The purpose of this work is to implement four additional inviscid flux methods in the U2NCLE solver being developed at Mississippi State University. The goal is that some or all of these methods may provide benefits over the current options with respect to accuracy or robustness. These four methods include both the Harten, Lax, Van Leer, Einfeldt (HLLE) and Harten, Lax, Van Leer ? Contact (HLLC) methods as well as the Advection Upstream Splitting Method (AUSM) and its successor AUSM+. The HLL family, which includes both HLLE and HLLC are based on the Riemann problem, which is divided into a number of states. The AUSM family attempts to combine the effects of both flux vector and flux difference splittings to create better schemes. Several simple and complex cases were run with each new method and compared to the methods currently available as well as experimental and analytical results when available. The results of the simple tests showed that all the methods were similarly suited for delivering accurate results on simple cases. In more complex cases, however, the AUSM family proved to be less robust and failed to converge for the final case. The HLLE method showed excellent robustness qualities but seemed to over predict the viscous values in several cases. The HLLC method proved equally as accurate and robust as Roe's Method.

Computational Fluid Dynamics (CFD) has been pivotal in scientific computing, providing critical insights into complex fluid dynamics unattainable through traditional experimental methods. Despite its widespread use, the accuracy of CFD results remains contingent upon the underlying modeling and numerical errors. A key aspect of ensuring simulation reliability is the accurate quantification of discretization error (DE), which is the difference between the simulation solution and the exact solution in the physical world. This study addresses quantifying DE through Error Transport Equations (ETE), which are an additional set of equations capable of quantifying the local DE in a solution. Historically, Richardson extrapolation has been a mainstay for DE estimation due to its simplicity and effectiveness. However, the method's feasibility diminishes with increasing computational demands, particularly in large-scale and high-dimensional problems. The integration of ETE into existing CFD frameworks is facilitated by their compatibility with existing numerical codes, minimizing the need for extensive code modification. By incorporating techniques developed for managing discontinuities, the study broadens ETE applicability to a wider range of scientific computing applications, particularly those involving complex, unsteady flows. The culmination of this research is demonstrated on unsteady discontinuous problems, such as Sod's problem. / Master of Science / In the ever-evolving field of Computational Fluid Dynamics (CFD), the quest for accuracy is paramount. This thesis focuses on discretization error estimation within CFD simulations, specifically on the challenge of predicting fluid behavior in scenarios marked by sudden changes, such as shock waves. At the core of this work lies an error estimation tool known as Error Transport Equations (ETE) to improve the numerical accuracy of simulations involving unsteady flows and discontinuities. Traditionally, the accuracy of CFD simulations has been limited by discretization errors, generally the largest numerical error, which is the difference between the numerical solution and the exact solution. With ETE, this research identifies these errors to enhance the simulation's overall accuracy. The implications of ETE research are far-reaching. Improved error estimation and correction methods can lead to more reliable predictions in a wide range of applications, from aeronautical engineering, where the aerodynamics of aircraft is critical, to plasma science, with applications in fusion and deep space propulsion.

O desenvolvimento de tecnologia na área de turbomáquinas é essencial ao desenvolvimento da indústria nacional e o Laboratório de Engenharia Ambiental e Térmica da Escola Politécnica da Universidade de São Paulo tem compreendido ações para este propósito. Este trabalho tem por objetivo desenvolver um compressor para uma turbina a gás de pequeno porte de 500 kW, primeiro passo para o projeto e construção da turbina como um todo. A partir da análise do ciclo termodinâmico e da análise de adimensionais, o tipo de compressor a ser utilizado foi determinado. Optou-se pelo projeto de um compressor centrífugo. Iniciou-se o projeto através de análise e correlações unidimensionais com previsão de desempenho, definindo algumas geometrias iniciais a serem avaliadas nas fases seguintes. Realizou-se a análise bidimensional do impelidor com a ferramenta computacional Vista TF que utiliza o método de curvatura de linhas de corrente. Por fim, a geometria tridimensional foi definida com uso de simulações de dinâmica de fluidos computacional. De acordo com as simulações, o compressor projetado tem desempenho condizente com os requisitos impostos. / Technology development in turbomachinery is essential to the national industry development and the Laboratory of Environmental and Thermal Engineering of the Polytechnic School of the University of São Paulo is engaged on this purpose. This work intends to design a compressor for a small 500 kW gas turbine, the first step in the whole turbine design and construction. The compressor type was determined from thermodynamical cycle and adimensional analysis. The centrifugal type compressor was chosen. The design was initialized using one-dimensional analysis and correlations with performance prediction models, defining initial geometries to be evaluated in the upcoming design phases. The impeller was analyzed with a two dimensional computational tool named Vista TF, which uses the streamline curvature method. The tridimensional geometry was defined using computational fluid dynamics. According to the simulations, the design compressor performs satisfying the imposed requirements.

Este trabalho apresenta uma metodologia para predição da distribuição de pressão sobre uma asa bi-trapezoidal genérica usando redes neurais artificiais. O toolbox de redes neurais do MatLab® foi utilizado para o treinamento e validação das redes neurais e os conjuntos de treinamentos foram obtidos por meio do software BLWF® versão 28 (Boundary Layer Wing-Fuselage) um código CFD (Computacional Fluid Dynamics) de potencial completo com correção de camada limite. Levando em consideração o nível de complexidade do problema, optou-se por dividir o estudo em três etapas de desenvolvimento. Inicialmente, uma rede neural foi treinada considerando apenas as variáveis de condição de voo e de forma em planta. Resultados promissores motivaram a criação de uma segunda rede neural, mais genérica, na qual foram adicionadas variáveis de três perfis distribuídos ao longo da asa. Porém apenas um desses perfis era variável enquanto que os demais eram parametrizados com relação à este perfil. Criou-se, por fim, uma rede neural ainda mais genérica, desta vez atentando também para as variáveis dos três perfis de modo independente. Os resultados obtidos mostram que esta metodologia pode ser usada como interessante ferramenta para obtenção de distribuição de pressão, especialmente em projetos de MDO (Multi-Disciplinary Optimization), uma vez que ela possibilita uma predição rápida, precisa e de fácil automatização de pressão em uma asa genérica. / This work shows a method for predicting pressure distribution over a generic bi-trapezoidal wing using artificial neural networks. The MatLab® Neural Network Toolbox was used for the neural network implementation and the training set was obtained using the BLWF® version 28 (Boundary Layer Wing-Fuselage), a full potential CFD (Computational Fluid Dynamics) code with boundary layer correction. The work was divided in three development phase, according with the problem complexibibility level. Initially, a neural network considering only flight conditions and plan form variables was trained. Promising results motivated the generation of a more generic neural network, considering also parameters of three airfoils distributed along the wing spanwise and chordwise. However only one airfoil was variable, the two other were parametrized in relation to the variable airfoil. At last, an even more generic neural network was generated, this time considering also the variables of the three profiles independently. The results show that this methodology can be successfully used as an interesting tool to obtain the pressure distribution, especially on the solution of MDO (Multi-Disciplinary Optimization) problems, since it allows fast prediction, automation facility and accurate measuring of the pressure distribution under a generic wing.

L'un des principaux leviers technologiques permettant le développement industriel de process de production énergétique renouvelable et à haute efficacité, consiste en l'élaboration d'une solution innovante de stockage de l'énergie. Ce système de stockage doit permettre de lisser la période de production et ainsi de suivre au plus près les besoins des consommateurs. Parmi les solutions existantes, le stockage thermique par chaleur latente présente de nombreux avantages qui font qu'aujourd'hui il fait l'objet de plusieurs travaux de recherche et de développement. Cette technologie est basée sur le principe que certaines classes de matériaux, appelés matériaux à changement de phase (MCP), libèrent (transition liquide/solide) ou accumulent (transition solide/liquide) de l'énergie lorsqu'ils sont soumis à un changement de phase. En amont du développement d'un design de stockage, il est essentiel de comprendre et de maitriser les processus thermiques entrant en jeu lors des phases de fusion et de solidification du matériau et cette compréhension passe par le développement de modèles numériques adaptés aux problématiques rencontrées. Dans le cadre de ce manuscrit, la filière technologique qui nous intéresse est celle des centrales solaires à concentration. Porté par l'ADEME dans le cadre du projet STARS (Stockage Thermique appliqué à l'extension de pRoduction d'énergie Solaire thermodynamique), le travail réalisé au sein du LaTEP consiste à analyser les performances d'une solution de stockage via la modélisation de cette dernière en considérant les phénomènes thermiques et hydrauliques. Le travail de modélisation est effectué à l'aide du logiciel de CFD libre de droit OpenFOAM dans lequel est développé et implémenté, par le laboratoire, un module dédié au problème qui nous concerne, à savoir la résolution eulérienne (maillage fixe) des équations de conservation pour un fluide incompressible, en présence d'un changement de phase solide-liquide dominé par des mouvements convectifs (convection-dominated phase change). Concernant les problèmes de transition de phase, diverses méthodes mathématiques et numériques ont été développées pour rendre compte finement de la physique de ces phénomènes. Après avoir effectué une revue de ces dernières dans la première partie du manuscrit, nous avons sélectionné deux formulations que nous avons implémenté dans OpenFOAM. Une fois ce travail réalisé nous avons taché de comparer les résultats renvoyés par ces différentes formulations en les confrontant aux résultats expérimentaux disponibles dans la littérature. Cela nous a permis d'une part de nous conforter dans l'utilisation de nos solveurs et sur la pertinence des résultats obtenus avec ces derniers et d'autre part de mettre en évidence les écarts entre les solutions renvoyées par chaque formulation. Fort de ce constat, nous avons souhaité évaluer l'impact de l'équation d'état utilisée pour relier l'enthalpie et la température, indispensable à la fermeture thermodynamique du système d'équations. Cette comparaison s'est faite par la simulation d'un échangeur type stockage thermique (simulations en 2D) et par l'analyse des performances de ce dernier lors des phases de stockage, de déstockage et au cours de plusieurs séries de cycles. Les résultats obtenus nous ont permis de conclure sur l'importance d'une bonne caractérisation des MCP afin de pouvoir modéliser leur comportement au plus juste via la formulation mathématique et la loi d'état la plus adaptée / A major technological lever to the industrial development of renewable energy production processes with high efficiency, is the development of an innovative solution to store the energy. This storage device should help to smooth the production period and to follow closely the demand. Among the existing solutions, latent heat thermal storage has many advantages that make today it is the subject of several research and development projects. This technology is based on the principle that certain classes of material, called phase change materials (PCMs), release (during liquid to solid transition) or accumulate (during solid to liquid transition) energy when subjected to a phase change. Upstream of the development of a new storage design, it is essential to understand and master the thermal processes involved in the melting and solidification phase of the material and this knowledge comes through the development of numerical models adapted to the problems encountered. In the particular context of this Phdthesis, the technological process that interests us is that of CSP (Concentrated Solar Power). Funded by ADEME under the STARS Project (Thermal STorage Applied to the expansion of pRoduction of thermodynamic Solar energy), the work done by the LaTEP is to analyze the performance of a storage solution by modeling the latter, considering the thermal and hydraulic phenomena. The modeling work is done with the free source OpenFOAM CFD software in which is developed and implemented by the laboratory, a new module dedicated to the problem we are concerned, namely the resolution of Eulerian (fixed grid) conservation equations for an incompressible fluid in the presence of a solid-liquid phase change dominated by convective motions. Regarding the phase transition problems, various mathematical and numerical methods have been developed to finely consider the physics of these phenomena. After conducting a review of methods in the first part of the Phd thesis, we selected two formulations that we have implemented in OpenFOAM. Once this work done, we have managed to compare the results returned from these formulations by comparing them with experimental results available in the literature and also with analytical cases. This allowed us firstly to strengthen us in the use of our solvers and the accuracy of the obtained results and secondly to highlight the differences between the solutions returned by each formulation. After that, we wanted to assess the impact of the equation of state used to connect the enthalpy to the temperature, essential for closing the thermodynamic equations. This comparison was made by the simulation of a thermal storage exchanger (2D simulation) and by analyzing the performance of this latter during the charge phase, the discharge one and during several series of cycles. The obtained results allowed us to conclude about the importance of a good characterization of PCM in order to model their behavior as accurately via the mathematical formulation and the most suitable state law

