Cette thèse de doctorat porte sur l’optimisation du bioprocédé de production d’acide lactique à partir de la farine de blé. L'acide lactique s’avère en effet de plus en plus attractif pour la production de PLA (acide poly lactique), un bio polymère, d’autant plus que différentes matières premières peu coûteuses comme la farine de blé sont désormais utilisées comme sources de carbone pour sa production. Cette thèse comprend trois parties principales. Une première partie propose pour l’optimisation du procédé de transformation du blé un schéma innovant composé de trois étapes successives : une liquéfaction, suivi d’une étape de saccharification et hydrolyse des protéines simultanées (SSPH) et une étape finale de saccharification, hydrolyse des protéines et fermentation simultanées (SSPHF). La deuxième partie s’intéresse à la modélisation de l’étape SSPHF (étape limitante) dans un bioréacteur continu. La détermination des paramètres du modèle ainsi que leur validation sont réalisées à l’aide de campagnes d’essais sur un bioréacteur de 5 L.Enfin, la dernière partie développe la mise en oeuvre de stratégies de commande permettant de maintenir le bioprocédé à son point optimal de fonctionnement. Pour ce faire, du fait de l’absence de capteurs pour la mesure en temps réel des concentrations des variables clé dans le bioréacteur, des estimateurs de ces concentrations ainsi que du taux de production en acide lactique sont tout d’abord élaborés. Des stratégies de commande régulant la concentration d’acide lactique à sa valeur optimale sont ensuite synthétisées et comparées en simulation. Une commande adaptative combinant une commande linéarisante par retour d’état et un estimateur du taux de production en acide lactique est finalement retenue et validée expérimentalement sur un réacteur instrumenté. Cette dernière s’est avérée robuste vis-à-vis des erreurs de modélisation et a permis lors des expériences de doubler la productivité de l’acide lactique. / This PhD thesis focuses on the optimization of the bioprocess of lactic acid production from wheat flour. Indeed, lactic acid has received much attention for the production of PLA (Poly Lactic Acid), a biopolymer, since different inexpensive raw material such as wheat flour are now used as carbon source for its production. This work was performed in three main steps. In the first step, an innovative wheat transformation process is proposed, whose main steps are the following: a liquefaction followed by a simultaneous saccharification, proteins hydrolysis (SSPH) and and a final simultaneous saccharification, proteins hydrolysis and fermentation (SSPHF). Secondly, the modeling of the SSPHF (limiting step) in a continuous bioreactor is considered. The determination and validation of model parameters is performed by means of experimental campaigns in a 5 L bioreactor.In the last step, the development of control strategies to maintain the process at its optimal operating point is considered. To do so, due to the absence of sensors for real-time measurement of the concentrations of key variables of the bioreactor, estimators of these concentrations and of the lactic acid production rate are first developed. Then, control strategies for regulating the lactic acid concentration at its optimal value are designed and compared in simulation. An adaptive control combining a state feedback linearizing control and an estimator of the lactic acid production rate is finally chosen to be experimentally validated on an instrumented reactor. This strategy showed good robustness features with respect to modeling mismatches and was able during experiments to increase twice the lactic acid productivity.

