No description available.

This work considers the organization and performance of computations on parallel computers of tree algorithms for the N-body problem where the number of particles is on the order of a million. The N-body problem is formulated as a set of recursive equations based on a few elementary functions, which leads to a computational structure in the form of a pyramid-like graph, where each vertex is a process, and each arc a communication link. The pyramid is mapped to three different processor configurations: (1) A pyramid of processors corresponding to the processes pyramid graph; (2) A hypercube of processors, e.g., a connection-machine like architecture; (3) A rather small array, e.g., $2 \\times 2 \\ times 2$, of processors faster than the ones considered in (1) and (2) above. Simulations of this size can be performed on any of the three architectures in reasonable time.

Over the past 15 years, modern PC graphics cards (GPUs) have changed from being pure graphics accelerators into parallel computing platforms.Several new parallel programming languages have emerged, including NVIDIA's parallel programming language for GPUs (CUDA). This report explores two related problems in parallel: How well-suited is CUDA for implementing algorithms that utilize non-trivial data structures?And, how does one develop a complex algorithm that uses a CUDA system efficiently? A guide for how to implement complex algorithms in CUDA is presented. Simulation of a dense 2D particle system is chosen as the problem domain foralgorithm optimization. Two algorithmic optimization strategies are presented which reduce the computational workload when simulating theparticle system. The strategies can either be used independently, or combined for slightly improved results. Finally, the resultingimplementations are benchmarked against a simpler implementation on a normal PC processor (CPU) as well as a simpler GPU-algorithm. A simple GPU solution is shown to run at least 10 times faster than a simple CPU solution. An improved GPU solution can thenyield another 10 times speed-up, while sacrificing some accuracy.

As imagens de ressonância magnética ponderadas por difusão retratam a difusividade de moléculas de água presentes em tecidos biológicos. Em estruturas biológicas altamente organizadas e compactas como fibras nervosas e musculares, a difusividade é maior na direção paralela às fibras do que perpendicularmente às mesmas. Essa propriedade permite a reconstrução digital das trajetórias das fibras, técnica denominada tractografia, representando uma das poucas formas não invasivas de investigação da conectividade anatômica e organização estrutural do cérebro e do coração. A metodologia de tractografia mais difundida faz uso da integração numérica da direção principal de difusividade para reconstruir essas trajetórias. Porém, esta técnica apresenta problemas como o erro intrínseco a métodos de integração numérica e o erro associado a regiões de incerteza nos dados de difusividade. Uma metodologia considerada mais robusta consiste da modelagem da tractografia como a simulação de um sistema de partículas. No entanto, tal metodologia possui diversos parâmetros que precisam ser otimizados para cada caso e apresenta alta complexidade computacional. Esta dissertação apresenta uma metodologia de tractografia global baseada em sistema de partículas, mas com custo computacional reduzido pois evita passos desnecessários da otimização para reconstrução das trajetórias. Avaliamos sua acurácia em conjuntos de dados com graus de complexidade crescentes utilizando imagens sintéticas de difusão construídas digitalmente e em imagens reais de difusão do miocárdio humano. Nesses testes foram observadas reduções no consumo de tempo e maior acurácia para metodologia global apresentada com relação às descritas na literatura. Essa metodologia possui o potencial de evidenciar a organização e arquitetura de diversos tecidos do corpo humano com maior fidelidade e menor tempo de reconstrução. / The diffusion magnetic resonance imaging portray the diffusivity of water molecules present on biological tissues. High organized and compact biological structures like neuronal fibres and muscles present higher diffusivity parallel to the fibres than perpendicular to those. This property allows the digital reconstruction of fibres trajectories, technique named tractography, being one of the few non invasive ways of investigation of the anatomical connectivity and structural organization of the brain and heart. The most common tractography methodology uses numerical integration following the main diffusion direction in order to reconstruct trajectories. Yet, this technique is prone to intrinsic error to numeric integration and the error associated to uncertainty regions on diffusion data. A methodology considered more robust to such problems consists on modelling tractography as a particle system simulation. However, such methodology has several parameters that require fine tuning for each case and has a high computational complexity. This dissertation presents a global tractography methodology based on particle system but at lower computational cost because of the avoidance of unnecessary optimization steps on trajectory reconstruction. We evaluate its accuracy on synthetic diffusion datasets of increasing complexity and on real human cardiac diffusion images. Theses tests evidence reduced time consumption and increased accuracy for the presented methodology compared to the ones described in the literature. This methodology has the potential to reveal the organization and architecture of several tissues from the human body with higher fidelity and lower reconstruction time.

