[pt] O sistema SPEC (Speech Predictive Encoding Communication) é um método de redução de faixa que emprega técnicas de interpolação (digital) de voz. As diferenças entre os valores estimados e atuais das amostras do sinal de voz são comparadas, em um Processador, com um limiar, que pode ser fixo ou proveniente de uma tabela. O valor estimado utilizado na comparação é o do instante anterior, o que pode ser considerado como uma predição de ordem zero. O propósito principal deste trabalho é a investigação do efeito do uso de preditores de ordem superior para estimar amostras no sistema SPEC. Paralelamente, é investigado o desempenho do sistema com um Processador, no qual as diferenças entre os valores estimados e atuais das amostras são ordenadas de acordo com suas magnitudes. Assim, são transmitidas as amostras que correspondem às maiores diferenças não-comprimidas, diminuindo a degradação. / [en] The SPEC system (Speech Predictive Encoding Communication), is a bandwidth reduction method, which employs digital speech interpolation techniques. The differences between the estimated and present values of the samples of speech signals are compared, in a Processor, with a threshold which can be fixed or derived from a table. The estimated value used in the comparison is the one of the previous instant, which can be considered as a zero-order prediction. The main purpose of this work is the investigation of the effect of the use of higher order predictors to estimate samples in the SPEC system. At the same time, the system performance with a Processor where the differences between the estimated and the present values of the samples are ordered according to their magnitude is investigated. With this Processor, the samples corresponding to the greater non-compressed differences are transmitted, which contributes to minimize the system overload noise.

Os avanços na engenharia de materiais têm aumentado a busca por materiais com atividade antimicrobiana para aplicação em áreas da saúde. Neste contexto, a presente pesquisa desenvolveu materiais para revestir bráquetes metálicos utilizados no tratamento ortodôntico. As sínteses de hidroxiuretanos (HU) foram realizadas empregando-se bis(ciclocarbonatos) derivados de epóxidos com segmentos siloxânicos por reação de aminólise. Dois ciclocarbonatos (CC) foram sintetizados, a partir da reação de cicloadição de CO2 ao anel epóxi do 1,3-bis(3-glicidiloxipropil)tetrametildisiloxano e do polidimetilsiloxano diglicidil éter catalisada pelo brometo de hexadeciltrimetilamônio. Os CCs obtidos foram denominados de CCMDMS e CCPDMS. Para aumentar o tamanho dos HUs e, consequentemente proporcionar um ganho nas propriedades mecânicas, realizou-se um processo de extensão pela reação da 3-(aminometil)-3,5,5-trimetilcicloexilamina com os CCs (CC-HUMDMS e CC-HUPDMS). O processo de finalização dos HUs foi realizado através da reação do CC-HUMDS com a 3-trietoxisililpropilamina e, o CC-HUPDMS com a 3-trietoxisililpropilamina e N-aminoetil-3-trimetoxisililpropilamina. Os HUs sintetizados foram chamados de HUMDMS, HUPDMS e HUPDMS2. Para elevar o caráter antibiofilme do HUMDMS, HUPDMS e HUPDMS2 foram incorporadas nanopartículas a suas estruturas por dois métodos. O primeiro, método coloidal, consistiu na redução dos cátions Ag+ em etanol misturados com os HUs, sendo os produtos denominados de HUMDMSCol, HUPDMSCol e HUPDMS2Col. O segundo, método fotoassistido, baseou-se na adição de nanopartículas de óxido de zinco (nanoZnO), nesta etapa, os HUs foram chamados de HUMDMSZnO, HUPDMSZnO e HUPDMS2ZnO, seguido pela imersão em solução aquosa de nitrato de prata e, posterior, redução com luz ultravioleta, os materiais obtidos foram chamados de HUMDMSIrr, HUPDMSIrr e HUPDMS2Irr. Os materiais foram caracterizados quanto à sua estrutura química, para tanto, utilizou-se técnicas espectroscópicas de infravermelho com reflexão total atenuada, ressonânciamagnética nuclear e fotoelétrons excitados por raios-X. A morfologia foi analisada por microscopia de transmissão eletrônica e difração de raios-X, avaliou-se a estabilidade térmica por análise termogravimétrica e a energia superficial por medidas de ângulo de contato. As técnicas de espectroscopia no ultravioleta e visível e espectrometria de emissão atômica com plasma acoplado indutivamente foram utilizadas para observar a banda plasmônica das nanopartículas e a sorção/dessorção de cátions, respectivamente. As nanoAg apresentaram absorção plasmônica em 415 e 440 nm para os métodos