In this work high-resolution numerical simulation (Large-Eddie Simulation, LES) has been used to study the characteristic factors causing and influencing the development of moist convective clouds. Through this work a 1D cloud-model was derived from first principles to represent the vertical profile of individual convective clouds. A microphysics framework was implemented to ensure identical behaviour in LES and cloud-model integration where the microphysical processes represented are numerically integrated using a novel adaptive step microphysics integration which uses the physical speed at which a process takes place to adjust the integration step size (in space and time). This work also introduces a simple representation of cloud-droplet formation which allows for super-saturation to exist in-cloud and through this provide more physical representation of the in-cloud state. Together with high-resolution simulation of isolated individual and interacting multiple clouds in environmental conditions leading to shallow convection, the 1D cloud-model was used to infer that the principal influence on moist convective clouds is the entrainment of air from a cloud’s immediate environment which is significantly different from the environmental mean state. This suggests that convection parameterisations must represent the influence of moist convective downdrafts to properly predict the vertical structure of convective clouds so as to correctly predict the cloud-top height and vertical transport. Finally it was found that cloud-base radius is not in itself adequate as a means of classification for defining cloud-types as clouds with the same cloud-base radius showed large variation (≈ 600m) in cloud-top height. Based on simulations of individual convective clouds it was found that 3D simulations are necessary to capture the full dynamic behaviour of convective clouds (2D axisymmetric simulations have too little entrainment) and that agreement with the 1D cloud-model could only be found when entrainment was diagnosed from simulation instead of being parameterised by the traditional Morton-Turner model and only for 2D axisymmetric simulations, suggesting that the 1D cloud-model will require further extension or the diagnosis of entrainment improved.

Atmospheric boundary layer flows around wind turbines distributed in a large wind farm can be examined by the use of large eddy simulation (LES), which is based on the assumption that large eddies in the flow are anisotropic and depend on the mean flow and the configuration geometry, while smaller eddies are isotropic and homogeneous, and can be modeled via subgrid scale models. In this thesis, a pseudo-spectral LES code with inflow conditions imposed through a precursor concurrent simulation is utilized to model the flow around a single wind turbine or a large wind farm operating in thermally-stratified conditions. The effect of the wind turbines on humidity is monitored through an additional scalar convection equation. It is found that on average, the effect of an individual wind turbine on the humidity is less than 1%, while the effect of the wind farm on humidity can reach 1-2% in the cumulative wakes.

Since its introduction in 1997, the use of Detached Eddy Simulation (DES) and similar hybrid turbulence techniques has become increasingly popular in the field of CFD. However, with increased use some of the limitations of the DES model have become apparent. One of these is the dependence of DES on grid construction, particularly regarding the point of transition between the Reynolds-Averaged Navier-Stokes and Large Eddy Simulation models. An additional issue that arises with unstructured grids is the definition of the grid spacing in the implementation of a DES length scale. To lay the ground work to study these effects the Spalart-Allmaras one-equation turbulence model, SA based DES hybrid turbulence model, and the Scale Adaptive Simulation hybrid turbulence model are implemented in an unstructured grid CFD code, UNCLE. The implemented SA based DES model is validated for flow over a three-dimensional circular cylinder for three different turbulent Reynolds numbers. Validation included studying the pressure, skin friction coefficient, centerline velocity distributions averaged in time and space. Tools to output the mean velocity profiles and Reynolds stresses were developed. A grid generation code was written to generate a two/three dimensional circular cylinder grid to simulate flow over the cylinder in UNCLE. The models implemented and validated, and the additional tools mentioned will be used in the future.