The gas turbine has an important role for the energy distribution due to its stability and flexibility. By increasing turbine inlet temperature (TIT) an increased thermal efficiency of the turbine can be achieved. The biggest limitation of the TIT is the material of the turbine components. To avoid this limitation, cooling is needed in the first stages of the turbine by air from the compressor. The downside of the cooling is the decrease of efficiency with excess of cooling air. To achieve an optimum cooling flow, the designing process is important. One major tool in the designing process is simulations by Computational Fluid Dynamics (CFD). For optimum and correct cooling design, the CFD simulations needs to accurate predict the temperature transport through the turbine. Therefore, this study focused to estimate the accuracy of different CFD methods in predicting the temperature distribution through a 1.5 stage turbine with experimental results. The CFD simulations were done by using Ansys CFX and divided into two study cases with steady RANS. One with different turbulence models;  –, Wilcox –  and SST – . The other with two different simulation approaches of interfaces for frame change; Mixing plane and Frozen rotor. All simulations included two configurations of swirlers clocking for interest of their differences within the turbine and validation of the CFD simulations; Passage (PA) and Leading Edge (LE) clockings. The experimental results showed a formation of gradually more uniformed temperature profile with the fluid. This could not be seen in the same extend with any of the simulations. The temperature difference between the hot and cold section with all simulations were marginally decreased in comparison of the measurements. All results with steady RANS simulations tended to over and under predict the temperatures of the hot respectively cold sections within the fluid flow through the turbine. This occurred already after the first stage guide vanes and the difference from the measurements increased after the first stage rotor. This since the steady RANS tended to under predict the mixing process through the turbine. Differences between the turbulence models were noticeable after the rotor blades, where the   – turbulence model predicted most mixing of the evaluated turbulence models but badly compared to the measurements. Another outcome from this study was that the frozen rotor interface with several positions of the rotor blades did not stated better results compared to mixing plane interface for temperature distribution in axial turbines. On the other hand, one simulation of one position of the rotor with frozen rotor interface could be used to simulate an approximatively similar circumferential average temperature as the mixing plane with better convergence with the disadvantage of bigger domain. The gas turbine has an important role for the energy distribution due to its stability and flexibility. By increasing turbine inlet temperature (TIT) an increased thermal efficiency of the turbine can be achieved. The biggest limitation of the TIT is the material of the turbine components. To avoid this limitation, cooling is needed in the first stages of the turbine by air from the compressor. The downside of the cooling is the decrease of efficiency with excess of cooling air. To achieve an optimum cooling flow, the designing process is important. One major tool in the designing process is simulations by Computational Fluid Dynamics (CFD). For optimum and correct cooling design, the CFD simulations needs to accurate predict the temperature transport through the turbine. Therefore, this study focused to estimate the accuracy of different CFD methods in predicting the temperature distribution through a 1.5 stage turbine with experimental results. The CFD simulations were done by using Ansys CFX and divided into two study cases with steady RANS. One with different turbulence models;  –, Wilcox –  and SST – . The other with two different simulation approaches of interfaces for frame change; Mixing plane and Frozen rotor. All simulations included two configurations of swirlers clocking for interest of their differences within the turbine and validation of the CFD simulations; Passage (PA) and Leading Edge (LE) clockings. The experimental results showed a formation of gradually more uniformed temperature profile with the fluid. This could not be seen in the same extend with any of the simulations. The temperature difference between the hot and cold section with all simulations were marginally decreased in comparison of the measurements. All results with steady RANS simulations tended to over and under predict the temperatures of the hot respectively cold sections within the fluid flow through the turbine. This occurred already after the first stage guide vanes and the difference from the measurements increased after the first stage rotor. This since the steady RANS tended to under predict the mixing process through the turbine. Differences between the turbulence models were noticeable after the rotor blades, where the   – turbulence model predicted most mixing of the evaluated turbulence models but badly compared to the measurements. Another outcome from this study was that the frozen rotor interface with several positions