Real time image guided radiotherapy has been proposed by integrating an in-line 6 MV linear accelerator (linac) to a magnetic resonance (MR) imager in either a parallel or transverse configuration. In either configuration, magnetic interference in the linac is caused by its immersion in the magnetic fringe fields of the MR imager. Thus in order to minimize the effect of the magnetic interference, investigations on linac performance in external magnetic fields was completed through various simulations. Finite difference and finite element methods as well as particle simulations were performed in order to design an electron gun and an in-line 6 MV linac waveguide. Monte Carlo simulations provided calculations of dose distributions in a water tank from the derived electron phase space at the linac target. The entire simulation was validated against measurements taken from a commercial medical in-line 6 MV linac, other simulation programs, and theory. The validated linac simulation was used to investigate linac performance in external magnetic fields. The results of this investigation showed that the linac had a much lower tolerance to transverse magnetic fields compared to longitudinal fields. While transverse magnetic fields caused a global deflection of the electron beam away from the central axis of the waveguide, longitudinal fields changed the optics of the electron gun in a suboptimal way. Both transverse and longitudinal magnetic fields caused excessive beam loss if the field strength was large enough. Heating caused by excessive beam loss in external magnetic fields was shown to have little effect on the resonant frequency of the waveguide, and any change in dosimetry, if it existed, was shown to be easily corrected using the jaws or multileaf collimators (MLCs). It was determined that the low-field parallel configuration linac-MR system investigated did not require any magnetic shielding, so the focus was on shielding the transverse configuration. Using beam loss, MLC motor tolerance to magnetic fields, and MR imager homogeneity as constraints, passive and active magnetic shielding was designed and optimized. Thus through the parallel configuration, or using magnetic shielding, magnetic interference has been reduced to within the linac operational tolerance. / Medical Physics

No description available.

No description available.

La combustion d’hydrocarbures représente encore aujourd’hui une part très majoritaire de la production d’énergie dans le monde, et en particulier dans l’industrie aéronautique. La plupart des brûleurs industriels sont alimentés par un carburant sous forme liquide, injecté directement dans la chambre de combustion, générant ainsi de fortes interactions entre le spray, l’écoulement turbulent et la flamme. Dans le but d’acquérir une meilleure compréhension de la structure complexe des flammes de spray, une étude numérique a été réalisée sur la configuration du brûleur diphasique KIAI, caractérisée de façon précise et complète expérimentalement. Une approche de type simulation des grandes échelles a été utilisée pour simuler la phase gazeuse tandis que la phase liquide était résolue selon un formalisme Lagrangien déterministe (LES-DPS). L’analyse détaillée de la structure de flamme de spray permet de mettre en exergue le rôle important de la combustion non prémélangée dans ce type de flamme. Cela a motivé dans une seconde étape le développement d’une nouvelle approche pour modéliser les flammes de diffusion turbulentes. Le modèle présenté s’appuie sur la réponse des flammes de diffusion laminaires au maillage, à l’étirement et au plissement. Le dégagement de chaleur global de la flamme a été analysé dans des configurations de complexité croissante, et la capacité du modèle à le décrire a été évaluée. / The combustion of hydrocarbons still represents a major part of the worldwide production of energy, especially in the aeronautical industry. Most industrial burners are fed with liquid fuel that is directly injected in the combustion chamber, generating a strong interaction between the spray, the turbulent flow and the flame. In order to provide a better understanding of turbulent spray flame complex structures, a numerical study has been performed on the two-phase flow burner KIAI which has been experimentally fully characterized. Numerical simulations consist of Large Eddy Simulation coupled to Discrete Particle Simulation for the dispersed phase (LES-DPS). A detailed analysis of the flame structure shows that non-premixed combustion plays an important role in this type of spray flame. This motivates, in a second step of the present work, the development of a new approach to model turbulent diffusion flames. The model is based on the response to the mesh, strain rate and wrinkling. The global flame heat release is analyzed through configurations of increasing complexity and the capacity of the model to describe it is evaluated.