The use of Large Eddy Simulation (LES) for modelling air flows in buildings is a growing area of Computational Fluid Dynamics (CFD). Compared to traditional CFD techniques, LES provides a more detailed approach to modelling turbulence in air. This offers the potential for more accurate modelling of low energy natural ventilation which is notoriously difficult to model using traditional CFD. Currently, very little is known about the performance of LES for modelling natural ventilation, and its computational intensity makes its practical use on desk top computers prohibitive. The objective of this work was to apply LES to a variety of natural ventilation strategies and to compile guidelines for practitioners on its performance, including the trade-off between accuracy and cost.

Subfilter scalar variance is a key quantity for scalar mixing at the small scales of a turbulent flow and thus plays a crucial role in large eddy simulation (LES) of combustion. While prior studies have mainly focused on the physical aspects of modeling subfilter variance, the current work discusses variance models in conjunction with numerical errors due to their implementation using finite difference methods. Because of the prevalence of grid-based filtering in practical LES, the smallest filtered scales are generally under-resolved. These scales, however, are often important in determining the values of subfilter models. A priori tests on data from direct numerical simulation (DNS) of homogenous isotropic turbulence are performed to evaluate the numerical implications of specific model forms in the context of practical LES evaluated with finite differences. As with other subfilter quantities, such as kinetic energy, subfilter variance can be modeled according to one of two general methodologies. In the first of these, an algebraic equation relating the variance to gradients of the filtered scalar field is coupled with a dynamic procedure for coefficient estimation. Although finite difference methods substantially underpredict the gradient of the filtered scalar field, the dynamic method is shown to mitigate this error through overestimation of the model coefficient. The second group of models utilizes a transport equation for the subfilter variance itself or for the second moment of the scalar. Here, it is shown that the model formulation based on the variance transport equation is consistently biased toward underprediction of the subfilter variance. The numerical issues stem from making discrete approximations to the chain rule manipulations used to derive convective and diffusive terms in the variance transport equation associated with the square of the filtered scalar. This set of approximations can be avoided by solving the equation for the second moment of the scalar, suggesting that model's numerical superiority. / text

The aim of the thesis is to develop and validate subgrid-scale models that are relevant for large eddy simulations of complex flows including scalar mixing. A stochastic Smagorinsky model with adjustable variance and time scale is developed by adding a stochastic component to the Smagorinsky constant. The stochastic model is shown to provide for backscatter of both kinetic energy and scalar variance without causing numerical instabilities. In addition, new models for the subgrid-scale stress and passive scalar flux are derived from modelled subgrid scale transport equations. These models properly account for the anisotropy of the subgrid scales and have potentials wall bounded flows. The proposed models are validated in wall bounded flows with and without rotation and show potential or significantly improve predictions for such cases. / <p>QC 20100826</p>

Numerical simulation is progressively taking importance in the design of an aero- nautical engine. However, concerning the particular case of cooling devices, the high number of sub-millimetric cooling holes is an obstacle for computational sim- ulations. A classical approach goes through the modelling of the effusion cooling by homogenisation. It allows to simulate a full combustor but failsin representing the jet penetration and mixing. A new approach named thickened-hole model was developed during this thesis to overcome this issue. A work on improving the mesh resolution onkey areas thanks to an automatic adaptive method is also presented, leading to a clear breakthrough. In parallel, as the flame tube temperature is a cornerstone for the combustor durability,a low-cost approach is proposed to predict it. To meet the time-constraints of design, it is based on thermal modelling instead of a direct thermal resolution.

Aero-engine intakes play a critical role in the performance of modern high-bypass turbofan engines. It is their function to provide uniformly distributed, steady air flow to the engine fan face under a variety of flow conditions. However, during situations of high incidence, high curvature of the intake lip can accelerate flow to supersonic speeds, terminating with a shock wave. This produces undesirable shock wave boundary layer interactions (SWBLIs). Reynolds-Averaged Navier Stokes (RANS) turbulence models have been shown to be insensitive to the effects of boundary layer relaminarisation present in these highly-accelerated flows. Further, downstream of the SWBLI, RANS methods fail to capture the distorted flow that propagates towards the engine fan face. The present work describes simulations of a novel experimental intake rig model that replicates the key physics found in a real intake- namely acceleration, shock and SWBLI. The model is a simple geometric configuration resembling a lower intake lip at incidence. Simulations are carried out at two angles of attack, $\alpha=23^{\circ}$ and $\alpha=25^{\circ}$, with the more aggressive $\alpha=25^{\circ}$ possessing a high degree of shock oscillation. RANS, Large Eddy Simulations (LES) and hybrid RANS-LES are carried out in this work. Modifications to the one-equation Spalart-Allmaras (SA) RANS turbulence model are proposed to account for the effects of re-laminarisation and curvature. The simulation methods are validated against two canonical test cases. The first is a subsonic hump model where RANS modifications give a noticeable improvement in surface pressure predictions, even for this mild acceleration case. However, RANS is shown to over-predict the separation size. LES performs much better here, as long as the Smagorinsky-Lilly SGS model is not used. The $\sigma$-SGS model is found to perform best, and is used to run a hybrid RANS-LES that predicts a separation bubble size within $4\%$ of LES. The second canonical test case is a transonic hump that features a normal shockwave and SWBLI. RANS performs well here, predicting shock location, surface pressure and separation with good agreement with experimental measurements. Hybrid RANS-LES also performs well, but predicts a shock downstream of that measured by experiment. The use of an improved shock sensor here is able to maintain solution accuracy. Simulations of the intake rig are then run. RANS modifications provide a significant improvement in prediction of the shock location and lip surface pressure compared to the standard SA model. However, RANS models fail to reproduce the post shock interaction flow well, giving incorrect shape of the flow distortion. Further, RANS is inherently unable to capture the unsteady shock oscillations and related flow features. LES and hybrid RANS-LES predict the shock location and SWBLI well, with the downstream flow distortion also in very good agreement with experimental measurements. LES and hybrid RANS-LES are able to reproduce the time averaged smearing of the shock which RANS cannot. However, shock oscillations in the $\alpha=25^{\circ}$ case present a particular challenge for costly LES, requiring long simulation time to obtain time averaged flow statistics. Hybrid RANS-LES offers a significant saving in computational expense, costing approximately $20\%$ of LES. The work proposes recommendations for simulation strategy for intakes at incidence based on computational cost and performance of simulation methods.

Tese de doutoramento. Engenharia Mecânica. Faculdade de Engenharia. Universidade do Porto. 2009

Turbulent flows present structures with a wide range of scales. The computation of the complete physics of a turbulent flow (termed DNS) is very expensive and is, for the time being, limited to low and medium Reynolds number flows. As a way to capture high Reynolds number flows, a part of the physics complexity has to be modeled. Large eddy simulation (LES) is a simulation strategy where the large turbulent eddies present on a given mesh are captured and the influence of the non-resolved scales onto the resolved ones is modeled. The present thesis reports on the development and validation of a methodology in order to apply LES for complex wall-bounded flows. Discretization methods and LES models, termed subgrid scale models (SGS), compatible with such a geometrical complexity are discussed. It is proved that discrete a kinetic energy conserving discretization of the convective term is an attractive solution to perform stable simulations without the use of an artificial dissipation, as upwinding. The dissipative effect of the SGS model is thus unaffected by any additional dissipation process. The methodology is first applied to a developed parallel fourth-order incompressible flow solver for cartesian non-uniform meshes. In order to solve the resulting Poisson equation, an efficient multigrid solver is also developed. The code is first validated using DNS (Taylor-Green vortex, channel flow, four-vortex system) and LES (channel flow), and finally applied to the investigation of an aircraft two-vortex system in ground effect. The methodology is then applied to improve a RANS-based industrial unstructured compressible flow solver, developed at CENAERO, to perform well for LES applications. The proposed modifications are tested successfully on the unsteady flow past a sphere at Reynolds of 300 and 10000, corresponding to the subcritical regime.