of the rotor blades did not stated better results compared to mixing plane interface for temperature distribution in axial turbines. On the other hand, one simulation of one position of the rotor with frozen rotor interface could be used to simulate an approximatively similar circumferential average temperature as the mixing plane with better convergence with the disadvantage of bigger domain. / Gasturbinen har en viktig roll i nutida och framtida energidistribution för elektricitet på grund av dess stabilitet samt flexibilitet. Genom att öka temperaturen in till turbinen ökar den termiska effektiviteten. Den största begränsning av denna temperaturökning är materialen av komponenterna i turbinen. För att kringgå detta används kylning i turbinen med luft från kompressorn. Effektiviteten kan däremot minskas vid överdriven användning av kylluft och därav är designen av kylningen viktig för optimal användning av kylluft. Ett verktyg som oftast används vid design av turbiner är simuleringar med Computational Fluid Dynamics (CFD). För att uppnå en optimal design av kylningen behöver CFD simuleringarna korrekt prediktera temperaturtransporten genom turbinen. Därför fokuserade denna studie på att uppskatta och validera olika CFD metoders förmåga att prediktera temperaturtransporten genom en 1,5 stegs axiell turbin med experimentella resultat. Stationära CFD simuleringar gjordes med RANS av olika turbulensmodeller; k – ε, Wilcox k – ω and SST k – ω. Dessutom jämfördes två olika sätt att simulera gränssnittet mellan stationära och roterande domän; Mixing plane och Frozen rotor. Samtliga simuleringsmetoder inkluderade två olika konfigurationer av virvlarepositioner; Passage (PA) och Leading edge (LE) klockningar. Experimentella resultat visade en stegvis mer enhetlig temperaturprofil med fluidflödet genom turbinen. Detta sågs dock inte i samma utsträckning i någon av simuleringarna. Temperaturskillnaden mellan de varma och kalla stråken i samtliga simuleringar minskade marginellt i jämförelse med de experimentella resultaten. Samtliga resultat med stationära RANS simuleringar tenderade att över och under prediktera temperaturen av de varma respektive kalla stråken. Detta inträffade redan efter förstastegsledskenorna, där skillnaden från de uppmätta temperaturerna ökade över första stegs rotor. Detta på grund av att mixningen i fluiden under predikterades. Skillnader mellan de olika turbulensmodellerna var synliga efter första stegs rotor där  – turbulensmodell predikterade mest mixning av samtliga simuleringar av turbulensmodeller. Däremot predikterade den marginellt bättre i jämförelse med mätningarna. Andra resultat från denna studie var att gränssnittet med frozen rotor med flera positioner inte anger bättre mixning av fluiden genom rotordomänen än vad gränssnittet med mixing plane där liknande radiella temperaturprofiler fås. Däremot gav en simulering med en position av rotorn liknande resultat med radiellt fördelade temperaturer som mixing plan och skulle kunna användas för approximativa simuleringar med bättre konvergens.

Dans le contexte actuel de croissance continue de la demande mondiale en combustible fossile, la sécurité de la production, du transport, ainsi que du stockage de l'énergie est un défi majeur de ce début de XXIème siècle. Les produits manipulés étant extrêmement volatils et inflammables, les éventuelles fuites qui peuvent survenir malgré les lourdes mesures de sécurité mises en place, peuvent engendrer des explosions désastreuses. Il existe donc un fort besoin d'être capable de prédire ces explosions afin de limiter les dégâts potentiels et d'assurer la sécurité des personnes et des biens. Dans cette optique, l'augmentation régulière des puissances de calcul permet à la CFD (Computational Fluid Dynamics) de se présenter comme une alternative intéressante aux expériences qui peuvent s'avérer couteuses et dangereuses. Les explosions sont des phénomènes multi-physiques qui sont principalement dirigés par la turbulence et la combustion et qui prennent place sur une très large gamme d'échelles nécessitant ainsi d'être modélisées. Aujourd'hui, des codes basés sur une approche URANS (Unsteady Reynolds Averaged Navier Stokes) sont généralement utilisés afin de simuler des explosions de gaz dans des configurations à échelle industrielle. Cependant, l'émergence de la LES (Large Eddy Simulation), qui a déjà montré son potentiel à donner des prédictions plus fiables que le URANS sur des configurations instationnaires complexes, ouvre de nouvelles perspectives pour le domaine de la sécurité explosion. Le but principal de cette thèse est d'évaluer l'apport des méthodes LES et de développer une méthodologie pour la prédiction des phénomènes réactifs turbulents transitoires que sont les explosions. Tout au long de cette étude, un intérêt particulier a été porté à l'approfondissement de la compréhension