Augmentation of the impact of an explosive is routinely achieved by packing metal particles in the explosive charge. When detonated, the particles in the charge are ejected and dispersed. The ejecta influences the post-detonation combustion processes that bolster the blast wave and determines the total impact of the explosive. Thus, it is vital to understand the dispersal and the combustion of the particles in the post-detonation flow, and numerical simulations have been indispensable in developing important insights. Because of the accuracy of Eulerian-Lagrangian (EL) methods in capturing the particle interaction with the post-detonation mixing zone, EL methods have been preferred over Eulerian-Eulerian (EE) methods. However, in most cases, the number of particles in the flow renders simulations using an EL method unfeasible. To overcome this problem, a combined EE-EL approach is developed by coupling a massively parallel EL approach with an EE approach for granular flows. The overall simulation strategy is employed to simulate the interaction of ambient particle clouds with homogenous explosions and the dispersal of particles after detonation of heterogeneous explosives. Explosives packed with aluminum particles are also considered and the aluminum particle combustion in the post-detonation flow is simulated. The effect of particles, both reactive and inert, on the combustion processes is analyzed. The challenging task of solving for clouds of micron and sub-micron particles in complex post-detonation flows is successfully addressed in this thesis.

Choosing correct boundary conditions, flow field characteristics and employing right thermal fluid properties can affect the simulation of convection heat transfer using nanofluids. Nanofluids have shown higher heat transfer performance in comparison with conventional heat transfer fluids. The suspension of the nanoparticles in nanofluids creates a larger interaction surface to the volume ratio. Therefore, they can be distributed uniformly to bring about the most effective enhancement of heat transfer without causing a considerable pressure drop. These advantages introduce nanofluids as a desirable heat transfer fluid in the cooling and heating industries. The thermal effects of nanofluids in both forced and free convection flows have interested researchers to a great extent in the last decade. Investigating the interaction mechanisms happening between nanoparticles and base fluid is the main goal of the study. These mechanisms can be explained via different approaches through some theoretical and numerical methods. Two common approaches regarding particle-fluid interactions are Eulerian-Eulerian and Eulerian-Lagrangian. The dominant conceptions in each of them are slip velocity and interaction forces respectively. The mixture multiphase model as part of the Eulerian-Eulerian approach deals with slip mechanisms and somehow mass diffusion from the nanoparticle phase to the fluid phase. The slip velocity can be induced by a pressure gradient, buoyancy, virtual mass, attraction and repulsion between particles. Some of the diffusion processes can be caused by the gradient of temperature and concentration. The discrete phase model (DPM) is a part of the Eulerian-Lagrangian approach. The interactions between solid and liquid phase were presented as forces such as drag, pressure gradient force, virtual mass force, gravity, electrostatic forces, thermophoretic and Brownian forces. The energy transfer from particle to continuous phase can be introduced through both convective and conduction terms on the surface of the particles. A study of both approaches was conducted in the case of laminar and turbulent forced convections as well as cavity flow natural convection. The cases included horizontal and vertical pipes and a rectangular cavity. An experimental study was conducted for cavity flow to be compared with the simulation results. The results of the forced convections were evaluated with data from literature. Alumina and zinc oxide nanoparticles with different sizes were used in cavity experiments and the same for simulations. All the equations, slip mechanisms and forces were implemented in ANSYS-Fluent through some user-defined functions. The comparison showed good agreement between experiments and numerical results. Nusselt number and pressure drops were the heat transfer and flow features of nanofluid and were found in the ranges of the accuracy of experimental measurements. The findings of the two approaches were somehow different, especially regarding the concentration distribution. The mixture model provided more uniform distribution in the domain than the DPM. Due to the Lagrangian frame of the DPM, the simulation time of this model was much longer. The method proposed in this research could also be a useful tool for other areas of particulate systems. / Thesis (PhD)--University of Pretoria, 2016. / Mechanical and Aeronautical Engineering / PhD / Unrestricted