coloidal e fotoassistido, respectivamente. O valor de pH satisfatório para sorção de Ag+ foi observado entre 7 e 8, com a seguinte ordem de sorção HUPDMS2 > HUMDMS > HUPDMS. As isotermas de sorção mostraram que o modelo de Sips justifica satisfatoriamente o processo de sorção dos cátions Ag+ pelos HUs. Os materiais foram termicamente estáveis para temperaturas inferiores a 200 °C. A resistência abrasiva dos HUs foi avaliada com base em ensaios de fricção utilizando-se escova e creme dental. Os filmes de HUPDMS apresentaram maior fragilidade, mesmo com aumento da resistência devido adição de nanoZnO. Todos os filmes apresentaram energia superficial inferior a 25 mN/m, indicando assim, resistência à formação de biofilme. A atividade antibiofilme dos materiais híbridos foi testada para as bactérias Escherichia coli, Enterococos feacalis e Estreptococos mutans, bem como, para o fungo Candida albicans. O processo de dessorção dos cátions Ag+ nos filmes com nanoAg potencializou a eficácia contra a formação de biofilme. / Advances in material engineering have increased the search for materials with antimicrobial activity for application in health areas. In this context, the present research developed materials to coat metallic brackets used in orthodontic treatment. The synthesis of hydroxyurethanes (HU) was carried out using bis(cyclocarbonates) derived from epoxides with siloxane segments by aminolysis reaction. Two cyclocarbonates (CC) were synthesized by cycloaddition reaction of CO2 to the epoxy ring of 1,3-bis(3-glycidyloxypropyl) tetramethyldisiloxane and the polydimethylsiloxane diglycidyl ether catalyzed by hexadecyltrimethylammonium bromide. The obtained CCs were called CCMDMS and CCPDMS. To increase the size of the HUs and consequently provide a gain in the mechanical properties, a process of extension was carried out by the reaction of 3-(aminomethyl)-3,5,5-trimethylcyclohexylamine with the CCs (CC-HUMDMS and CC-HUPDMS). The HU-termination process was performed by reacting CC-HUMDS with 3-triethoxysilylpropylamine and CC-HUPDMS with 3-triethoxysilylpropylamine and N-aminoethyl-3-trimethoxysilylpropylamine. The HUs synthesized were called HUMDMS, HUPDMS and HUPDMS2. To raise the antibiofilm character of the HUMDMS, HUPDMS and HUPDMS2 nanoparticles were incorporated into their structures by two methods. The first, colloidal method consisted in the reduction of the Ag+ cations in ethanol mixed with the HUs, the products being called HUMDMSCol, HUPDMSCol and HUPDMS2Col. The second, photo-assisted method was based on the addition of nanoparticles of zinc oxide (nanoZnO), at this stage the HUs were called HUMDMSZnO, HUPDMSZnO and HUPDMS2ZnO, followed by immersion in aqueous solution of silver nitrate and subsequent reduction by ultraviolet light, the obtained materials were called HUMDMSIrr, HUPDMSIrr and HUPDMS2Irr. The materials were characterized in terms of their chemical structure, using infrared spectroscopy techniques with attenuated total reflection, nuclear magnetic resonance and X-ray excited photoelectrons, the morphology was analyzed by transmission electron microscopy and X-ray diffraction, the thermal stability was evaluated by thermogravimetric analysis and the surface energy by contact angle measurements. Ultraviolet and visible spectroscopy techniques and inductively coupled plasma atomic emission spectrometry were used to observe the plasmon band of nanoparticles and cation sorption/desorption, respectively. The nanoAg presented plasmid absorption at 415 and 440 nm for the colloidal and photoassayed methods, respectively. The pH value satisfactory for Ag+ sorption was observed between 7 and 8, with the following sorption order HUPDMS2 > HUMDMS > HUPDMS. The sorption isotherms showed that the Sips model satisfactorily justifies the sorption process of the Ag+ cations by HUs. The materials were thermally stable at temperatures below 200 °C. The abrasive strength of the HUs were evaluated based on friction tests using brush and toothpaste. The films of HUPDMS presented greater fragility, even with increased resistance due to the addition of nanoZnO. All films presented surface energy lower than 25 mN/m, thus indicating resistance to biofilm formation. The antibiofilm activity of the hybrid materials was tested for Escherichia coli, Enterococos feacalis and Streptococcus mutans strains, as well as for Candida albicans fungus. The desorption process of Ag+ cations in films with nanoAg enhanced the antibiofilm efficacy.

This dissertation consists of three projects in matched case-control studies. In the first project, we employ a general bias preventive approach developed by Firth (1993) to handle the bias of an estimator of the log-odds ratio parameter in conditional logistic regression by solving a modified score equation. The resultant estimator not only reduces bias but also can prevent producing infinite value. Furthermore, we propose a method to calculate the standard error of the resultant estimator. A closed form expression for the estimator of the log-odds ratio parameter is derived in the case of a dichotomous exposure variable. Finite sample properties of the estimator are investigated via a simulation study. Finally, we apply the method to analyze a matched case-control data from a low-birth-weight study. In the second project of this dissertation, we propose a score typed test for checking adequacy of a functional form of a covariate of interest in matched case-control studies by using penalized regression splines to approximate an unknown function. The asymptotic distribution of the test statistics under the null model is a linear combination of several chi-square random variables. We also derive the asymptotic distribution of the test statistic when the alternative model holds. Through a simulation study we assess and compare the finite sample properties of the proposed test with that of Arbogast and Lin (2004). To illustrate the usefulness of the method, we apply the proposed test to a matched case-control data constructed from the breast cancer data of the SEER study. Usually a logistic model is needed to associate the risk of the disease with the covariates of interests. However, this logistic model may not be appropriate in some instances. In the last project , we adopt idea to matched case-control studies and derive an information matrix based test for testing overall model adequacy and investigate the properties against the cumulative residual based test in Arbogast and Lin (2004) via a simulation study. The proposed method is less time consuming and has comparative power for small parameters. It is suitable to explore the overall model fitting.

Differential systems appear modelling many natural phenomena in different branches of sciences, in biological and physical applications among other areas. Differential systems usually have invariant curves and we can obtain a better description of the qualitative behaviour of their solutions using them. Such invariant curves may be algebraic or not and in the case where they are closed, isolated in the set of periodic orbits and without singular points, they are called limit cycles. There is a very famous problem, proposed by David Hilbert in 1900 what ask about the maximum number of limit cycle that all polynomial differential systems of a given degree could present. In this work we investigate the existence of some invariant curves in quadratic polynomial differential systems and in discontinuous piecewise differential systems (they are known as Filippovs systems). Even after hundreds of studies on the phase portraits of real planar quadratic vector fields the complete characterization of their phase portraits is a quite complex task, they depend on twelve parameters, after affine transformations and time rescaling, we have families with five parameters, which is still a large number. So many subclasses have been considered instead of the complete system. In this work we investigate conditions under the parameters of the system for a planar quadratic differential system present invariant algebraic curve of degree 3 (a cubic curve) and a Darboux invariant and obtain all the topological non-equivalent phase portraits of these systems. The increasing interest in the theory of nonsmooth vector fields has been mainly motivated by their strong relation with physics, engineering, biology, economy, and other branches of sciences. In the study of the Filippovs systems, we investigate the number of periodic orbits that they can present. In this study we apply the averaging theory. Such theory is used to study some classical models and we also present generalization of such technique for a class of nonsmooth systems. In addition, we also show how the LyapunovSchmidt reduction method can be used to consider cases where the averaging theory is not sufficient to study periodic solutions. / Sistemas diferenciais aparecem na modelagem de muitos fenômenos naturais em diferentes ramos da ciência, em aplicações biológicas e físicas, entre outras áreas. Sistemas diferenciais geralmente possuem curvas invariantes e podemos obter uma melhor descrição do comportamento qualitativo de suas soluções utilizando-as. Tais curvas invariantes podem ser algébricas ou não e, no caso de serem fechadas, isoladas no conjunto de órbitas periódicas e sem pontos singulares, são chamadas de ciclos limites. Há um problema muito famoso, proposto por David Hilbert em 1900, que questiona o número