des phénomènes d'explosion ainsi qu'à la mise en valeur des points cruciaux de modélisation qui permettent une reproduction correcte des phénomènes considérés. Notre approche peut alors se résumer en deux temps : - Dans un premier temps nous nous sommes concentrés sur l'étude LES des déflagrations dans une chambre de combustion de petite échelle : la configuration expérimentale de l'Université de Sydney. La LES associée à un modèle de flamme épaissie a ainsi été appliquée à cette configuration à l'aide du code AVBP (développé par le CERFACS et l'IFP-EN) et a permis de mettre en place une méthodologie de calcul. Une étude de Quantification d'Incertitude (UQ) a ensuite été réalisée sur ces simulations afin d'évaluer la fiabilité de ces résultats, ce qui est primordial dans ce contexte d'étude de sécurité. - Dans un second temps, le but a été d'extrapoler les résultats obtenus sur la configuration de petite échelle à des configurations de plus grande échelle, plus représentatives des configurations industrielles réelles de plateformes pétrolières ou de dépôts de carburants qui constituent l'objectif final visé. Une campagne expérimentale a ainsi été lancée afin de construire des répliques de la configuration de Sydney à des échelles plus importantes et de les étudier numériquement grâce à la méthodologie LES mise en place sur la configuration de petite échelle. Afin de replacer notre étude dans le contexte actuel et de le relier à l'état-de-l'art en matière d'étude de risque d'explosions, d'autres calculs de ces configurations d'explosion ont été réalisés en parallèle de l'étude LES, premièrement avec un code phénoménologique développé dans le cadre de cette thèse, ainsi qu'avec le code URANS FLACS. Ceci a permis de mettre en évidence leurs limitations ainsi que l'apport de la LES pour ce type d'étude. / Within the current context of increasing global demand of fossil fuels, the safety of production, transport, and storage of energy is a major challenge of this early 20th century. The products used are highly volatile and flammable. The eventual leakages which could occur (in spite of the strong safety measures) can lead to dramatic explosions. As a consequence, we need to be able to predict these explosions in order to limit their potential damages and ensure the human and material safety. To this end, the growing of computational power makes the CFD (Computational Fluid Dynamics) an interesting alternative to experiments which can be expensive and dangerous. Gas explosions are multi-physics phenomena mainly driven by turbulence and combustion which take place over a wide range of scales and need to be modeled. Today, CFD codes based on the URANS (Unsteady Reynolds Averaged Navier Stokes) approach are usually used to simulate gas explosions at industrial scale. However, the emergence of LES (Large Eddy Simulation) has already shown its potential to give more accurate prediction than URANS on complex unsteady configurations. This opens new perspectives for the field of explosion safety. The main aim of this thesis is to assess the benefits of using LES for gas explosion studies and to develop a methodology to predict these unsteady turbulent reactive phenomena. All along this thesis, efforts have been made to increase our understanding of explosions and to highlight key points of modeling which enable an accurate reproduction of the considered phenomena. Our work can be summed up in two parts: - First, the focus was on the LES study of deflagrations in a small scale explosion chamber: the experimental setup of the University of Sydney. LES combined with a thickened flame approach has been applied to this configuration with the AVBP code (developed by CERFACS and IFP-EN) and enabled to set up a computation methodology. An Uncertainty Quantification (UQ) study has then been performed over these simulations in order to asses the reliability of these results, which is essential in this context of safety related studies. - Then, the aim was to extend the conclusions obtained for the small scale configurations to larger scales, more representative of real industrial cases of oil platforms or fuel storage facilities which are the final aim. An experimental campaign has consequently be launched in order to build replicas of the Sydney test-case at larger scales and to study them numerically using the LES methodology developed with the small scale configuration. In order to put our study back into the current context and to link it to the state-of-the-art of explosion risk assessment studies, several other simulations of these explosion configurations have been performed, first using a 0D phenomenological code developed in the framework of this thesis, and then using the URANS CFD code FLACS. This enabled to highlight the limitations of these approaches and the advantages of LES for this type of study.

Este trabalho apresenta uma metodologia para predição da distribuição de pressão sobre uma asa bi-trapezoidal genérica usando redes neurais artificiais. O toolbox de redes neurais do MatLab® foi utilizado para o treinamento e validação das redes neurais e os conjuntos de treinamentos foram obtidos por meio do software BLWF® versão 28 (Boundary Layer Wing-Fuselage) um código CFD (Computacional Fluid Dynamics) de potencial completo com correção de camada limite. Levando em consideração o nível de complexidade do problema, optou-se por dividir o estudo em três etapas de desenvolvimento. Inicialmente, uma rede neural foi treinada considerando apenas as variáveis de condição de voo e de forma em planta. Resultados promissores motivaram a criação de uma segunda rede neural, mais genérica, na qual foram adicionadas variáveis de três perfis distribuídos ao longo da asa. Porém apenas um desses perfis era variável enquanto que os demais eram parametrizados com relação à este perfil. Criou-se, por fim, uma rede neural ainda mais genérica, desta vez atentando também para as variáveis dos três perfis de modo independente. Os resultados obtidos mostram que esta metodologia pode ser usada como interessante ferramenta para obtenção de distribuição de pressão, especialmente em projetos de MDO (Multi-Disciplinary Optimization), uma vez que ela possibilita uma predição rápida, precisa e de fácil automatização de pressão em uma asa genérica. / This work shows a method for predicting pressure distribution over a generic bi-trapezoidal wing using artificial neural networks. The MatLab® Neural Network Toolbox was used for the neural network implementation and the training set was obtained using the BLWF® version 28 (Boundary Layer Wing-Fuselage), a full potential CFD (Computational Fluid Dynamics) code with boundary layer correction. The work was divided in three development phase, according with the problem complexibibility level. Initially, a neural network considering only flight conditions and plan form variables was trained. Promising results motivated the generation of a more generic neural network, considering also parameters of three airfoils distributed along the wing spanwise and chordwise. However only one airfoil was variable, the two other were parametrized in relation to the variable airfoil. At last, an even more generic neural network was generated, this time considering also the variables of the three profiles independently. The results show that this methodology can be successfully used as an interesting tool to obtain the pressure distribution, especially on the solution of MDO (Multi-Disciplinary Optimization) problems, since it allows fast prediction, automation facility and accurate measuring of the pressure distribution under a generic wing.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-03-30T13:38:45Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Tese_Mikele Cândida Sousa de Sant'Anna.pdf: 2684133 bytes, checksum: 61e61bb0edcf743d1f2af9c55724a92e (MD5) / Made available in DSpace on 2016-03-30T13:38:45Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Tese_Mikele Cândida Sousa de Sant'Anna.pdf: 2684133 bytes, checksum: 61e61bb0edcf743d1f2af9c55724a92e (MD5) Previous issue date: 2015-11-18 / CAPEs / O presente trabalho trata da simulação numérica de um gaseificador de leito fluidizado borbulhante fazendo uso da CFD para o sistema composto por gás-biomassa-areia. Inicialmente, simulou-se os sistemas gás-areia e gás-biomassa. O modelo computacional empregado foi validado empregando-se os resultados experimentais de Taghipuor et al. (2005).A seguir, foi realizado um planejamento fatorial 23, onde se variou a massa específica e o diâmetro da partícula e o percentual de biomassa na fase sólida. Para realizar as simulações foram utilizados os softwares ANSYS CFX 15.0 e ANSYS FLUENT 15.0, adotando-se a abordagem euleriana, com a Teoria Cinética de Escoamento Granular. As seguintes velocidades superficiais do gás foram testadas: 0,03, 0,1, 0,38 0,46 e 0,51 m.s-1. Para o sistema gás-areia, o leito permaneceu fixo nas velocidades de 0,03 e 0,10 m.s-1. Aos 2,50 s de simulação transiente, o leito encontrava-se fluidizado para as velocidades maiores ou iguais a 0,38 m.s-1 e assim permaneceu alcançando um estado pseudo-estacionário. No sistema gás-biomassa, o leito manteve-se fixo apenas na velocidade de 0,03 m.s-1. Dois sistemas foram testados com três componentes (gás-areia-biomassa) diferenciando-se entre si pelos tamanhos das partículas de areia e biomassa. Para grandes diferenças entre estes tamanhos, o sistema apresentou segregação durante a fluidização. No sistema com menor diferença nestes tamanhos, a fluidização ocorreu mais facilmente, uma vez que os efeitos de segregação foram atenuados. Foram obtidos perfis de fração volumétrica do gás, areia e biomassa para as 17 condições do planejamento fatorial, bem como um modelo que prediz a expansão do leito em sistemas fluidizados. O ensaio que apresentou maior altura final do leito (0,50 m), mantendo-se em regime borbulhante, foi aquele com 15% de partículas de biomassa com 375 m de diâmetro e 85% de areia, sendo, portanto, uma condição ótima para a fluidização. / This work has studied a bubbling fluidized bed gasifier though numerical simulation using computational fluid dynamics (CFD) for the system composed of gas - biomass - sand. Initially, gas-sand and gas-biomass systems were simulated. The computer model used was validated employing experimental results from Taghipuor et al. (2005) .Then, a 23 factorial design was employed, where the density, the particle diameter and the the fraction of biomass were varied. All simulations were carried out using ANSYS CFX 15.0 and ANSYS FLUENT 15.0. An eulerian approach coupled to the Kinetic Theory of Granular Flow were used. The following gas superficial velocities were tested: 0.03, 0.1, 0.38, and 0.46 0.51 m s-1. For gas-sand system, a fixed bed was obtained for gas velocities of 0.03 and 0.10 m s-1. After 2.50 s of transient simulation, the bed became fluidized for gas velocities greater or equal to 0.38 m s-1 staying in a pseudo-steady state. For the biomassgas system, the bed remained fixed only at the speed of 0.03 m s-1. Two systems were tested using the three components (gas, sand and biomass) differing from each other only by the size of sand and biomass particles. For high differences between these sizes, the system showed segregation during fluidization. In the system with lower size difference, the fluidization occurred more easily, since the segregation effects were attenuated. Volumetric fraction profiles of gas, sand and biomass were obtained for the 17 factorial design conditions used as well as a model that predicts the bed expansion in fluidized systems. The assay that showed higher final height of the bed (0.50 m) staying in a bubbling regime was one with 15% biomass particles with 375 mm in diameter and 85% of sand, being, therefore, a good condition to carry out fluidization.

