporate, mass loading is 2% and jet Reynolds number Re =2000. A series of grid refinements were performed and on the largest grid, gird spacing Δy =7.5η, where ηis an estimate of the Kolmogorov length scale based on flow conditions. Decay of axial velocity on the centerline showed variations with grid refinement, tending to the experimentally measured value as the grid is refined. Variation of turbulence intensities along the centerline revealed a jump in axial velocity fluctuations at the breakdown location, while radial and azimuthal velocities showed a smooth increase to their asymptotic value. This jump was resolved on grid refinement and on fine grids axial velocity fluctuations followed the other two quantities closely in their rise to asymptotic state. Comparison of these quantities with a jet without particles revealed that the flow features are same for a jet with and without particles, and at the mass loading studied, particles have negligible effect on jet breakdown. Another study performed at a higher Reynolds number of Re =11,000, under similar flow conditions showed similar behavior. To assess the ability of predicting dispersed phase, simulations of particle-laden flows at low Stokes number were performed and compared against an experiment by Lau & Nathan (2014). The experiment studies variation of velocity and particle concentration along the centerline, and half widths of a jet velocity and concentration. Particles are injected into a pipe along with air, and the two phase flow is fully developed by the time it exits the pipe into a wind tunnel along with a co-flow. Particles are mono-disperse with a density of 1200 kg/m3. Mass loading is 40% so that particles have a significant effect on the continuous phase. Two cases at particle Stokes number of 1.4, one with Re =10,000, bulk velocity of 12 m/s and particle diameter of 20µm and another with Re =22,500, bulk velocity of 36 m/s and particle diameter of 10µm were simulated. Simulations of both the cases showed good match with experimental measurements of centerline decay for the continuous phase. For the dispersed case, simulations with larger particles showed good match with experimental results, while smaller particles showed differences. This was understood to be the effect of lateral migration which is prominent in case of smaller particles, the models for which have not been used in the present simulation study.

In dieser Dissertation wird das kompressible, nicht-hydrostatische und dreidimensionale Modell All Scale Atmospheric Model (ASAM) für Grobstruktur- bzw. Large-Eddy-Simulationen (LES) angewendet, um lokale Inseleffekte in der karibischen Passatwindzone zu untersuchen. Da das Modell bis dato noch keine Anwendung im Bereich von LES feuchter atmosphärischer Grenzschichten und heterogener Oberflächen fand, wurden einige Bestandteile zum Modellcode hinzugefügt oder überarbeitet. Ein Hauptaugenmerk liegt dabei auf das Einbeziehen orographischer Strukturen mittels angeschnittener Zellen (engl. cut cells). Sowohl die räumliche und zeitliche Diskretisierung der Modellgleichungen als auch die nötigen physikalischen Parameterisierungen werden in einer umfassenden Modellbeschreibung zusammengefasst. Die Robustheit und Stabilität der Modellformulierung wird durch eine Reihe von Simulationen idealisierter Testfälle bestätigt. Large-Eddy-Simulationen werden für das Gebiet der Karibikinsel Barbados zur Untersuchung von Inseleffekten bezüglich Grenzschichtmodifikation, Wolkenbildung und vertikaler Durchmischung von Aerosolen durchgeführt. Durch das Vorhandensein einer topographisch strukturierten Inseloberfläche in der Mitte des Modellgebietes muss das Modellsetup offene seitliche Randbedingungen beinhalten. Damit das einströmende Windfeld konsistent mit der Dynamik einer turbulenten, marinen Grenzschicht ist, wird eine neue Methode implementiert und angewendet, welche auf Störungen des potentiellen Temperaturfeldes mittels finiter Amplituden basiert. Beobachtungen aus der SALTRACE-Messkampagne werden benutzt, um die Modellläufe anzutreiben. Die Ergebnisse einiger Sensitivitätstests zeigen Probleme der Modellierung im Bereich der \"Terra incognita\" auf. Dabei handelt es sich um die Modellierung auf räumlichen Skalen, welche zwischen denen von LES und wolkenauflösenden Modellen liegen. Außerdem werden Auswirkungen von entweder turbulent oder laminar anströmenden Windfeldern auf die Simulationsergebnisse untersucht. Besonders die Wolkeneigenschaften im Lee von Barbados werden in diesen Simulationen merklich beeinflusst. Ergebnisse einer weiteren Simulation mit einer sehr starken Passatinversion bringt deren Einfluss auf die Dicke und Höhe der simulierten Wolkenschichten zum Vorschein. Die Veränderung von Saharastaubschichten, welche Barbados über weiträumigen Transport über den Atlantik erreichen, wird analysiert. Die Auswirkungen beinhalten sowohl eine Ausdünnung und ein Absinken dieser Schichten als auch turbulenter Transport in Richtung Erdoberfläche. Die genaue Position der beeinflussten Schichten und die Stärke des turbulenten Mischens werden hauptsächlich von der atmosphärischen Schichtung, der Inversionsstärke und Windscherung gesteuert. Vergleiche zwischen den LES-Modellergebnissen und Daten aus Doppler-Windlidarmessungen zeigen gute Übereinstimmungen in der Formierung der konvektiven Strukturen tagsüber und des Vertikalwindfeldes. / In this thesis, the fully compressible, three-dimensional, nonhydrostatic atmospheric model called All Scale Atmospheric Model (ASAM) is utilized for large eddy simulations (LES) to investigate local island effects at the Caribbean. Since the model has not been applied to LES for moist boundary layers and heterogeneous surfaces so far, several parts are added to the model code or reworked. A special focus lies on the inclusion of orographical structures via the cut cell method. Spatial and temporal discretization as well as necessary physical parameterizations are summarized in a thorough model description. The robustness of the model formulation is confirmed by a set of idealized test case simulations. Large eddy simulations are performed for the area of the Caribbean island Barbados to investigate island effects on boundary layer modification, cloud generation and vertical mixing of aerosols. Due to the presence of a topographically structured island surface in the domain center, the model setup has to be designed with open lateral boundaries. In order to generate inflow turbulence consistent with the upstream marine boundary layer forcing, the newly developed cell perturbation method based on finite amplitude perturbations is applied. Observations from the SALTRACE field campaign are used to initialize the model runs. Several numerical sensitivity tests are carried out to demonstrate the problems related to \"gray zone modeling\" beyond LES scales or when the turbulent marine boundary layer flow is replaced by laminar winds. Especially cloud properties west of Barbados (downwind) are markedly affected in these simulations. Results of an additional simulation with a strong trade-wind inversion reveal its effect on cloud layer depth and height. The modification of Saharan dust layers reaching Barbados via long-range transport over the North Atlantic is analyzed. Effects of layer thinning, subsidence and turbulent downward transport near the layer bottom become apparent. The position of these layers and strength of downward mixing is found to be mainly controlled atmospheric stability, inversion strength and wind shear. Comparisons of LES model output with wind lidar data show similarities in the formation of the daytime convective plume and the vertical wind structure.

Most flows in nature and engineering are turbulent, and many are wall-bounded. Further, in turbulent flows, the turbulence generally has a large impact on the behavior of the flow. It is therefore important to be able to predict the effects of turbulence in such flows. The Navier-Stokes equations are known to be an excellent model of the turbulence phenomenon. In simple geometries and low Reynolds numbers, very accurate numerical solutions of the Navier-Stokes equations (direct numerical simulation, or DNS) have been used to study the details of turbulent flows. However, DNS of high Reynolds number turbulent flows in complex geometries is impractical because of the escalation of computational cost with Reynolds number, due to the increasing range of spatial and temporal scales. In Large Eddy Simulation (LES), only the large-scale turbulence is simulated, while the effects of the small scales are modeled (subgrid models). LES therefore reduces computational expense, allowing flows of higher Reynolds number and more complexity to be simulated. However, this is at the cost of the subgrid modeling problem. The goal of the current research is then to develop new subgrid models consistent with the statistical properties of turbulence. The modeling approach pursued here is that of "Optimal LES". Optimal LES is a framework for constructing models with minimum error relative to an ideal LES model. The multi-point statistics used as input to the optimal LES procedure can be gathered from DNS of the same flow. However, for an optimal LES to be truly predictive, we must free ourselves from dependence on existing DNS data. We have done this by obtaining the required statistics from theoretical models which we have developed. We derived a theoretical model for the three-point third-order velocity correlation for homogeneous, isotropic turbulence in the inertial range. This model is shown be a good representation of DNS data, and it is used to construct optimal quadratic subgrid models for LES of forced isotropic turbulence with results which agree well with theory and DNS. The model can also be filtered to determine the filtered two-point third-order correlation, which describes energy transfer among filtered (large) scales in LES. LES of wall-bounded flows with unresolved wall layers commonly exhibit good prediction of mean velocities and significant over-prediction of streamwise component energies in the near-wall region. We developed improved models for the nonlinear term in the filtered Navier-Stokes equation which result in better predicted streamwise component energies. These models involve (1) Reynolds decomposition of the nonlinear term and (2) evaluation of the pressure term, which removes the divergent part of the nonlinear models. These considerations significantly improved the performance of our optimal models, and we expect them to apply to other subgrid models as well. / text