Este trabalho propõe o desenvolvimento de Algoritmos Evolutivos (AEs) para estimação dos parâmetros que modelam sinais elétricos (frequência, fase e amplitude) em tempo-real. A abordagem proposta deve ser robusta a ruídos e harmônicos em sinais distorcidos, por exemplo devido à presença de faltas na rede elétrica. AEs mostram vantagens para lidar com tais tipos de sinais. Por outro lado, esses algoritmos quando implementados em software não possibilitam respostas em tempo-real para uso da estimação como relé de frequência ou Unidade de Medição Fasorial. O desenvolvimento em FPGA apresentado nesse trabalho torna possível paralelizar o cálculo da estimação em hardware, viabilizando AEs para análise de sinal elétrico em tempo real. Além disso, mostra-se que AEs multiobjetivos podem extrair informações não evidentes das três fases do sistema e estimar os parâmetros adequadamente mesmo em casos em que as estimativas por fase divirjam entre si. Em outras palavras, as duas principais contribuições computacionais são: a paralelização do AE em hardware por meio de seu desenvolvimento em um circuito de FPGA otimizado a nível de operações lógicas básicas e a modelagem multiobjetiva do problema possibilitando análises dos sinais de cada fase, tanto independentemente quanto de forma agregada. Resultados experimentais mostram superioridade do método proposto em relação ao estimador baseado em transformada de Fourier para determinação de frequência e fase / This work proposes the development of Evolutionary Algorithms (EAs) for the estimation of the basic parameters from electrical signals (frequency, phase and amplitude) in real time. The proposed approach must be robust to noise and harmonics in signals distorted, for example, due to the presence of faults in the electrical network. EAs show advantages for dealing with these types of signals. On the other hand, these algorithms when implemented in software cant produce real-time responses in order to use their estimations as frequency relay or Phasor Measurement Unit. The approach developed on FPGA proposed in this work parallelizes in hardware the process of estimation, enabling analyses of electrical signals in real time. Furthermore, it is shown that multi-objective EAs can extract non-evident information from the three phases of the system and properly estimate parameters even when the phase estimates diverge from each other. This research proposes: the parallelization of an EA in hardware through its design on FPGA circuit optimized at level of basic logic operations and the modeling of the problem enabling multi-objective analyses of the signals from each phase in both independent and aggregate ways. Experimental results show the superiority of the proposed method compared to an estimator based on Fourier transform for determining frequency and phase

A presente tese trata do estudo dos estimadores algébricos aplicados a sistemas de controle. Tais estimadores são filtros variantes no tempo que estimam as derivadas dos sinais de entrada e funcionam como uma alternativa ao uso de observadores. Este trabalho inclui uma contribuição teórica que permite determinar um limite superior do erro do estimador algébrico. Também, mostramos um resultado que garante estabilidade em malha fechada no contexto do princípio da separação para esta classe particular de sistemas. Um exemplo de levitação magnética baseado em linearização exata com estimadores algébricos de derivadas até segunda ordem implementados digitalmente é tratado com detalhe. Simulações em computador são apresentadas mostrando excelentes resultados. / This thesis deals with the algebraic estimators and their application in closed loop control systems. The algebraic estimators can be implemented as time-varying filters that produce an estimation of the derivatives of the input signal, and they can be an interesting alternative for substituting classical observers. This work includes a theoretic contribution that allows to compute a bound of the estimation error of the algebraic estimator. Furthermore it is shown that any derivative estimator that respects this bound will assure closed loop stability in the context of the separation principle, at least for a class of nonlinear systems. An example of magnetic levitation based on exact linearization and in closed loop with an estimator of the first and second derivatives is considered. A digital implementation of the algebraic estimation is also discussed, and simulations are presented, showing excellent results.