máximo de ciclos limites que os sistemas diferenciais polinomiais de um determinado grau poderiam apresentar. Neste trabalho investigamos a existência de algumas curvas invariantes em sistemas diferenciais polinomiais quadráticos e em sistemas diferenciais contínuos por partes (eles são conhecidos como sistemas de Filippov). Mesmo após centenas de estudos sobre os retratos de fase dos campos vetoriais reais planares e quadráticos, a caracterização completa de seus retratos de fase é uma tarefa bastante complexa. Eles dependem de doze parâmetros e após transformações afins e reescalonamento de tempo, temos famílias com cinco parâmetros, o que ainda é um grande número. Assim muitas subclasses tem sido consideradas em vez do sistema completo. Neste trabalho investigamos condições sob os parâmetros do sistema para que um sistema diferencial planar quadrático apresente uma curva algébrica invariante de grau 3 (curva cúbica) e um invariante de Darboux e obtemos todos os retratos de fase não equivalentes destes sistemas. O crescente interesse na teoria dos campos de vetores suaves por partes tem sido motivado, principalmente, por sua forte relação com a física, engenharia, biologia, economia e outros ramos das ciências. No estudo dos sistemas de Filippov, investigamos o número de órbitas periódicas que eles podem apresentar. Para este estudo, aplicamos a teoria do averaging. Tal teoria é usada para estudar alguns modelos clássicos e também apresentamos a generalização de tal técnica para uma classe de sistemas suaves por partes. Além disso, mostramos também como o método de redução de Lyapunov Schmidt pode ser usado para considerar casos em que a teoria do averaging sozinha não é suficiente para estudar soluções periódicas.

Le recours à des simulations numériques pour l’étude de l’influence des paramètres d’entrée (matériaux, chargements, conditions aux limites, géométrie, etc.) sur les différentes quantités d’intérêt en soudage (contraintes résiduelles, distorsion, etc.) s’avère trop long et coûteux vu l’aspect multi-paramétrique de ces simulations. Pour explorer des espaces paramétriques de grandes dimensions, avec des calculs moins coûteux, il parait opportun d’utiliser des approches de réduction de modèle. Dans ce travail, d’une façon a posteriori, une stratégie non-intrusive est développée pour construire les abaques dédiées aux études paramétriques du soudage. Dans une phase offline, une base de données (‘snapshots’) a été pré-calculée avec un choix optimal des paramètres d'entrée donnés par une approche multi-grille (dans l’espace des paramètres). Pour explorer d’autres valeurs de paramètres, une méthode d’interpolation basée sur la variété Grassmannienne est alors proposée pour adapter les bases réduites espace-temps issues de la méthode SVD. Cette méthode a été constatée plus performante que les méthodes d’interpolation standards, notamment en non-linéaire. Afin d’explorer des espaces paramétriques de grandes dimensions, une méthode de type décomposition tensorielle (i.e. HOPGD) a été également étudiée. Pour l’aspect d’optimalité de l’abaque, nous proposons une technique d’accélération de convergence pour la HOPGD et une approche ‘sparse grids’ qui permet d’échantillonner efficacement l’espace des paramètres. Finalement, les abaques optimaux de dimension jusqu’à 10 à précision contrôlée ont été construits pour différents types de paramètres (matériaux, chargements, géométrie) du procédé de soudage. / The use of standard numerical simulations for studies of the influence of input parameters (materials, loading, boundary conditions, geometry, etc.) on the quantities of interest in welding (residual stresses, distortion, etc.) proves to be too long and costly due to the multiparametric aspect of welding. In order to explore high-dimensional parametric spaces, with cheaper calculations, it seems to be appropriate to use model reduction approaches. In this work, in an a posteriori way, a non-intrusive strategy is developed to construct computational vademecum dedicated to parametric studies of welding. In an offline phase, a snapshots database is pre-computed with an optimal choice of input parameters given by a “multi-grids” approach (in parameter space). To explore other parameter values, an interpolation method based on Grassmann manifolds is proposed to adapt both the space and time reduced bases derived from the SVD. This method seems more efficient than standard interpolation methods, especially in non-linear cases. In order to explore highdimensional parametric spaces, a tensor decomposition method (i.e. HOPGD) has also been studied. For the optimality aspect of the computational vademecum, we propose a convergence acceleration technique for HOPGD and a “sparse grids” approach which allows efficient sampling of the parameter space. Finally, computational vademecums of dimension up to 10 with controlled accuracy have been constructed for different types of welding parameters (materials, loading, geometry).

Submitted by Liliane Ferreira (ljuvencia30@gmail.com) on 2018-04-11T12:39:47Z No. of bitstreams: 2 Dissertação - Carlos Alirio Rico Acevedo - 2018.pdf: 1289579 bytes, checksum: 0b60c803c3d9f6f61772e58e7d624086 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2018-04-12T11:29:32Z (GMT) No. of bitstreams: 2 Dissertação - Carlos Alirio Rico Acevedo - 2018.pdf: 1289579 bytes, checksum: 0b60c803c3d9f6f61772e58e7d624086 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2018-04-12T11:29:32Z (GMT). No. of bitstreams: 2 Dissertação - Carlos Alirio Rico Acevedo - 2018.pdf: 1289579 bytes, checksum: 0b60c803c3d9f6f61772e58e7d624086 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2018-03-16 / Conselho Nacional de Pesquisa e Desenvolvimento Científico e Tecnológico - CNPq / Let $ k \geq 2.$ an integer. The recurrence $ \fk{n} = \sum_ {i = 0}^k \fk{n-i} $ for $ n> k $, with initial conditions $F_{-(k-2)}^{(k)}=F_{-(k-3)}^{(k)}=\cdots=F_{0}^{(k)}=0$ and $F_1^{ (k)} = 1$, which is called the $k$-generalized Fibonacci sequence. When $ k = 2 ,$ we have the Fibonacci sequence $ \{ F_n \}_{n\geq 0}.$ We will show that the equation $F_{n}^{x}+F_{n+1}^x=F_{m}$ does not have no non-trivial integer solutions $ (n, m, x) $ to $ x> 2 $. On the other hand, for $ k \geq 3,$ we will show that the diophantine equation $\epi$ does not have integer solutions $ (n, m, k, x) $ with $ x \geq 2 $. In both cases, we will use initially Matveev's Theorem, for linear forms in logarithms and the reduction method due to Dujella and Pethö, to limit the variables $ n, \; m $ and $ x $ at intervals where the problem is computable. In addition, in the case for $ k\geq 3 $, we will use the fact that the dominant root the $k$-generalized Fibonacci sequence is exponentially close to 2 to bound $k$, a method developed by Bravo and Luca. / Seja $k\geq 2$ inteiro, considere-se a recorrência $\fk{n}=\sum_{i=0}^{k}\fk{n-i}$ para $n>k$, com condições iniciais $F_{-(k-2)}^{(k)}=F_{-(k-3)}^{(k)}=\cdots=F_{0}^{(k)}=0$ e $F_{1}^{(k)}=1$, que é a sequência de Fibonacci $k$-generalizada. No caso quando $k=2$, é dizer, para a sequência de Fibonacci $\{F_n\}_{n\geq 0}$, vai-se mostrar que a equação $F_{n}^{x}+F_{n+1}^x=F_{m}$ não possui soluções inteiras não triviais $(n,m,x)$ para $x>2$. Por outro lado para, $k\geq 3$ se mostrar que a equação diofantina $\epi$ não possui soluções inteiras $(n,m,k,x)$ com $x\geq 2$. Em ambos casos, inicialmente são usados resultados como o Teorema de Matveev, para formas lineares em logaritmos e o método de redução de Dujella e Pethö, para limitar as variáveis $n, \; m$ e $x$ em intervalos onde o problema seja computável. Adicionalmente, no caso para $k\geq 3$ é usado que a raiz dominante da sequência de Fibonacci $k$-generalizada e exponencialmente próxima a 2, para limitar $k$, o que é um método desenvolvido por Bravo e Luca.

Dans le domaine automobile, les modules électroniques de puissance des produits mécatroniques voient leur puissance sans cesse s'accroître, tout en étant confinés dans des volumes de plus en plus réduits. Au cours de leur fonctionnement, les composants semi-conducteurs et leur assemblage subissent ainsi des contraintes électro-thermo-mécaniques sévères, susceptibles d'entraîner leur destruction et de provoquer la défaillance du produit. L'étude de la fiabilité et le calcul de la durée de vie de tels produits dépendent des températures de jonction calculées au niveau des puces des composants de puissances. De surcroît, le contexte d'applications embarquées requiert de maîtriser, outre les paramètres électriques et mécaniques, les paramètres thermiques tels que les températures de jonctions et les puissances dissipées au niveau des composants, qu'il est nécessaire de réguler et contrôler en temps réel afin d'assurer le bon fonctionnement du produit. L'objectif de cette thèse est ainsi de proposer une méthode d'identification de modèles réduits dans le but d'estimer le comportement thermique des modules électroniques de puissance, en se fondant uniquement sur les données d'entrées et les résultats issus d'une simulation numérique d'un modèle détaillé du système étudié. Dans cette thèse, une nouvelle méthode d'identification, nommée « Kernel Identification Method », est développée. Cette méthode a été validée sur une application industrielle traitant d'un problème thermique couplé solide/fluide dont le comportement est essentiellement régi par de la convection forcée. Une étude exploratoire portant sur l'identification de problèmes non linéaires où la convection naturelle joue le rôle dominant est ensuite proposée. A cet effet, deux méthodes d’identification non-paramétrique sont proposées : (i) une première méthode basée sur l’extension de la méthode Kernel Identification Method ; et (ii) une deuxième méthode basée sur la variante dite « Unscented » du filtre de Kalman. / In automotive applications, the thermal dissipation of power electronics modules in mechatronic products is constantly increasing, whereas these products are confined in increasingly reduced volumes. During their operation, the semiconductor components and their environment are then submitted to severe electro-thermo-mechanical stresses that could cause their damage and lead to the product failure. The reliability and lifetime prevision of such products depend on the temperature junction located at the chip of power components. Furthermore, in order to ensure the safe operation of embedded applications, it is essential to perform a real-time control of thermal parameters such as the junction temperatures and power dissipated on the power components, in addition to the electrical and mechanical parameters. The objective of this thesis is to develop an identification method aimed at producing reduced thermal models to estimate the thermal behaviour of power electronic modules. Designed in a non-intrusive framework, this method post-processes the input data and the results produced by the numerical simulation of a detailed of the system under study. In this thesis, a new identification method, called "Kernel Identification Method" is developed. It has been validated on an industrial application dealing with a thermally coupled solid / fluid problem mainly governed by forced convection. An exploratory study of nonlinear problems identification where the natural convection plays the dominant role is then proposed. To this end, two identification methods of nonparametric nature are proposed: (i) a method based on the extension of the Kernel Identification Method; and (ii) a second method based on the "unscented" variant of the Kalman filter.

Made available in DSpace on 2016-06-02T20:06:09Z (GMT). No. of bitstreams: 1 5618.pdf: 988258 bytes, checksum: 1ce6234a919d1f5b4a4d4fd7482d543c (MD5) Previous issue date: 2013-11-26 / Financiadora de Estudos e Projetos / The analysis of bivariate data can be found in several areas of knowledge, when the data of interest are obtained in a paired way and present correlation between counts. In this work the Holgate bivariate Poisson, bivariate generalized Poisson and bivariate zero-inflated Poisson models are presented, which are useful to the modeling of bivariate count data correlated. Illustrative applications are presented for these models and the comparison between them is made by using criteria of model selection AIC and BIC, as well as the asymptotic likelihood ratio test. Particularly, we propose a Bayesian approach to the Holgate bivariate Poisson and bivariate zero-inflated Poisson models, based in the Gibbs sampling algorithm with data augmentation. / A análise de dados bivariados pode ser encontrada nas mais diversas áreas do conhecimento, quando os dados de interesse são obtidos de forma pareada e apresentam correlação entre as contagens. Neste trabalho são apresentados os modelos Poisson bivariado de Holgate, Poisson generalizado bivariado e Poisson bivariado inflacionado de zeros, os quais são úteis na modelagem de dados de contagem bivariados correlacionados. Aplicações ilustrativas serão apresentadas para estes modelos e a comparação entre eles será realizada pelos critérios de seleção de modelos AIC e BIC, assim como pelo teste da razão de verossimilhança assintótico. Particularmente, propomos uma abordagem Bayesiana para os modelos Poisson bivariado de Holgate e Poisson Inflacionado de zeros, baseada no algoritmo Gibbs sampling com dados ampliados.

Le but de ce travail est d'introduire un cadre d'approximation, la Reference Points Method, afin de réduire la complexité de calcul des opérations algébriques lorsqu'elles concernent des approximations à variables séparées dans le cadre de la Proper Generalized Decomposition.La PGD a été introduite dans [1] dans le cadre de la méthode LaTIn pour résoudre efficacement des équations différentielles non linéaires et dépendants du temps en mécanique des structures. La technique consiste à chercher la solution d'un problème dans une base d'ordre réduit (ROB) qui est automatiquement et à la volée générée par la méthode LaTIn. La méthode LaTIn est une stratégie itérative qui génère les approximations de la solution sur l'ensemble du domaine espace-temps-paramètres par enrichissements successifs. Lors d'une itération particulière, la ROB, qui a déjà été formée, est d'abord utilisée pour calculer un nouveau modèle réduit (ROM) et, donc, pour trouver une nouvelle approximation de la solution. Si la qualité de cette approximation ne suffit pas, la ROB est enrichie avec la génération d'un nouveau produit de fonctions PGD en utilisant un algorithme de type 'greedy'.Les techniques de réduction de modèle sont particulièrement efficaces lorsque le ROM a besoin d'être construit qu'une seule fois. Ce n'est pas le cas pour les techniques de réduction de modèle quand elles concernent des problèmes non linéaires. En effet, dans un tel cas, les opérateurs qui sont impliqués dans la construction du ROM varient au cours du processus itératif et des calculs préliminaires ne peuvent pas être effectués à l'avance pour accélérer le processus 'online'.Par conséquent, la construction du ROM est un élément coûteux de la stratégie de calcul en terme de temps de calcul. Il en découle la nécessité d'évaluer, à chaque itération, la fonction non linéaire de grande dimension (et éventuellement sa jacobienne) et ensuite sa projection pour obtenir les opérateurs réduits. Cela représente un point de blocage des stratégies de réduction de modèle dans le cadre non linéaire. Le présent travail a comme but une réduction ultérieure du coût de calcul, grâce à l'introduction d'un nouveau cadre de rapprochement dédiée à la stratégie de calcul LaTIn-PGD. Il est basé sur la notion de temps, de points et de paramètres de référence et permet de définir une version compressée des données. Comparé à d'autres techniques similaires [3,4] cela ne se veut pas une technique d'interpolation, mais un cadre algébrique qui permet de donner une première approximation, peu coûteuse, de toutes les quantités sous une forme à variable séparés par des formules explicites. L'espace de données compressées présente des propriétés intéressantes qui traitent les opérations algébriques élémentaires. Le RPM est introduit dans le solveur LaTIn-PGD non linéaire pour calculer certaines opérations répétitives. Ces opérations sont liées à la résolution du problème du temps / paramètre qui implique la mise à jour de l'opérateur tangent et la projection de ce dernier sur la base réduite. La RPM permet de simplifier et de réduire le nombre d'opérations nécessaires.