Submitted by Biblioteca do Instituto de Física (bif@ndc.uff.br) on 2017-06-09T19:40:00Z No. of bitstreams: 1 DissertacaoMestrado_FabianaMonteiro.pdf: 63170351 bytes, checksum: c19ab17f1c6a5a21e7a5a6ed4744c86e (MD5) / Made available in DSpace on 2017-06-09T19:40:00Z (GMT). No. of bitstreams: 1 DissertacaoMestrado_FabianaMonteiro.pdf: 63170351 bytes, checksum: c19ab17f1c6a5a21e7a5a6ed4744c86e (MD5) / Conselho Nacional de Desenvolvimento Científico e Tecnológico / Durante os últimos 1000 anos ocorreram significativas mudanças climáticas em nosso planeta, que tem sido caracterizada como uma Pequena Idade do Gelo (LIA). Este foi um período de esfriamento global, que ocorreu após o Período Quente Medieval, estendendo-se entre os séculos XIV e XIX, aproximadamente de 1350 a 1850. A compreensão deste fenômeno, relacionado a mudanças globais, possui grande interesse científico. No Hemisfério Norte há vários trabalhos científicos que relacionam os resultados encontrados com a Pequena Idade do Gelo e o Período Quente Medieval, já no Hemisfério Sul esses eventos são incertos devido a escassez de material datado dessa época. Isto tem provocado uma grande controvérsia, uma vez que embora alguns dos poucos trabalhos existentes tenham identificado que o mesmo período de resfriamento também tenha ocorrido na América do Sul, outros autores discordam que esse período frio possa ser aplicado globalmente. Desta forma, ainda há uma questão em aberto, onde trabalhos que exploram a reconstrução de encostas (superfície do mar), vazão de rios e a produtividade biológica sul americanas, podem contribuir na elucidação deste problema. Dentro deste contexto, esta dissertação propôs uma contribuição no estudo da formação de encostas brasileiras, mais especificamente, na enseada do Pântano Sul, Ilha de Santa Catarina, SC, onde testemunhos sedimentares demonstraram uma variação significativa na composição de seus sedimentos, que podem estar associados a eventos de mudanças climáticas significativas. Para tanto, foi utilizada a técnica de AMS (Accelerator Mass Spectrometry), através da datação de radiocarbono de amostras de conchas e sedimentos, oriundas destes testemunhos sedimentares, numa colaboração entre pesquisadores do Instituto de Física e Departamento de Biologia Marinha, ambos da Universidade Federal Fluminense (UFF). Em adição, esta dissertação apresenta um grande destaque como marco tecnológico, pois é o primeiro trabalho aplicando a técnica AMS, desenvolvida integralmente em um país Latino-Americano. Assim, são apresentados os primeiros resultados para a utilização de um acelerador do tipo SSAMS (Single Stage Accelerator Mass Spectrometry System) de 250kV, recém instalado no Instituto de Física da UFF. Desta forma, a técnica de datação 14C-AMS já pode ser totalmente realizada aqui no Brasil e na América Latina. / During the last 1000 years significant climatic changes have occured in our planet, which have been characterized as a Little Ice Age (LIA). This was a period of global coldness, after the Warm Medieval Period, between the XIV and XIX centuries, aproximately from 1350 to 1850. The understandind of such fenomenum, related to global changes, is of great scientifc interest. In the northern hemisphere there are many works relating its results to the Little Ice Age and the Warm Medieval Period. In the southern hemisphere however, these events are not clear due the lack of events. This has lead to controversy since a few works have identifyed the same cold period in South America, others authors disagree that it was a global phenomenum. This way, there is still an open question, where studies that explores coastal reconsctruction (ocean surface), river flow and the biological productivity in South America can contribute to solving the problem. in this sence, this work proposes a contribution to the study of Brazilian coastal formation, more specifically in the Pantano Sul Inlet, Santa Catarina Island, SC, where sediment cores have shown a significant climatic changes. In this study, we used the Accelerator Mass Spectrometry (AMS) thecniche, for the radiocarbon dating of shells and sediment from the collected cores in a collaborative work between researches from the Physics Institute and the department of Marine Biology, both from the Fluminense Federal University. In addition, this work is a technological mark since it is the first results of Single Stage Mass Spectrometry Accelerator System, recently installed in the Physics Institute of UFF. Therefore, the 14C-AMS dating technique can now be throroughly permormed in Brazil and Latin America.

No description available.

Le travail présenté dans cette thèse a été effectué au sein de la collaboration internationale LHCb qui a conçue et qui exploite un détecteur pour la physique des particules auprès de l’accélérateur proton-proton, le LHC, au CERN à Genève. Ces travaux concerne l’opération de l’expérience dans son ensemble. Ils ont montré toutes leurs forces pendant la première année de prise de données qui a débutée fin 2009. Ils couvrent plusieurs systèmes qui sont très dépendant les uns des autres. Deux systèmes sont plus particulièrement étudiés. Le premier est en charge de la surveillance des faisceaux, du niveau des bruits de fond et de la luminosité. Le second permet la visualisation, l’analyse et l’optimisation des conditions expérimentales. Ces deux systèmes sont fortement interconnectés. En effet, l’amélioration de la qualité des faisceaux de la machine et la diminution du bruit de fond augmentent le nombre de collisions utiles pour la physique. En même temps, comprendre les paramètres clefs qui gouvernent l’opération de l’expérience permet de les optimiser et d’améliorer la qualité des données collectées. / There are two main central topics in the thesis: the LHCb beam, background and luminosity monitoring systems and the LHCb optimization systems of experimental conditions. These systems are heavily connected to each other, as improving the machine beam, background and luminosity conditions will automatically improve global operation by maximizing the ratio of luminosity recorded over signal background. At the same time, improving the operation of the experiment will help improve luminosity, by studying more accurately the beam and background conditions and therefore improving the LHC machine settings. In this thesis, the systems to accomplish the requirements of these two main topics are described in detail.

This thesis investigates the possibility of porting a neural network model trained and modeled in TensorFlow to a low-power AI inference accelerator for IoT edge computing. A slightly modified LeNet-5 neural network model is presented and implemented such that an input frequency of 10 frames per second is possible while consuming 4mW of power. The system is simulated in software and synthesized using the FreePDK45 technology library. The simulation result shows no loss of accuracy, but the synthesis results do not show the same positive results for the area and power. The default version of the accelerator uses single-precision floating-point format, float32, while a modified accelerator using the bfloat16 number representation shows significant improvements in area and power with almost no additional loss of accuracy.

This work deals with design of communication protocol for data transmission between control computer and computing cores implemented on FPGA chips. The purpose of the communication is speeding the performance demanding software algorithms of non-stream data processing by their hardware computation on accelerating system. The work defines a terminology used for protocol design and analyses current solutions of given issue. After that the work designs structure of the accelerating system and communication protocol. In the main part the work describes the implementation of the protocol in VHDL language and the simulation of implemented modules. At the end of the work the aplication of designed solution is presented along with possible extension of this work.

The main object of this master´s thesis is to increase productivity glueing and lowering the weight of the bus roof while maintaining its toughness. The first part describes the present glueing technologies. Subsequently, there is adduced theoretically various hypothesis about the glue seam. The proposed solution is to use the glue containing accelerators. Two glues from the companies DINITROL and SIKA are suggested and compared. The following chapter details the suggested technology and describes its possibilities. The final part presents the project of ultralight roof made from sandwich panels.

ThomX, un démonstrateur de source Compton compacte de rayons X d’énergie réglable entre 45 et 90 keV, est en construction sur le campus de l'Université Paris-Saclay à Orsay. La thèse s’inscrit dans le cadre de l’upgrade du linac de ThomX qui consiste à réaliser une section accélératrice compacte à fort gradient en bande S (3 GHz) pour porter les faisceaux de ThomX de 50 MeV à 70 MeV. Un accord de collaboration R&D est signé entre LAL et PMB-Alcen pour développer une structure accélératrice en cuivre (OFHC) compacte en bande S à fort gradient. Une étude électromagnétique, thermique et dynamique de faisceau a été effectuée au LAL pour proposer une géométrie optimale de la section accélératrice pour atteindre des gradients accélérateurs très élevés. PMB est en charge d’améliorer les processus de fabrication en commençant par la réalisation des prototypes pour valider les choix technologiques et ensuite fabriquer la section finale pour répondre aux spécifications demandées. Dans un premier temps une étude couplée électromagnétique-thermique-structurelle du canon HF a été effectuée en utilisant le Logiciel d'analyse des éléments finis 3D (ANSYS). Ensuite l’étude électromagnétique et l’optimisation de la géométrie des cellules accélératrices ainsi que la conception des coupleurs de puissance pour constituer les prototypes à impédance constante avec un certain nombre de cellules réduit et la section accélératrice compacte à gradient constant ont été effectuées en utilisant les logiciels CST MWS et HFSS. Puis une étude thermomécanique de la structure accélératrice a été réalisée avec ANSYS pour concevoir et optimiser le circuit de refroidissement pour extraire la chaleur générée par la puissance HF dissipée dans les parois de la structure et garantir une répartition uniforme de la température au long de la structure. Les simulations du vide ont été également réalisées avec le code Monte Carlo pour envisager la meilleure solution de pompage pour le prototype de cuivre et la section finale. En outre, les principales étapes suivies dans la fabrication du Canon HF au LAL et le prototype en aluminium à 7 cellules chez PMB-Alcen ont été présentées. Des tests HF bas niveau du prototype ont été effectués afin de valider la géométrie « processus d’usinage ». Compte tenu des résultats expérimentaux, des problèmes techniques et des contraintes technologiques ont été abordées et certaines solutions ont été proposées pour la fabrication des prototypes en cuivre et de la section finale. Les simulations de la dynamique des faisceaux du Linac de ThomX ont été effectuées en utilisant le code ASTRA. Le but est de réduire autant que possible la dispersion en énergie et l’emittance transverse du paquet d’électrons au point d’interaction avec les impulsions laser, pour préserver la pureté spectrale de rayons X produits. Un modèle aussi proche que possible des caractéristiques des composants réels, tels que le canon HF, la section accélératrice à onde progressive (OP) et les solénoïdes a été pris en compte dans les simulations. Des résultats importants ressortent de ces simulations concernant les paramètres du laser (taille et durée du spot), le champ magnétique maximal des solénoïdes pour la compensation de l'effet de charge espace, le déphasage entre l’onde RF et le laser et l'effet du champ électromagnétique sur la dynamique des électrons. Différentes options pour les paramètres de fonctionnement de la machine et une nouvelle configuration de la position des solénoïdes ont été proposées. L’optimisation des caractéristiques du paquet d’électrons a été obtenue en utilisant un algorithme génétique et les performances finales du faisceau d’électrons ont été mises en évidence. / The ThomX source should provide quasi-monochromatic high-quality X-rays (range 45-90 keV). The framework of the thesis is the electron beam linac energy upgrade from 50 MeV to 70 MeV necessary to achieve X-rays of 90 keV. For this purpose, the development of a compact high-gradient S-band electron accelerating structure is needed. It implies a research and development (R&D) activity at LAL in partnership with a French company (PMB-Alcen) in the High-Gradient (HG) technology of accelerating structures. The LAL-PMB-Alcen collaboration aims at the fabrication of a normal-conducting HG S-band structure by tackling the technological aspects that limit the achievement of high-gradient acceleration mostly due to vacuum RF breakdown and pulsed heating fatigue. Basically, the electromagnetic and thermal design of the HG S-band accelerating section has been performed at LAL. Meanwhile, PMB-Alcen was in charge to perform the fabrication, tuning and low power tests of prototypes and the final accelerating section. In this work, a fully coupled electromagnetic-thermal-structural finite element analysis on the THOMX RF gun has been performed with Ansys workbench. The HG accelerating section final regular cell dimensions and the power coupler design have been optimized. In particular, the electromagnetic simulation techniques and outcomes have been applied to constant impedance (CI) TW prototypes and also to a constant gradient (CG) final accelerating section. This allowed to verify the geometry choice, validate the fabrication procedure and check the fulfilment of the normal operating conditions. Moreover, a coupled thermo-mechanical study on a CI copper prototype has been performed. The water cooling system has been simulated to validate the capability to extract the heat generated by the dissipated power on the walls of the structure and guarantee a uniform temperature distribution along the section. Also, vacuum simulations have been performed on a 16-cells CI copper prototype and also on the final CG accelerating section. In addition, the main steps for the fabrication of the RF gun at LAL and a 7-cells aluminium prototype at PMB-Alcen have been presented. RF low power tests on the prototype have been performed in order to validate the 3D geometry design and the machining process. Taking into account the experimental results, mechanical problems and technological constraints have been tackled and some solutions have been proposed for the future copper prototype fabrication. Finally, beam dynamics simulations of the ThomX linac has been carried out by ASTRA code. The aim is to reduce as much as possible the energy spread and the transverse emittance to preserve the spectral purity of the produced X-rays, at the electron-photon interaction point. A model as close as possible to the characteristics of the real components, such as RF gun, TW section and solenoids has been considered. Important results came out from these simulations regarding laser parameters (spot size and duration), the maximum magnetic field of solenoids for high space charge effect compensation, dephasing between the RF and laser in the gun and effect of the travelling wave electromagnetic field on the particle dynamics. Different options for the parameter settings of machine operation and a new configuration of the solenoids position have been proposed. An optimization of the beam dynamics properties has been obtained by using a genetic algorithm and the ultimate performances of the electron beam have been highlighted.

abstract: Deep neural networks (DNNs) have had tremendous success in a variety of statistical learning applications due to their vast expressive power. Most applications run DNNs on the cloud on parallelized architectures. There is a need for for efficient DNN inference on edge with low precision hardware and analog accelerators. To make trained models more robust for this setting, quantization and analog compute noise are modeled as weight space perturbations to DNNs and an information theoretic regularization scheme is used to penalize the KL-divergence between perturbed and unperturbed models. This regularizer has similarities to both natural gradient descent and knowledge distillation, but has the advantage of explicitly promoting the network to and a broader minimum that is robust to weight space perturbations. In addition to the proposed regularization, KL-divergence is directly minimized using knowledge distillation. Initial validation on FashionMNIST and CIFAR10 shows that the information theoretic regularizer and knowledge distillation outperform existing quantization schemes based on the straight through estimator or L2 constrained quantization. / Dissertation/Thesis / Masters Thesis Computer Engineering 2019

Die vorliegende Arbeit beschreibt einen Architekturansatz, der im Rahmen einer Machbarkeitsstudie bei IBM entwickelt wurde. Dadurch wird der IBM DB2 Analytics Accelerator als eine Data-Warehouse-Appliance dazu in die Lage versetzt, über angepasste Schnittstellen Data-Mining-Modelle über entsprechende Algorithmen direkt auf dem Accelerator zu erstellen. Neben dieser Beschreibung wird die bisherige Verwendung des DB2 Analytics Accelerators sowie das zugehörige Umfeld von Datenbanksystemen bis zum System z Mainframe vorgestellt. Darauf aufbauend werden praxisnahe Anwendungsfälle präsentiert, die unter Anwendung von intelligenten Methoden auf gespeicherten Kundendaten statistische Modelle erstellen. Für diesen Prozess wird die Datengrundlage zuerst vorbereitet und angepasst, um sie dann in dem zentralen Data-Mining-Schritt nach neuen Zusammenhängen zu durchsuchen.

Innerhalb der handschriftlichen Überlieferung, die aus dem Mittelalter auf uns gekommen ist, gehören Palimpseste zu den Zeugnissen, die schon immer in besonderer Weise das Interesse auf sich gezogen und Phantasien beflügelt haben.