As imagens de ressonância magnética ponderadas por difusão retratam a difusividade de moléculas de água presentes em tecidos biológicos. Em estruturas biológicas altamente organizadas e compactas como fibras nervosas e musculares, a difusividade é maior na direção paralela às fibras do que perpendicularmente às mesmas. Essa propriedade permite a reconstrução digital das trajetórias das fibras, técnica denominada tractografia, representando uma das poucas formas não invasivas de investigação da conectividade anatômica e organização estrutural do cérebro e do coração. A metodologia de tractografia mais difundida faz uso da integração numérica da direção principal de difusividade para reconstruir essas trajetórias. Porém, esta técnica apresenta problemas como o erro intrínseco a métodos de integração numérica e o erro associado a regiões de incerteza nos dados de difusividade. Uma metodologia considerada mais robusta consiste da modelagem da tractografia como a simulação de um sistema de partículas. No entanto, tal metodologia possui diversos parâmetros que precisam ser otimizados para cada caso e apresenta alta complexidade computacional. Esta dissertação apresenta uma metodologia de tractografia global baseada em sistema de partículas, mas com custo computacional reduzido pois evita passos desnecessários da otimização para reconstrução das trajetórias. Avaliamos sua acurácia em conjuntos de dados com graus de complexidade crescentes utilizando imagens sintéticas de difusão construídas digitalmente e em imagens reais de difusão do miocárdio humano. Nesses testes foram observadas reduções no consumo de tempo e maior acurácia para metodologia global apresentada com relação às descritas na literatura. Essa metodologia possui o potencial de evidenciar a organização e arquitetura de diversos tecidos do corpo humano com maior fidelidade e menor tempo de reconstrução. / The diffusion magnetic resonance imaging portray the diffusivity of water molecules present on biological tissues. High organized and compact biological structures like neuronal fibres and muscles present higher diffusivity parallel to the fibres than perpendicular to those. This property allows the digital reconstruction of fibres trajectories, technique named tractography, being one of the few non invasive ways of investigation of the anatomical connectivity and structural organization of the brain and heart. The most common tractography methodology uses numerical integration following the main diffusion direction in order to reconstruct trajectories. Yet, this technique is prone to intrinsic error to numeric integration and the error associated to uncertainty regions on diffusion data. A methodology considered more robust to such problems consists on modelling tractography as a particle system simulation. However, such methodology has several parameters that require fine tuning for each case and has a high computational complexity. This dissertation presents a global tractography methodology based on particle system but at lower computational cost because of the avoidance of unnecessary optimization steps on trajectory reconstruction. We evaluate its accuracy on synthetic diffusion datasets of increasing complexity and on real human cardiac diffusion images. Theses tests evidence reduced time consumption and increased accuracy for the presented methodology compared to the ones described in the literature. This methodology has the potential to reveal the organization and architecture of several tissues from the human body with higher fidelity and lower reconstruction time.

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第20089号 / 工博第4256号 / 新制||工||1659(附属図書館) / 33205 / 京都大学大学院工学研究科電気工学専攻 / (主査)教授 大村 善治, 教授 松尾 哲司, 准教授 小嶋 浩嗣 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM