This thesis aims to explore the potential for using large eddy simulation (LES) as a predictive tool for gas-turbine compressor flows. Compressors present a significant challenge for the Reynolds Averaged Navier-Stokes (RANS) based CFD methods commonly used in industry. RANS models require extensive calibration to experimental data, and thus cannot be used predictively. This thesis explores how LES can offer a more predictive alternative, by exploring the sensitivity of LES to sources of uncertainty. Specifically, the importance of the numerical scheme, the Sub-Grid Scale (SGS) model, and the correct specification of inflow turbulence is examined. The sensitivity of LES to the numerical scheme is explored using the Taylor-Green vortex test case. The numerical smoothing, controlled by a user defined smoothing constant, is found to be important. To avoid tuning the numerical scheme, a locally adaptive smoothing (LAS) scheme is implemented. But, this is found to perform poorly in a forced isotropic turbulence test case, due to the intermittency of the dispersive error. A novel scheme, the LAS with windowing (LASW) scheme, is thus introduced. The LASW scheme is shown to be more suitable for predictive LES, as it does not require tuning to a known solution. The LASW scheme is used to perform LES on a compressor cascade, and results are found to be in close agreement with direct numerical simulations. Complex transition mechanisms, combining characteristics of both natural and bypass modes, are observed on the pressure surface. These mechanisms are found to be sensitive to numerical smoothing, emphasising the importance of the LASW scheme, which returns only the minimum smoothing required to prevent dispersion. On the suction surface, separation induced transition occurs. The flow here is seen to be relatively insensitive to numerical smoothing and the choice of SGS model, as long as the Smagorinsky-Lilly SGS model is not used. These findings are encouraging, as they show that, with the LASW scheme and a suitable SGS model, LES can be used predictively in compressor flows. In order to be predictive, the accurate specification of inflow conditions was shown to be just as important as the numerics. RANS models are shown to over-predict the extent of the three dimensional separation in the endwall - suction surface corner. LES is used to examine the challenges for RANS in this region. The LES shows that it is important to accurately capture the suction surface transition location, with early transition leading to a larger endwall separation. Large scale aperiodic unsteadiness is also observed in the endwall region. Additionally, turbulent anisotropy in the endwall - suction surface corner is found to be important. Adding a non-linear term to the RANS model leads to turbulent stresses that are in better agreement with the LES. This results in a stronger corner vortex which is thought to delay the corner separation. The addition of a corner fillet reduces the importance of anisotropy, thereby reducing the uncertainty in the RANS prediction.

Cette étude a pour objectif d’étudier différentes approches de modélisation de la turbulence sur un compresseur centrifuge industriel à fort taux de pression afin d’essayer d’élargir notre compréhension des différents phénomènes physiques mis en jeu et leur interaction avec la turbulence. D’abord, la sensibilité au maillage et au modèle turbulence est évalué sur des calculs RANS. Une analyse de simulations LES est ensuite effectuée. En particulier, une étude de l’effet de la turbulence sur l’écoulement et une comparaison aux résultats RANS et expérimentaux est réalisée. Enfin, deux approches hybrides DES sont étudiées afin d’exposer les problèmes rencontrés par ces modèles sur cette configuration. Il en résulte une évaluation des différentes méthodes et de leur applicabilité future dans un contexte industriel. / This study aims to assess the abilities of existing numerical simulation methods to predict the complex physical phenomena occurring in an industrial centrifugal compressor and especialy the effect of turbulence on the different flow features. RANS simulations are first performed using various turbulence model, then LES simulations and finally, two simulations using RANS/LES hybrid models of DES type are carried out. The whole compressor operating range is simulated using RANS, but because of LES and DES high computational costs, attention is focussed on the nominal operating point. Particular care is devoted to determine the impact of grid refinement on the simulation results. To this end, simulations are performed on three grids, respectively composed of over 8, 26 and 165 million cells. Even though the grids used do not fulfill the mesh refinement criteria recommended in the litterature for an accurate wall-resolved LES simulation, the simulation performed on the denser grid provides interesting conclusions on the turbulence generation and its interaction with the mean flow.The hybrid DES approches used involve a shield function to prevent the boundary layers to be computed in LES. However, this function is found to be unsuited to this centrifugal compressor flow. Indeed, the RANS and LES regions are not correctly located and most of the tip leakage flow is resolved using a RANS approach, preventing the development of turbulence.This work allowed us to evalute the various approches and highlight some of the problems and advantages of each for the simulation of this centrifugal compressor.

In this thesis a numerical study has been undertaken to investigate turbulent flow and heat transfer in a number of flow problems, representing the gas-cooled reactor core flows. The first part of the research consisted of a meticulous assessment of various advanced RANS models of fluid turbulence against experimental and numerical data for buoyancy-modified mixed convection flows, such flows being representative of low-flow-rate flows in the cores of nuclear reactors, both presently-operating Advanced Gas-cooled Reactors (AGRs) and proposed ‘Generation IV’ designs. For this part of the project, an in-house code (‘CONVERT’), a commercial CFD package (‘STAR-CD’) and an industrial code (‘Code_Saturne’) were used to generate results. Wide variations in turbulence model performance were identified. Comparison with the DNS data showed that the Launder-Sharma model best captures the phenomenon of heat transfer impairment that occurs in the ascending flow case; v^2-f formulations also performed well. The k-omega-SST model was found to be in the poorest agreement with the data. Cross-code comparison was also carried out and satisfactory agreement was found between the results.The research described above concerned flow in smooth passages; a second distinct contribution made in this thesis concerned the thermal-hydraulic performance of rib-roughened surfaces, these being representative of the fuel elements employed in the UK fleet of AGRs. All computations in this part of the study were undertaken using STAR-CD. This part of the research took four continuous and four discrete design factors into consideration including the effects of rib profile, rib height-to-channel height ratio, rib width-to-height ratio, rib pitch-to-height ratio, and Reynolds number. For each design factor, the optimum configuration was identified using the ‘efficiency index’. Through comparison with experimental data, the performance of different RANS turbulence models was also assessed. Of the four models, the v^2-f was found to be in the best agreement with the experimental data as, to a somewhat lesser degree were the results of the k-omega-SST model. The k-epsilon and Suga models, however, performed poorly. Structured and unstructured meshes were also compared, where some discrepancies were found, especially in the heat transfer results. The final stage of the study involved a simulation of a simplified 3-dimensional representation of an AGR fuel element using a 30 degree sector configuration. The v^2-f model was employed and comparison was made against the results of a 2D rib-roughened channel in order to assess the validity and relevance of the precursor 2D simulations of rib-roughened channels. It was shown that although a 2D approach is extremely useful and economical for ‘parametric studies’, it does not provide an accurate representation of a 3D fuel element configuration, especially for the velocity and pressure coefficient distributions, where large discrepancies were found between the results of the 2D channel and azimuthal planes of the 3D configuration.

A low-emissions power generator comprising a solid oxide fuel cell coupled to a gas turbine has been developed by Rolls-Royce Fuel Cell Systems. As part of the cycle, a fraction of the unreacted fuel (the off-gas) and oxidizer streams is reacted in a burner, which is the main source of pollutant formation. In this thesis a computational model of the burner has been developed which captures the formation of NOx and the oxidation of CO. This model gives accurate predictions at low computational cost, making it suitable for use as a design tool in future burner design optimization through parametric studies. A key factor in increasing computational efficiency was the development of a reduced H2/CO/N2 kinetic mechanism; from a starting mechanism of 30 species to 10 and 116 reactions to 6. The results of laminar opposed-flow diffusion flames have been used to validate the reduced mechanism. Several different turbulent combustion models have been evaluated by creating an interface between the reduced kinetic mechanism and the commercial CFD solver FLUENT. Comparison of model predictions with well-characterized turbulent syngas flames, which share a similar fuel composition to the experimental work conducted on the off-gas burner, shows acceptable agreement. These studies have demonstrated the sensitivity of modelling constants. Improved predictions were achieved by calibrating these constants and including radiative heat losses. Following suitable modification to reflect the predominantly laminar flow present in the current burner design, the relevant modelling approaches were applied to the off-gas burner. Comparison was made to previous detailed measurements, showing that the important trends of NOx and CO are captured in general. The model was extended to high pressure conditions, similar to those in the actual off-gas burner, with the emissions predictions within design limits. The outcome of this work is a fast, accurate design tool for CFD which has capabilities to simulate beyond the laminar burner studied here. It may be applied to more general types of off-gas/syngas burners where turbulence-chemistry interaction is expected to be more significant.

This Thesis presents simulation of flow past laboratory-scale and full-scale tidal stream turbines (TST) using EDF's open-source CFD solver Code_Saturne. The work shows that detailed results may be obtained with confidence and that greater information on the loading and wake structure is available than other methods, such as blade element momentum theory.Results are obtained using a new sliding-mesh method that has been implemented in Code_Saturne as part of this work. The sliding-mesh method uses internal Dirichlet boundary conditions with values on the interface prescribed via a halo-point method. Parallel performance is optimised by a carefully-chosen method of exchanging information between specific processes. Validation is provided for flow past a rotating cylinder and a sphere.For the laboratory-scale TST, Reynolds-Averaged Navier-Stokes models are used to model turbulence. The k-omega-SST and Launder-Reece-Rodi (LRR) models yield good agreement with experimental values of power and thrust coefficients as a function of tip-speed ratio (TSR). The standard k-epsilon model is shown to perform poorly due to an overprediction of turbulent kinetic energy upstream of the rotor plane. The k-omega-SST model is then used to examine wake behaviour for parametric studies of turbulence intensity and TSR. Increased turbulence levels are shown to reduce the downstream propagation of the wake because of increased mixing. The near wake is influenced by the TSR, whilst the far wake is independent of TSR.The predicted effect of tidal conditions typical of the EMEC test site are considered for flow past Tidal Generation Limited's 1MW TST. The effect of sheared-velocity profiles leads to an increase in loading on an individual turbine blade at the point of a rotation where velocity shear is greatest. The effect of increased yaw angle leads to large fluctuations of the power coefficient, but smaller fluctuations of the thrust coefficient. Mean values of thrust and power decrease as a function of the cosine of the yaw angle and yaw angle squared respectively.

[pt] Apesar dos recentes avanços tecnológicos e do surgimento de computadores extremamente rápidos, a simulação numérica direta de escoamentos turbulentos ainda é proibitivamente cara para a maioria das aplicações de engenharia e até mesmo para algumas aplicações de pesquisa. As simulações utilizadas são, no geral, baseadas em grandezas médias e altamente dependentes de modelos de turbulência. Apesar de amplamente utilizados, tais modelos não conseguem prever adequadamente o escoamento médio em muitas aplicações, como o escoamento em um duto quadrado. Com o reflorescimento do Aprendizado de Máquina nos últimos anos, muita atenção está sendo dada ao uso de tais técnicas para substituir os modelos tradicionais de turbulência. Este trabalho estudou o uso de Redes Neurais como alternativa para aprimorar a simulação de escoamentos turbulentos. Para isso, a técnica PIV-Estereoscópico foi aplicada ao escoamento em um duto quadrado para obter dados experimentais de estatísticas do escoamento e campos médios de velocidade de 10 casos com diferentes números de Reynolds. Um total de 10 metodologias foram avaliadas para entender quais grandezas devem ser previstas por um algoritmo de aprendizado de máquina para obter simulações aprimoradas. A partir das metodologias selecionadas, excelentes resultados foram obtidos com uma Rede Neural treinada a partir dos dados experimentais para prever o termo perpendicular do Tensor de Reynolds e a viscosidade turbulenta. As simulações turbulentas auxiliadas pela Rede Neural retornaram campos de velocidade com menos de 4 por cento de erro, em comparação os dados medidos. / [en] Although the technological advances that led to the development of fast computers, the direct numerical simulation of turbulent flows is still prohibitively expensive to most engineering and even some research applications. The CFD simulations used worldwide are, therefore, based on averaged quantities and heavily dependent on mathematical turbulence models. Despite widely used, such models fail to proper predict the averaged flow in many practical situations, such as the simple flow in a square duct. With the re-blossoming of machine learning methods in the past years, much attention is being given to the use of such techniques as a replacement to the traditional turbulence models. The present work evaluated the use of Neural Networks as an alternative to enhance the simulation of turbulent flows. To this end, the Stereoscopic-PIV technique was used to obtain well-converged flow statistics and velocity fields for the flow in a square duct for 10 values of Reynolds number. A total of 10 methodologies were evaluated in a data-driven approach to understand what quantities should be predicted by a Machine Learning technique that would result in enhanced simulations. From the selected methodologies, accurate results could be obtained with a Neural Network trained from the experimental data to predict the nonlinear part of the Reynolds Stress Tensor and the turbulent eddy viscosity. The turbulent simulations assisted by the Neural Network returned velocity fields with less than 4 percent in error, in comparison with those previously measured.

Les écoulements transsoniques rencontrés dans le cadre de la propulsion aéronautique et spatiale sont associés à l'apparition d'ondes de choc. En impactant la couche limite se développant sur une paroi, un gradient de pression adverse est généré qui conduit à l'épaississement ou au décollement de la couche limite. Lors de la vibration de la structure, l'onde de choc oscille et interagit avec la couche limite, générant une fluctuation de la pression statique à la paroi. Il s'ensuit alors un échange d'énergie entre le fluide et la structure qui peut être stabilisant ou au contraire conduire à une instabilité aéroélastique (flottement). La modélisation de la réponse instationnaire de l'interaction onde de choc / couche limite pour l'étude des interactions fluide-structure est l'objet de ce travail de recherche. Il s'appuie sur la résolution des équations de Navier-Stokes moyennées (RANS) et la modélisation de la turbulence. Les méthodes et modèles utilisés ont été validés à partir de résultats expérimentaux issus d'une tuyère transsonique dédiée à l'étude des interactions fluide-structure. Ces travaux sont ensuite appliqués à l'amélioration de la prédiction du flottement en turbomachine. Une méthode linéarisée en temps permettant la résolution des équations RANS dans le domaine fréquentiel est utilisée. Nous confirmons l'importance de la dérivation du modèle de turbulence lors de la prédiction d'une interaction forte entre une onde de choc et une couche limite décollée. Une méthode de régularisation est présentée puis appliquée aux opérateurs non dérivables du modèle de turbulence k-! de Wilcox (2006). La prédiction de la réponse instationnaire de l'interaction onde de choc / couche limite dans une tuyère est évaluée à partir de simulations bidimensionnelles et présente un bon accord avec les données expérimentales. En évaluant l'influence de la fréquence réduite, une instabilité aéroélastique de type flottement transsonique est identifiée. Un dispositif de contrôle, reposant sur la génération d'ondes de pression rétrogrades à l'aval de la tuyère, est proposé puis validé numériquement. Enfin, une méthodologie est proposée pour comprendre les mécanismes aérodynamiques conduisant au flottement. Pour cela, il a été réalisé un dessin provisoire d'une soufflante transsonique à fort taux de dilution. Cette soufflante, l'ECL5, est destinée à l'étude expérimentale des instabilités aérodynamiques et aéroélastiques. La méthodologie proposée repose sur la simulation 2D d'une coupe de tête et met à profit la linéarisation pour analyser la contribution de sources locales en fonction de la fréquence réduite, du diamètre nodal et de la déformée modale / Transonic flows, which are common in aeronautical and spatial propulsion systems, produce shock-waves over solid boundaries. When a shock-wave impacts the boundary layer, an adverse pressure gradient is generated and a thickening or even a separation of the boundary layer is induced. If the solid boundary vibrates, the shock-wave oscillates, interacts with the boundary layer and produce a fluctuation of the static pressure at the wall. This induces an exchange of energy between the fluid and the structure which can be stabilising or lead to an aeroelastic instability (flutter).The main objective of this PhD thesis is the modelling of the unsteady behaviour the simulation of the shock-wave/boundary layer interaction for fluid-structure interaction studies. To this end, simulations have been carried out to solve Reynolds-Averaged Navier-Stokes equations using two equations turbulence model. The method is validated thanks to experimental data obtained on a transonic nozzle dedicated to aeroelastic studies. This method is then use to increase the predictability of flutter events in turbomachinery.A time linearised frequency-domain method is applied to RANS equations. It is shown that the unsteady behaviour of the turbulent boundary-layer contributes to the fluctuating static pressure when the shock-wave boundary layer interaction is strong. Hence, the frozen turbulence assumption is not valid and the turbulence model must be derivated. Thus, the regularisation of the non derivable operators is proposed and applied on k-? Wilcox (2006) turbulence model.The unsteady behaviour of the shock-wave/boundary-layer interaction in a transonic nozzle is evaluated thanks to 2D numerical simulations and shows good agreement with experimental data. When varying the reduced frequency an aeroelastic instability is found, known as transonic flutter. An active control device generating backward travelling pressure waves is then designed and numerically validated.Finally, a methodology is proposed to understand the aerodynamic onsets of transonic flutter. To this end, a preliminary design of a high bypass ratio transonic fan has been carried out. This fan, named ECL5, is dedicated to experimental aerodynamic and aeroelastic studies. The methodology relies on 2D simulations of a tip blade passage and uses linearisation to analyse the contribution of local sources as a function of reduced frequency, nodal diameter and mode shape

The marine current turbine (MCT) is an exciting proposition for the extraction of renewable tidal and marine current power. However, the numerical prediction of the performance of the MCT is difficult due to its complex geometry, the surrounding turbulent flow and the free surface. The main purpose of this research is to develop a computational tool for the simulation of a MCT in turbulent flow and in this thesis, the author has modified a 3D Large Eddy Simulation (LES) numerical code to simulate a three blade MCT under a variety of operating conditions based on the Immersed Boundary Method (IBM) and the Conservative Level Set Method (CLS). The interaction between the solid structure and surrounding fluid is modelled by the immersed boundary method, which the author modified to handle the complex geometrical conditions. The conservative free surface (CLS) scheme was implemented in the original Cgles code to capture the free surface effect. A series of simulations of turbulent flow in an open channel with different slope conditions were conducted using the modified free surface code. Supercritical flow with Froude number up to 1.94 was simulated and a decrease of the integral constant in the law of the wall has been noticed which matches well with the experimental data. Further simulations of the marine current turbine in turbulent flow have been carried out for different operating conditions and good match with experimental data was observed for all flow conditions. The effect of waves on the performance of the turbine was also investigated and it has been noticed that this existence will increase the power performance of the turbine due to the increase of free stream velocity.

Among the different types of flows encountered in practical applications, the physics of turbulent separated flows and turbulence induced secondary motion are not fully understood despite the large amount of previous experimental and numerical work. The objectives of this work are to study theoretically and computationally the conditions at the separation and reattachment point, the numerical simulation of turbulence induced secondary motion in non-circular ducts, and to provide a comprehensive test of different RANS models of these types of flow. In a theoretical study of flow separation, a Lagrangian approach was first used to derive an Eulerian criterion, which associates separation and reattachment points to a critical point in the eigenvalues of the Cauchy-Green tensor. A turbulent separated boundary layer under the influence of an adverse pressure gradient was simulated using DNS and LES techniques. A bootstrapping method was used to obtain high fidelity results at a relatively high Reynolds number with which the performance of some of the most commonly used eddy-viscosity turbulence models was evaluated. The DNS and LES results were used to assess the consistency of the different terms in the k−e , ζ −f , k −ω and Spalart-Allmaras models. Different wall-modelling techniques were employed for the calculation of separated boundary layers. The exact values of the modelled terms were calculated using the reference DNS and LES dataset. These results were used for both a priori and a posteriori tests. It was determined that the eddy-viscosity assumption works well, and that anisotropic effects are not significant in separated boundary layer. For the secondary flow calculation in non-circular ducts, direct numerical simulations of turbulent flow in square and skewed ducts were carried out to determine the effect of the duct (rhombus) included angle on both the mean and turbulence energy budgets. Two skewed ducts, with included angles of 30 and 60 degrees, were simulated. The capability of different turbulence models to predict the secondary velocity field was investigated. Results obtained from a non-linear stress-strain constitutive relation was found to be fairly accurate for the flows at the range of Reynolds number considered in this study. / Thesis (Ph.D, Mechanical and Materials Engineering) -- Queen's University, 2010-11-26 13:52:18.361

Vortex shedding is an important mechanism, by which the flow around bluff bodies create forces that excite vibratory motion. Vortex-induced vibration (VIV) is studied for a single circular cylinder by means of Computational Fluid Dynamics (CFD) simulations. An arbitrary Lagrangian Eulerian (ALE) formulation is used to achieve the grid deformation needed for VIV. In this thesis, a multifaceted approach is undertaken by which response dynamics and wake interaction are addressed. Four major aspects are considered in the study: the Reynolds number (Re); the mass and damping; the degree-of-freedom for VIV; and the mutual effect between VIV and heat transfer.As attention is paid towards high pre-critical Re flow, the turbulent flow around the cylinder is treated by two turbulence modelling approaches: unsteady Reynolds Averaged Navier Stokes (uRANS), and Large Eddy Simulation (LES). The wake-VIV interaction is analyzed by looking at mean velocities and Reynolds stresses, where decomposition of flow scales is undertaken to explore the evolution of coherent eddy structures, downstream of the cylinder. Conversely, the VIV response is analyzed by considering oscillation amplitude and frequency, in addition to the excitation and inertial dynamics.High turbulence in the separated shear layers disorders the cylinder's VIV response and induces higher amplitudes. The sensitivity for Re is found more pronounced in cylinders with low mass and damping. Meanwhile, VIV is found to enhance wake mixing, and to significantly change the near wake Reynolds stresses. It is suggested that the increase in Re brings a change to the wake patterns, which are known in VIV at lower Re. The kinetic energy production, of near wake eddy structures, is qualitatively altered with the presence of VIV. Furthermore, the surface heat flux is found to cause a noticeable increase in VIV amplitude, as long as it does not disorder the wake correlation. The cylinder's oscillation increases the average value of the Nusselt number (Nu), while the local variance of Nu rises markedly post-separation.