This thesis investigates the impact of upstream duct convolution on the plume development for high speed jets. In particular, investigations are carried out into an unconventional aero-engine exhaust systems comprised of a modified convergent-divergent rectangular nozzle where the converging section of the nozzle includes an S-bend in the duct. The motivation for this work comes from both the military and civilian sectors of the aerospace industry. The growing interest into highly efficient engines in the civilian sector and increasing complexities involved in stealth technologies for military applications has led to new design constraints on aero-engine exhaust systems that require further research into flows through more complex duct geometries. Due to a lack of experimental data into this area in the open literature validation studies are undertaken into flows through an S-bend duct and exhaust plume development from a rectangular convergent-divergent nozzle. The validation work is simulated using RANS CFD with common industrial turbulence models as well as LES with artificial inlet conditions. Subsequently, a CFD investigation into three unconventional aero-engine exhaust systems, with over-expanded conditions, with differing angles of curvature across the converging S-bend is undertaken using both RANS and LES methodologies governed by the validation work. As the curvature of the S-bend was increased it was found that the thrust and effective NPR both decrease. Whilst these changes were within acceptable levels (with some optimisation) for a circumferential extent of up to 53.1 the losses became prohibitive large at extents. For the ducts with a greater circumferential extents separation was seen to occur at the throat of the nozzle; this changes the design parameters of the nozzle leading to a higher Mach number and could potentially be harnessed to improve performance of the engine creating a `variable throat' nozzle. The impact of using different numerical solvers to simulate the flow through an unconventional aero-engine exhaust system has also been considered. The use of LES has shown that the octagonal, hexahedral and trapezoidal shapes initially observed in the development of the plumes of the RANS cases are likely to be an artifact caused by the RANS solver, as would the transverse total pressure gradients observed in the RANS cases at the nozzle exit as they are both absent from all of the LES results. Likewise the implementation of realistic inlet conditions has a significant impact on the development of the plume, particularly in the length of the potential core and the number of shock cells.

Large Eddy Simulation (LES) is carried out using the Rolls-Royce Hydra CFD code in order to investigate and give further insight into highly turbulent, unsteady flow structures for automotive applications. LES resolves time dependent eddies that are modelled in the steady-state by Reynolds-Averaged Navier-Stokes (RANS) turbulence models. A standard Smagorinsky subgrid scale model is used to model the energy transfer between large and subgrid scales. Since Hydra is an unstructured algorithm, a variety of unstructured hexahedral, tetrahedral and hybrid grids are used for the different cases investigated. Due to the computational requirements of LES, the cases in this study replicate and analyse generic flow problems through simplified geometry, rather than modelling accurate race car geometry which would lead to infeasible calculations. The first case investigates the flow around a diffuser-equipped bluff body at an experimental Reynolds number of 1.01 times 10 to the power 6 based on model height and inlet velocity. LES is carried out on unstructured hexahedral grids of 10 million and 20 million nodes, with the latter showing improved surface pressure when compared to the experiments. Comparisons of velocity and vorticity between the LES and experiments at the diffuser exit plane show a good level of agreement. Flow visualisation of the vortices in the diffuser region and behind the model from the mean and instantaneous flow attempts to explain the relation or otherwise between the two. The main weakness of the simulation was the late laminar to turbulent transition in the underbody region. The size of the domain and high experimental Reynolds number make this case very challenging. After the challenges faced by the diffuser-equipped bluff body, the underbody region is isolated so that increased grid refinement can be achieved in this region and the calculation is run at a Reynolds number of 220, 000, reducing the computational requirement from the previous case. A vortex generator mounted onto a flat underbody at an onset angle to the flow is modelled to generate vortices that extend along the length of the underbody and its interaction with the ground is analysed. Since the vortex generator resembles a slender wing with an incidence to the flow, a delta wing study is presented as a preliminary step since literature on automotive vortex generators in ground effect is scarce. Results from the delta wing study which is run at an experimental Reynolds number of 1.56 times 10 to the power 6 are in very good agreement with previous experiments and Detached Eddy Simulation (DES) studies, giving improved detail and understanding. Axial velocity and vorticity contours at several chordwise stations show that the leading edge vortices are predicted very well by a 20 million node tetrahedral grid. Sub-structures that originate from the leading edge of the wing and form around the core of the leading edge vortex are also captured. Large Eddy Simulation for the flow around an underbody vortex generator over a smooth ground and a rough ground is presented. A hexahedral grid of 40 million nodes is used for the smooth ground case, whilst a 48 million node hybrid grid was generated for the rough ground case so that the detailed geometry near the ground could be captured by tetrahedral cells. The geometry for the rough surface is modelled by scanning a tarmac surface to capture the cavities and protrusions in the ground. This is the first time that a rough surface representing a tarmac road has been computed in a CFD simulation, so that its effect on vortex decay can be studied. Flow visualisation of the instantaneous flow has shown strong interaction with the ground and the results from this study have given an initial understanding in this area.

In the last 10-15 years, large-eddy simulation (LES) has become well established for non-reacting flows, and several successful models have been developed for the transfer of momentum and kinetic energy to the subfilter-scales (SFS). However, for reacting flows, LES is still undergoing significant development. In particular, for many premixed combustion applications, the chemical reactions are confined to propagating surfaces that are significantly thinner than the computational grids used in practical LES. In these situations, the chemical kinetics and its interaction with the turbulence are not resolved and must be entirely modelled. There is, therefore, a need for accurate and robust physical modelling of combustion at the subfilter-scales. In this thesis, modelled transport equations for progress variable and flame surface density (FSD) were implemented and coupled to the Favre-filtered Navier-Stokes equations for a compressible reactive thermally perfect mixture. In order to reduce the computational costs and increase the resolution of simulating combusting flows, a parallel adaptive mesh (AMR) refinement finite-volume algorithm was extended and used for the prediction of turbulent premixed flames. The proposed LES methodology was applied to the numerical solution of freely propagating flames in decaying isotropic turbulent flow and Bunsen-type flames. Results for both stoichiometric and lean flames are presented. Comparisons are made between turbulent flame structure predictions for methane, propane, hydrogen fuels, and other available numerical results and experimental data. Details of subfilter-scale modelling, numerical solution scheme, computational results, and capabilities of the methodology for predicting premixed combustion processes are included in the discussions. The current study represents the first application of a full transport equation model for the FSD to LES of a laboratory-scale turbulent premixed flame. The comparisons of the LES results of this thesis to the experimental data provide strong support for the validity of the modelled transport equation for the FSD. While the LES predictions of turbulent burning rate are seemingly correct for flames lying within the wrinkled and corrugated flamelet regimes and for lower turbulence intensities, the findings cast doubt on the validity of the flamelet approximation for flames within the thin reaction zones regime.

This study investigates the complex flow structure generated by riparian emergent vegetation along the edge of floodplain. Detailed velocity and boundary shear stress measurements were carried out for various arrangements of emergent rigid cylindric rods of 3 mm, 6 mm and 9 mm diameters and for three different rod densities. In addition, the impact of foliage on the flow field was assessed during a series of experiments where brushes were used instead of smooth rods. The results of these new experiments are first presented. In addition to the laboratory data, field data was obtained through Acoustic Doppler Current Profiler measurements for two flood events in a stretch of the river Rhône that can be approximated to a straight compound channel with vegetated banks. The analysis of the flow structure highlights the presence of strong secondary circulation and increased vorticity on the river banks. The rods on the edge of the floodplain increase significantly flow resistance, reducing velocity and decreasing boundary shear stress. Flow rate was seen to decrease with increasing vegetative density for all cases except when foliage was added. This suggests that an optimum threshold density, for which a smaller density would lead to an increased flow rate might exist. Wakes trailing downstream of the vegetation stem, planform coherent structures advected between the main channel and the floodplain, and eddying motion in the flow due to enhanced turbulence anisotropy are among the defining patterns observed in the studied compound channel flows with one line of emergent vegetation along the edge of the floodplain. The Shiono and Knight Method (SKM) was modified in order to account for the increased turbulence activity due to the rods. The drag force term was introduced in the same way as in the work of Rameshwaran and Shiono (2007). However, a new term was added to the transverse shear stress term in the form of an Elder formulation, incorporating a friction drag coefficient which can be derived from the experimental data. In this proposed version, the advection term was set to zero. Another version of the SKM, similar to Rameshwaran and Shiono (2007), was also tested with the addition of a local drag friction only applied in the rod region. The proposed SKM version without the advection term was favored as it can be more closely related to the experimental data and to physical processes. Finally, the capabilities of Telemac-2D were tested against the experimental data for various turbulence models. The Large Eddy Simulation turbulence model highlighted some unsteady flow patterns that were observed during experiments, while satisfactorily predicting the lateral velocity and boundary shear stress distributions.

A large eddy simulation (LES) model has been developed and validated for turbulent non-premixed and partially premixed combustion systems. LES based combustion modelling strategy has the ability to capture the detailed structure of turbulent flames and account for the effects of radiation heat loss. Effects of radiation heat loss is modelled by employing an enthalpy-defect based non-adiabatic flamelet model (NAFM) in conjunction with a steady non-adiabatic flamelet approach. The steady laminar flamelet model (SLFM) is used with multiple flamelet solutions through the development of pre-integrated look up tables. The performance of the non-adiabatic model is assessed against experimental measurements of turbulent CH4/H2 bluff-body stabilized and swirl stabilized jet flames carried out by the University of Sydney combustion group. Significant enhancements in the predictions of mean thermal structure have been observed with both bluff body and swirl stabilized flames by the consideration of radiation heat loss through the non-adiabatic flamelet model. In particular, mass fractions of product species like CO2 and H2O have been improved with the consideration of radiation heat loss. From the Sydney University data the HM3e flame was also investigated with SLFM using multiple flamelet strategy and reasonably fair amount of success has been achieved. In this work, unsteady flamelet/progress variable (UFPV) approach based combustion model which has the potential to describe both non-premixed and partially premixed combustion, has been developed and incorporated in an in-house LES code. The probability density function (PDF) for reaction progress variable and scalar dissipation rate is assumed to follow a delta distribution while mixture fraction takes the shape of a beta PDF. The performance of the developed model in predicting the thermal structure of a partially premixed lifted turbulent jet flame in vitiated co-flow has been evaluated. The UFPV model has been found to successfully predict the flame lift-off, in contrast SLFM results in a false attached flame. The mean lift-off height is however over-predicted by UFPV-δ function model by ~20% for methane based flame and under-predicted by ~50% for hydrogen based flame. The form of the PDF for the reaction progress variable and inclusion of a scalar dissipation rate thus seems to have a strong influence on the predictions of gross characteristics of the flame. Inclusion of scalar dissipation rate in the calculations appears to be successful in predicting the flame extinction and re-ignition phenomena. The beta PDF distribution for the reaction progress variable would be a true prospect for extending the current simulation to predict the flame characteristics to a higher degree.