Sensores virtuais empregando modelos de inferência da composição(responsável pela qualidade) dos produtos de uma coluna de destilação correspondem a medidores implementados em software, capazes de estimar, em tempo real, a composição dos produtos da mesma, a partir de informações do tipo temperaturas e pressões em diversos pontos da coluna e vazões de entrada, de saída e de reciclo. O objetivo deste trabalho é obter esse tipo de sensor para uma coluna de destilação, capaz de estimar instantaneamente a composição dos produtos no topo de uma coluna de destilação multicomponente com condensador parcial, empregando redes neurais artificiais. Foi desenvolvido um simulador dinâmico baseado em modelo não-linear da coluna para aquisição de dados. Neste projeto foi incluído um estudo sobre a influência do treinamento parcial no desempenho do sensor virtual. A idéia é estudar o desempenho para o caso de um sensor virtual treinado de antemão, com dados coletados a partir de um simulador da coluna. Este procedimento disponibiliza um sensor operacional, treinado através de um conjunto de dados simulados ou através de um pequeno conjunto de pontos e retreinado, quando dados reais ou um conjunto maior de dados estiver disponível. Outra contribuição importante é o estudo realizado sobre os principais erros que podem ocorrer neste tipo de sensores, que são raramente tratados em publicações científicas. É também proposta uma metodologia para detecção e correção destes erros que foram encontrados e que afetam o comportamento do sensor, alterando sua precisão e capacidade de ser utilizado em um controle inferencial da planta. / Soft sensors for composition inference models (that are responsible for the quality) of distillation column products, correspond to virtual instruments implemented in software. This software is able to estimate, in real time, the composition of the output products of the column, based on information such as temperature and pressure on several points of the column and on input, output and recycle flow. The purpose of this work is to obtain a soft sensor that estimates the instantaneous composition of the product at the top of a multicomponent distillation column with a partial condenser, employing artificial neural networks. The chosen architecture was the feedforward neural network with three layers. It was chosen based on many tested options. It was developed a dynamical simulator of this column for data acquisition based on a non-linear model. In this study, it was included an investigation about the influence of partial training in the performance of the soft sensor. The goal is to study the results achieved in the case of a soft sensor trained beforehand, with data acquired from the simulator of this column. This procedure makes possible to have an operational soft sensor, trained based on a simulated data set or on a small amount of points and then retrained when a real or larger data set is available. Another important contribution is the study performed about the main errors that may appear in this kind of sensor. These errors are rarely mentioned in scientific papers. It also aims at implementing techniques to enable detection and correction of those errors that the soft sensor may present, and that affect the performance of the soft sensor, changing its precision and making it inadequate for inferential control.

O fenômeno de detonação que ocorre em motores de combustão interna é uma restrição de desempenho para baixas emissões de poluentes, altas solicitações de torque (as quais resultariam em mais potência entregue pelo motor), economia de combustível e também é uma fonte de ruídos desagradáveis ao condutor do veículo. No Brasil, tal fenômeno pode ter ainda uma utilidade benéfica se identificado de forma leal, tendo em vista que a detonação é mais fácil de ocorrer com gasolina do que com álcool. Assim, a ocorrência de detonação pode ser um auxílio no gerenciamento otimizado da injeção, ignição, bem como pode possibilitar a identificação do tipo de combustível. Para isso, técnicas de processamento de sinais podem ser utilizadas para caracterizar tal fenômeno. Este trabalho tem como foco a identificação (ocorrência) e caracterização (intensidade e duração) do sinal de detonação utilizando filtragem adaptativa e estimadores de potência, onde o desempenho individual e combinado destes algoritmos é avaliado. Inicialmente, um modelo do sinal de detonação é proposto e amostras do sinal real são obtidas e com isso, mostra-se que as técnicas propostas rendem eficientes caracterizadores do fenômeno. / The knock phenomenon that happens in internal combustion engines is a constraint against performance in terms of low pollution rates, higher torque requests (which would imply higher delivered power by the engine), mileage as well as it is an annoying source of noise to the user. In Brazil, such phenomenon may be useful if identified accurately, due to the fact that knock has a higher likelihood to happen with gasoline than ethanol. Thus, the occurrence of knock may be a support in an optimized management of fuel injection and sparking time, as well as it may provide means to identify the fuel being used. In order to achieve such goal, signal processing techniques may be used for characterization of such phenomenon. This work has the focus of identifying (occurrence) and characterize (intensity and duration) of knock signal using adaptive filtering and power estimators, where the individual and combined performance are assessed. To begin with, a knock signal model is proposed as well as samples of the real signal are taken, with that, it is shown that the proposed techniques offer efficient detectors of the phenomenon.

Some boundary value problems yield anisotropic solutions, e.g. solutions with boundary layers. If such problems are to be solved with the finite element method (FEM), anisotropically refined meshes can be advantageous. In order to construct these meshes or to control the error one aims at reliable error estimators. For isotropic meshes such estimators are known but they fail when applied to anisotropic meshes. Rectangular (or cuboidal) anisotropic meshes were already investigated. In this paper an error estimator is presented for tetrahedral or triangular meshes which offer a much greater geometrical flexibility.