Orientador: José Roberto Nunhez / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Química / Made available in DSpace on 2018-08-17T07:13:04Z (GMT). No. of bitstreams: 1 Olino_AnaLeticiaMonteiro_M.pdf: 2635476 bytes, checksum: 1044b4e4ad54faed575660cb81c06763 (MD5) Previous issue date: 2010 / Resumo: Um impelidor ideal para suspensões deve prover a suspensão completa ou homogeneização utilizando o mínimo de energia. O fluxo em um tanque agitado depende da geometria do impelidor, do diâmetro e da localização deste, do diâmetro e do fundo do tanque e da geometria de internos. A Fluidodinâmica Computacional (CFD) é uma ferramenta poderosa para predizer fluxos tridimensionais e distribuição de concentração de sólidos produzidos por impelidores de qualquer configuração geométrica. Este estudo pretende aumentar a capacidade de bombeamento de um impelidor de bombeamento axial, chamado KPC, enquanto mantém o baixo consumo de potência. Para avaliar a capacidade de bombeamento do KPC, diferentes ângulos entre a raiz e a ponta das pás foram estudados. O ângulo da raiz foi mantido em 45º, enquanto o ângulo da ponta da pá foi modificado, resultando em diferentes pás. As simulações em CFD foram feitas com água e com água e areia. O modelo de turbulência utilizado foi o Shear Stress Transport (SST) e foi utilizada a técnica de simulação Sliding Grid. O modelo escolhido para as simulações foi validado experimentalmente. O descolamento da camada limite na ponta das pás dos impelidores foi visualizado, e o objetivo principal do trabalho, encontrar o impelidor KPC otimizado, foi atingido. Este impelidor possui 45º na raiz da pá e 10º na ponta da pá. O impelidor otimizado confirmou seu melhor desempenho quando comparado ao impelidor inicial na simulação de uma suspensão de areia e água / Abstract: An ideal impeller for solid suspension should provide complete suspension or homogenization consuming a minimum of energy. The flow in a stirred tank depends on impeller design, diameter and the location of impellers, vessel diameter, bottom design and internals. Computational Fluid Dynamics (CFD) is a powerful tool to predict the three dimensional flow and solids concentration distribution produced by different impellers of any geometric configuration. This study aims to improve the pumping of an axial pumping impeller, named KPC, while maintaining low power consumption. In order to evaluate the pumping capacity of the KPC impeller, differents angles between the root and the tip of the blades have been studied. The angle at the root was maintained at 45º, and the angle at the blade tip has been modified, resulting in different blades. CFD simulations were performed for water and for water and sand. The model used the Shear Stress Transport (SST) turbulence model and the Sliding Grid simulation strategy. The model was validated against experimental data. The impelller blade flow separation was visualized, and the main objective of the work, which was to find the optimized KPC impeller, was reached. This impeller possess 45º in the blade root and 10º in the tip of the blade. The optimized impeller confirmed its better performance when compared to the initial impeller in a simulation of a suspension of sand and water / Mestrado / Desenvolvimento de Processos Químicos / Mestre em Engenharia Química