When designing modern engines it is important to construct a cooling system that cools the engine structure efficiently. Within the cooling system there is always a certain amount of air which can accumulate and form air pockets in critical areas, such as the water jacket, which can lead to wall degradation. A Computational Fluid Dynamics (CFD) method in STAR-CCM+ from CD-adapco, was derived at Volvo Cars in order to study the accumulation of air bubbles in the water jacket. The method was derived by investigating and evaluating already existing methods. The method initially considered as the best suited was the Eulerian-Eulerian approach. The method was validated against three simpler geometries where experimental data was available. The Eulerian-Eulerian approach treats both phases, liquid and gas, as continuous phases. The idea with the method is to solve the Navier-Stokes equation, the continuity equation and the energy equation for both phases using the Eulerian approach, therefore called Eulerian-Eulerian. The interaction between the two phases was important to model properly which was done by including several interaction models within STAR-CCM+. By tuning different coefficients, which were investigated by a thorough parameter study, the method resembled the experimental data in a satisfying way. The best suited mesh for these simpler geometries was a directed mesh. However, the mesh in the water jacket was automatically generated by STAR-CCM+ and the simpler cases were therefore validated with an automated mesh as well. To capture the experimental data the convection scheme for volume fraction had to be of second order when simulating with automated mesh. This resulted in convergence issues when implementing the method on the water jacket. Instead first order convection scheme, which did not present as satisfying results as second order, had to be implemented. Simulations of the water jacket were performed with two different velocities, that were 10 m/s and 19 m/s, and different flow split ratios for the three outlets. Air with volume fraction 0.1 was injected at the inlet during the first 0.5 s followed by 0.5-1.1 s of further simulation without injecting air. Increased velocity resulted in increased flow through of gas, whereas no big difference could be seen between the different outlet flow split ratios. At two different zones lower pressure was found which resulted in gas holdup. To be able to validate the results from the water jacket, experiments would be necessary to perform in order to provide experimental data for comparison. Velocity profiles from the derived two-phase method resemble the velocity profiles from the one-phase simulation from Volvo, which indicated that the two-phase method did not affect the solution in a remarkable way. Granted that the zones of lower pressure and gas holdup normally coincides, the pressure field from the one-phase simulation could be directly studied, which would lower the computational costs significantly.

Fluidized bed reactor technology has been widely adopted within the industry as vital component for numerous manufacturing, power generation and gasification processes due to its enhanced mixing characteristics. Computational modeling of fluidized bed hydrodynamics is a significant challenge that has to be tackled for increasing predictive accuracy. The distributor plate of a fluidized bed is typically modeled using a uniform inlet condition, when in reality the inlet is non-uniform inlet. The regions of bed mass that do not fluidize because of the non-uniform inlet conditions form deadzones and remain static between the jets. A new model based on the mass that contributes to the pressure drop is proposed to model a fluidized bed, and has been investigated for a cylindrical reactor for glass beads, ceramic single solids particles, and glass-ceramic, and ceramic-ceramic binary mixtures. The adjusted mass model was shown to accurately predict fluidization characteristics such as pressure drop and minimum fluidization velocity. The effectiveness of the adjusted mass model was further illustrated by applying it to fluidized beds containing coal, switchgrass, poplar wood, and cornstover biomass particles and coal-biomass binary mixtures. The adjusted mass model was further analyzed for bed expansion heights of different mixtures, and for solids distribution by analyzing the solids volume fraction. The effect of increasing the percent biomass in the mixture was also investigated. To further model the non-uniform inlet condition, two different distributor configurations with 5 and 9 jets was considered for a quasi-2D bed, and simulations were performed in both 2D and 3D. Fluidization characteristics and mixing of the bed were analyzed for the simulation. Furthermore, the deadzones formed due to multiple jet configurations of the distributor are quantified and their distributions over the plate were analyzed. / Ph. D.

A novel higher-order moment-closure method is applied for the Eulerian treatment of gas-particle multiphase flows characterized by a dilute polydisperse and polythermal particle phase. Based upon the polydisperse Gaussian-moment model (PGM) framework, the proposed model is derived by applying an entropy-maximization moment-closure formulation to the transport equation of the particle-number density function, which is equivalent to the Williams-Boltzmann equation for droplet sprays. The resulting set of first-order robustly-hyperbolic balance laws include a direct treatment for local higher-order statistics such as co-variances between particle distinguishable properties (i.e., diameter and temperature) and particle velocity. Leveraging the additional distinguishing variables, classical hydrodynamic droplet evaporation theory is considered to describe unsteady droplet vaporization. Further, studying turbulent multiphase flow theory, a first-order hyperbolicity maintaining approximation to turbulent flow diffusion-inertia effects is proposed. Investigations into the predictive capabilities of the model are evaluated relative to Lagrangian-based solutions for a range of flows, including aerosol dispersion and fuel-sprays. Further, the model is implemented in a massively parallel discontinuous-Galerkin framework. Validation of the proposed turbulence coupling model is subsequently performed against experimental data, and a qualitative analysis of the model is given for a qualitative liquid fuel-spray problem.

Une étude précise des écoulements gaz-particules poly-solides et réactifs rencontrés dans les lits fluidisés circulants (LFC) appliqués au procédé de Chemical Looping Combustion (CLC) est indispensable pour prédire un point de fonctionnement stable et comprendre l'influence de la réaction et de la polydispersion sur l'hydrodynamique des LFC. Dans ce but, des simulations avec le code NEPTUNE_CFD ont été confrontées aux expériences menées à l'Université Technologique de Compiègne par ALSTOM. Cette modélisation a été validée sur des LFC non réactifs mono-solides et poly-solides. L'influence des caractéristiques des particules et de la position des injecteurs sur l'entrainement de solide est étudiée. Un modèle de prise en compte de la production locale de gaz au cours de la réaction est présenté. L'étude locale de l'écoulement a permis de comprendre l'influence des collisions interparticulaire et de la production locale de gaz sur l'écoulement. Finalement, un point de fonctionnement a été proposé pour le pilote CLC en construction à Darmstadt. Ce travail a montré que NEPTUNE_CFD pouvait prédire l'hydrodynamique de LFC poly-solides à l'échelle du pilote industriel et participer au dimensionnement de centrales de types CLC. / This work deals with the development, validation and application of a model of Chemical Looping Combustion (CLC) in a circulating fluidized bed system. Chapter 1 is an introduction on Chemical Looping Combustion. It rst presents the most important utilizations of coal in the energy industry. Then, it shows that because of the CO2 capture policy, new technologies have been developed in the frame of post-combustion, pre-combustion and oxy-combustion. Then, the Chemical Looping Combustion technology is presented. It introduces multiple challenges: the choice of the Metal Oxide or the denition of the operating point for the fuel reactor. Finally, it shows that there are two specicities for CFD modeling: the influence of the collisions between particles of different species and the local production of gas in the reactor due to the gasication of coal particles. Chapter 2 outlines the CFD modeling approach: the Eulerian-Eulerian approach extended to flows involving different types of particles and coupled with the chemical reactions. Chapter 3 consists in the validation of the CFD model on mono-solid (monodisperse and poly-disperse) and poly-solid flows with the experimental results coming from an ALSTOM pilot plant based at the Universite Tchnologique de Compiegne (France). The relevance of modeling the polydispersity of a solid phase is shown and the influence of small particles in a CFB of large particles is characterized. This chapter shows that the pilot plant hydrodynamics can be predicted by an Eulerian-Eulerian approach. Chapter 4 consists in the validation of the CFD model on an extreme bi-solid CFB of particles of same density but whith a large particle diameter ratio. Moreover, the terminal settling velocity of the largest particles are twice bigger than the fluidization velocity: the hydrodynamics of the large particles are given by the hydrodynamics of the smallest. An experiment performed by Fabre (1995) showed that large particles can circulate through the bed in those operating conditions. Our simulations predicted a circulation of large particles, but underestimated it. It is shown that it can be due to mesh size eect. Finally, a simulation in a periodic box of this case was dened and allowed us to show the major influence of collisions between species. Chapter 5 presents the simulation of a hot reactive CLC pilot plant under construction in Darmstadt (Germany). The simulations account for the chemical reactions and describe its eect on the hydrodynamics. Different geometries and operating conditions are tested.

Le contexte général de cette thèse est la Simulation Numérique Directe des écoulements diphasiques dilués anisothermes. Un accent particulier est mis sur la détermination précise de la dispersion des particules et du transfert de chaleur entre la phase porteuse et dispersée. Cette dernière est décrite à l’aide d’une approche Eulérienne aux moments : le Formalisme Eulérien Mésoscopique (FEM) [41, 123], récemment étendu aux écoulements anisothermes [78]. Le principal objectif de ce travail est de déterminer si ce formalisme est capable de prendre en compte de manière précise l’inertie dynamique et thermique des particules dans un écoulement turbulent, et particulièrement dans une configuration avec un gradient moyen. Le code de calcul utilisé est AVBP. La simulation numérique d’un spray dilué avec une approche Eulerienne soulève des questions supplémentaires sur les méthodes numériques et les modèles employés. Ainsi, les méthodes numériques spécifiques aux écoulements diphasiques implémentées dans AVBP [69, 103, 109] ont été testées et revisitées. L’objectif est de proposer une stratégie numérique précise et robuste qui résiste aux forts gradients de fraction volumique de particule provoqués par la concentration préférentielle [132], tout en limitant la diffusion numérique. Ces stratégies numériques sont comparées sur une série de cas tests de complexité croissante et des diagnostics pertinents sont proposés. Par exemple, les dissipations dues `a la physique et au numérique sont extraites des simulations et quantifiées. Le cas test du tourbillon en deux dimensions chargé en particules est suggéré comme une configuration simple pour mettre en évidence l’impact de l’inertie des particules sur leur champ de concentration et pour discriminer les stratégies numériques. Une solution analytique est aussi proposée pour ce cas dans la limite des faibles nombres de Stokes. Finalement, la stratégie numérique qui couple le schéma centré d’ordre élevé TTGC et une technique de stabilisation, aussi appelée viscosité artificielle, est celle qui fournit les meilleurs résultats en terme de précision et de robustesse. Les paramètres de viscosité artificielle (c'est-à-dire les senseurs) doivent néanmoins être bien choisis. Ensuite, la question des modèles nécessaires pour d´écrire correctement la dispersion des particules dans une configuration avec un gradient moyen est abordée. Pour ce faire, un des modèles RUM (appel´e AXISY-C), proposé par Masi [78] et implémenté dans AVBP par Sierra [120], est validé avec succès dans deux configurations: un jet plan diphasique anisotherme 2D et 3D. Contrairement aux anciens modèles RUM, les principales statistiques de la phase dispersée sont désormais bien prédites au centre et aux bords du jet. Finalement, l’impact de l’inertie thermique des particules sur leur température est étudié. Les résultats montrent un effet important de cette inertie sur les statistiques mettant en évidence la nécessité pour les approches numériques de prendre en compte ce phénomène. Ainsi, l’extension du FEM aux écoulements anisothermes, c’est-à-dire les flux de chaleur RUM (notés RUM HF), est implémentée dans AVBP. L’impact des RUM HF sur les statistiques de température des particules est ensuite évalué sur les configurations des jets 2D et 3D. Les champs Eulériens sont comparés à des solutions Lagrangiennes de référence calculées par B. Leveugle au CORIA et par E. Masi à l’IMFT pour les jets 2D et 3D, respectivement. Les résultats montrent que les RUM HF améliorent la prédiction des fluctuations de température mésoscopique, et dans une moindre mesure la température moyenne des particules en fonction de la configuration. Les statistiques Lagrangiennes sont retrouvées lorsque les RUM HF sont pris en compte alors que les résultats sont dégradés dans le cas contraire. / This work addresses the Direct Numerical Simulation of non-isothermal turbulent flows laden with solid particles in the dilute regime. The focus is set on the accurate prediction of heat transfer between phases and of particles dispersion. The dispersed phase is described by an Eulerian approach : the Mesoscopic Eulerian Formalism [41, 123], recently extended to non-isothermal flows [78]. The main objective of this work is to assess the ability of this formalism to accurately account for both dynamic and thermal inertia of particles in turbulent sheared flows. The CFD code used in this work is AVBP. The numerical simulation of dilute sprays with an Eulerian approach calls for specific modelling and raises additional numerical issues. First, the numerical methods implemented in AVBP for two-phase flows [69, 103, 109] were tested and revisited. The objective was to propose an accurate and robust numerical strategy that withstands the steep gradients of particle volume fraction due to preferential concentration [132] with a limited numerical diffusion. These numerical strategies have been tested on a series of test cases of increasing complexity and relevant diagnostics were proposed. In particular, the two-dimensional vortex