The periods of time-varying turbulence in the atmospheric boundary layer, i.e.\ the morning and evening transitions, are often overlooked or highly idealised by dispersion models. These transitions make up a significant portion of the diurnal cycle and are known to affect the spread of pollution due to the properties of turbulence in the residual and stable layers, resulting in phenomena such as lofting, trapping, and fumigation.\\ Two main simulation techniques are presented for the purpose of modelling the dispersion of passive tracers in both convective and evening transitional boundary layers: Lagrangian stochastic (LS) modelling for 1D, inhomogeneous, non-stationary turbulence; and large-eddy simulation (LES) with a particle model tracing pollutant paths using a combination of the resolved flow velocities and a random displacement model to represent sub-grid scale motions.\\ In the convective boundary layer, LS models more accurately representing the state of turbulence, and including the effect of skewness, are shown to produce dispersion results in closer agreement with LES. By considering individual particle trajectories, a reflective top boundary in LS models is shown to produce un-physical, sharp changes in velocity and position. By applying a correction to the vertical velocity variance based on representing the stable potential temperature gradient above the boundary layer, particles are contained within the boundary layer in a physically accurate way. \\ An LS model for predicting dispersion in time-varying, skewed turbulence is developed and tested for various particle releases in transitional boundary layers with different rates of decay, showing an improvement in accuracy compared with previous LS models. Further improvement is made by applying a correction to the vertical velocity variance to represent the effect of a positive potential temperature gradient developing over the course of the transition. Finally, a developing stable boundary layer is shown to have a significant trapping effect on particles released near the surface. \\

La recherche en bruit de combustion est de nos jours majoritairement consacrée au développement d'outils de calcul du bruit rayonné par les flammes. Les méthodes actuelles de CFD telles que la LES ou la DNS sont capables de fournir le champ acoustique rayonné par des sources de bruit, mais elles sont cependant limitées à des domaines de faible taille, ceci dû à leur fort coût de calcul. Pour surmonter cette limitation, on a vu l'émergence de méthodes hybrides. Dans cette approche, les sources de bruit sont découplées du son rayonné. Les sources sont alors calculées par les méthodes de DNS et de LES tandis que l'analogie acoustique permet de calculer le son rayonné par des codes acoustiques, moins coûteux en temps de calcul.L'objet de cette étude est de développer un outil numérique sur la base de l'analogie acoustique de Phillips pour de faibles nombres de Mach. Il permet de prendre en compte l'impact des conditions limites sur le champ acoustique résultant. La LES et le code de calcul acoustique développé ont été utilisés pour évaluer le bruit produit par une flamme turbulente confinée. Les deux techniques donnent des résultats en accord tant que les bonnes quantités sont comparées: il a été observé que le signal de pression obtenu directement à partir de la LES contient une quantité non négligeable d'hydrodynamique, laquelle doit être négligée si on cherche à comparer seulement les champs acoustiques issus des deux codes.L'hypothèse d'un nombre de Mach faible est totalement réaliste si l'on considère l'écoulement présent dans une chambre de combustion. Elle conduit à des simplifications significatives lorsque les analogies acoustiques sont considérées. Cependant, cette hypothèse ne peut pas être utilisée pour l'écoulement en amont (entrée d'air, compresseur) ni en aval (turbine, tuyère) des chambres de combustion aéronautiques. Un outil numérique a été développé pour pallier ce problème. Il est basé sur les équations d'Euler Quasi-1D, qui prennent en compte des écoulements convectifs, non isentropiques et non isenthalpiques. Cet outil permet d'estimer les conditions limites acoustiques qui doivent être imposées sur les entrées/sorties d'une chambre de combustion pour prendre en compte la présence d'un écoulement de nombre de Mach non négligeable, alors que les calculs acoustiques sont eux effectués sous cette hypothèse fortement restrictive. / Today, much of the current effort in combustion noise is the development of efficient numerical tools to calculate the noise radiated by flames. Although unsteady CFD methods such as LES or DNS can directly provide the acoustic field radiated by noise sources, this evaluation is limited to small domains due to high computational costs. Hybrid methods have been developed to overcome this limitation. In these schemes, the noise sources are decoupled from the radiated sound. The sources are still calculated by DNS or LES codes whereas the radiated sound is evaluated by acoustic codes using an acoustic analogy.In the present study, a numerical tool based on the Phllips' analogy for low Mach numbers flows has been developed. This tool accounts for the role of the boundary conditions in the resulting acoustic field. Both LES and the acoustic code developed here are used to assess the noise produced by a turbulent confined flame of a turbulent swirled-stabilized staged combustor. Good agreements are obtained between both techniques as long as the good quantities are compared: the pressure signal obtained directly from LES contains a non negligible amount of hydrodynamics that must be removed when a suitable acoustics-acoustics comparison is sought. The low Mach number assumption is completely realistic when considering the flow within a combustion chamber; it also conducts to considerable simplifications when leading with acoustic analogies. However, it cannot be used for the upstream (air-intake, compressors) and downstream (turbines, nozzle) of an aeronautical combustion chamber. A numerical tool is developed based on the quasi-1D Linearized Euler Equations in order to account for convective, non-isentropic and non-isenthalpic flows. By means of this tool, it is possible to estimate the acoustic boundary conditions that should be imposed at the inlet/oultlet of a given combustion chamber when performing low-Mach number acoustic computations.