[1] Ladevèze P., Sur une famille d’algorithmes en mécanique des structures, Comptes Rendus Académie des Sciences. Paris. Ser. II 300, pp.41-44, 1985.[2] Chinesta, F., Ladevèze, P., and Cueto, E. A short review on model order reduction based on proper generalized decomposition. Archives of Computational Methods in Engineering, 18, pp.395-404, 2011.[3] Barrault M., Maday Y., Nguyen N., Patera A., An ’empirical interpolation’ method: application to efficient reduced-basis discretization of partial differential equations, Comptes Rendus Académie des Sciences. Paris. Ser. I, 339, pp. 667-672, 2004.[4] Chaturentabut S., Sorensen D., Nonlinear model reduction via discrete empirical interpolation, Society for Industrial and Applied Mathematics 32(5), pp.2737-2764, 2010. / The aim of this work is to introduce an approximation framework, called Reference Points Method (RPM), in order to decrease the computational complexity of algebraic operations when dealing with separated variable approximations in the Proper Generalized Decomposition (PGD) framework.The PGD has been introduced in [1] in the context of the LATIN method to solve efficiently time dependent and/or parametrized nonlinear partial differential equations in structural mechanics (see, e.g., the review [2] for recent applications). Roughly, the PGD technique consists in seeking the solution of a problem in a relevant Reduced-Order Basis (ROB) which is generated automatically and on-the-fly by the LATIN method. This latter is an iterative strategy which generates the approximations of the solution over the entire time- space-parameter domain by successive enrichments. At a particular iteration, the ROB, which has been already formed, is at first used to compute a projected Reduced-Order Model (ROM) and find a new approximation of the solution. If the quality of this approximation is not sufficient, the ROB is enriched by determining a new functional product using a greedy algorithm.However, model reduction techniques are particularly efficient when the ROM needs one construction only. This is not the case for the model reduction techniques when they are addressed to nonlinear problems. Indeed, in such a case, the operators which are involved in the construction of the ROM change all along the iterative process and no preliminary computations can be performed in advance to speed up the online process. Hence, the construction of the ROM is an expensive part of the calculation strategy in terms of CPU. It ensues from the need to evaluate the high-dimensional nonlinear function (and eventually its Jacobian) and then to project it to get the low-dimensional operators at each computational step of a solution algorithm. This amounts to being the bottleneck of nonlinear model reduction strategies.The present work is then focused on a further reduction of the computational cost, thanks to the introduction of a new approximation framework dedicated to PGD-based nonlinear solver. It is based on the concept of reference times, points and parameters and allows to define a compressed version of the data. Compared to other similar techniques [3,4] this is not an interpolation technique but an algebraic framework allowing to give an inexpensive first approximation of all quantities in a separated variable form by explicit formulas. The space of compressed data shows interesting properties dealing the elementary algebraic operations. The RPM is introduced in the PGD-based nonlinear solver to compute some repetitive operations. These operations are related to the resolution of the time/parameter problem that involves the update of the tangent operator (for nonlinear problems) and the projection of this latter on the Reduced Order Basis. For that the RPM allows to simplify and reduce the number of operations needed.[1] Ladevèze P., Sur une famille d’algorithmes en mécanique des structures, Comptes Rendus Académie des Sciences. Paris. Ser. II 300, pp.41-44, 1985.[2] Chinesta, F., Ladevèze, P., and Cueto, E. A short review on model order reduction based on proper generalized decomposition. Archives of Computational Methods in Engineering, 18, pp.395-404, 2011.[3] Barrault M., Maday Y., Nguyen N., Patera A., An ’empirical interpolation’ method: application to efficient reduced-basis discretization of partial differential equations, Comptes Rendus Académie des Sciences. Paris. Ser. I, 339, pp. 667-672, 2004.[4] Chaturentabut S., Sorensen D., Nonlinear model reduction via discrete empirical interpolation, Society for Industrial and Applied Mathematics 32(5), pp.2737-2764, 2010.

No description available.