laden with solid particles was suggested as a simple configuration to illustrate the effect of particle inertia on their concentration profile and to test numerical strategies. An analytical solution was also derived in the limit of small inertia. Moreover, dissipations due to numerics and to physical effects were explicitly extracted and quantified. Eventually, the numerical strategy coupling the highorder centered scheme TTGC with a stabilization technique –the so called artificial viscosity– proved to be the most accurate and robust alternative in AVBP if an adequate set-up is used (i.e. sensors). Then, the issue of the accurate prediction of particle dispersion in configurations with a mean shear was adressed. One of the RUM model (denoted AXISY-C), proposed by Masi [78] and implemented by Sierra [120], was successfully validated in a two-dimensional and a three-dimensional non-isothermal jet laden with solid particles. Contrary to the former RUM models [63, 103], the main statistics of the dispersed phase were recovered at both the center and the edges of the jet. Finally, the impact of the thermal inertia of particles on their temperature statistics has been investigated. The results showed a strong dependency of these statistics to thermal inertia, pinpointing the necessity of the numerical approaches to account for this phenomenon. Therefore, the extension of the MEF to non isothermal conditions, i.e. the RUM heat fluxes, has been implemented in AVBP. The impact of the RUM HF terms on the temperature statistics was evaluated in both configurations of 2D and 3D jets. Eulerian solutions were compared with Lagrangian reference computations carried out by B. Leveugle at CORIA and by E. Masi at IMFT for the 2D and 3D jets, respectively. Results showed a strong positive impact of the RUM HF on the fluctuations of mesoscopic temperature, and to a lesser extent on the mean mesoscopic temperature depending of the configuration. Neglecting the RUM HF leads to erroneous results whereas the Lagrangian statistics are recovered when they are accounted for.

The evolution of numerical methods and computational facilities allow re- searchers to explore complex physical phenomenons such as multiphase flows. The specific regime of incompressible, turbulent, bubbly two-phase flow (where a car- rier fluid is infused with bubbles or particles) is also receiving increased attention due to it’s appearance in major industrial processes. The main challenges arise from coupling individual aspects of the physics into a unified model and to provide a robust numerical framework. The presented work aimed at to achieve the second part by employing the most frequently used dispersed two-phase flow model and another incompressible, turbulent single phase solver as a base flow provider for coupled Lagrangian or surface tracking tools. Among the numerical techniques, the finite element method is a powerful can- didate when the need arises for multiphysics simulations (for example coupling with an electrochemical module) where the counterpart has a node based ap- proach. Stabilization schemes such as PSPG/SUPG/BULK provide remedies for the pressure decoupling and the inherent instability of the central discretization when applied for convective flow problems. As an alternative to unsteady solvers based upon an explicit or a fully im- plicit nonlinear treatment of the convective terms, a semi-implicit scheme results in a method of second order accurate in both space and time, has absolute linear stability and requires only a single or two linear system solution per time step. The application of the skew symmetric approach to the convective term further stabilizes the solution procedure and in some cases it even prevents divergence. The Eulerian-Eulerian two-phase flow model poses various issues to be over- come. The major difficulty is the density ratio between the phases; for an ordinary engineering problem it is in the order of thousands or more. The seemingly minus- cule differences in the formulation of the stabilizations can cause very different end results and require careful analysis. Volume fraction boundedness is of concern as well, but it is treatable by solving for its logarithm. Since the equations allow jumps (even separation of the phases) in the volume fraction field, discontinuity capturing techniques are also needed. Besides the standard ’spatial’ stabilization temporal smoothing is also necessary, otherwise the limitation in time step size becomes too stringent. Designing a flow solver is one side of the adventure, but verification is equally important. Comparison against analytical solution (such as the single and two- phase Taylor-Green testcase) provides insight and confirmation about the mathe- matical and physical properties. Meanwhile comparing with real life experiments prove the industrialization and usability of a code, dealing with low quality meshes and effective utilization of computer clusters. / Doctorat en Sciences de l'ingénieur et technologie / info:eu-repo/semantics/nonPublished