Dans le contexte de la théorie bayésienne et de théorie de la décision, l'estimation d'une densité prédictive d'une variable aléatoire occupe une place importante. Typiquement, dans un cadre paramétrique, il y a présence d’information additionnelle pouvant être interprétée sous forme d’une contrainte. Cette thèse porte sur des stratégies et des améliorations, tenant compte de l’information additionnelle, pour obtenir des densités prédictives efficaces et parfois plus performantes que d’autres données dans la littérature. Les résultats s’appliquent pour des modèles avec données gaussiennes avec ou sans une variance connue. Nous décrivons des densités prédictives bayésiennes pour les coûts Kullback-Leibler, Hellinger, Kullback-Leibler inversé, ainsi que pour des coûts du type $\alpha-$divergence et établissons des liens avec les familles de lois de probabilité du type \textit{skew--normal}. Nous obtenons des résultats de dominance faisant intervenir plusieurs techniques, dont l’expansion de la variance, les fonctions de coût duaux en estimation ponctuelle, l’estimation sous contraintes et l’estimation de Stein. Enfin, nous obtenons un résultat général pour l’estimation bayésienne d’un rapport de deux densités provenant de familles exponentielles. / Abstract: In the context of Bayesian theory and decision theory, the estimation of a predictive density of a random variable represents an important and challenging problem. Typically, in a parametric framework, usually there exists some additional information that can be interpreted as constraints. This thesis deals with strategies and improvements that take into account the additional information, in order to obtain effective and sometimes better performing predictive densities than others in the literature. The results apply to normal models with a known or unknown variance. We describe Bayesian predictive densities for Kullback--Leibler, Hellinger, reverse Kullback-Leibler losses as well as for α--divergence losses and establish links with skew--normal densities. We obtain dominance results using several techniques, including expansion of variance, dual loss functions in point estimation, restricted parameter space estimation, and Stein estimation. Finally, we obtain a general result for the Bayesian estimator of a ratio of two exponential family densities.

We consider tail empirical processes for long memory stochastic volatility models with heavy tails and leverage. We show a dichotomous behaviour for the tail empirical process with fixed levels, according to the interplay between the long memory parameter and the tail index; leverage does not play a role. On the other hand, the tail empirical process with random levels is not affected by either long memory or leverage. The tail empirical process with random levels is used to construct a family of estimators of the tail index, including the famous Hill estimator and harmonic moment estimators. The limiting behaviour of these estimators is not affected by either long memory or leverage. Furthermore, we consider estimators of risk measures such as Value-at-Risk and Expected Shortfall. In these cases, the limiting behaviour is affected by long memory, but it is not affected by leverage. The theoretical results are illustrated by simulation studies.

Some boundary value problems yield anisotropic solutions, e.g. solutions with boundary layers. If such problems are to be solved with the finite element method (FEM), anisotropically refined meshes can be advantageous. In order to construct these meshes or to control the error one aims at reliable error estimators. For \emph{isotropic} meshes many estimators are known, but they either fail when used on \emph{anisotropic} meshes, or they were not applied yet. For rectangular (or cuboidal) anisotropic meshes a modified error estimator had already been found. We are investigating error estimators on anisotropic tetrahedral or triangular meshes because such grids offer greater geometrical flexibility. For the Poisson equation a residual error estimator, a local Dirichlet problem error estimator, and an $L_2$ error estimator are derived, respectively. Additionally a residual error estimator is presented for a singularly perturbed reaction diffusion equation. It is important that the anisotropic mesh corresponds to the anisotropic solution. Provided that a certain condition is satisfied, we have proven that all estimators bound the error reliably.

Singularly perturbed problems often yield solutions ith strong directional features, e.g. with boundary layers. Such anisotropic solutions lend themselves to adapted, anisotropic discretizations. The quality of the corresponding numerical solution is a key issue in any computational simulation. To this end we present a new robust error estimator for a singularly perturbed reaction-diffusion problem. In contrast to conventional estimators, our proposal is suitable for anisotropic finite element meshes. The estimator is based on the solution of a local problem, and yields error bounds uniformly in the small perturbation parameter. The error estimation is efficient, i.e. a lower error bound holds. The error estimator is also reliable, i.e. an upper error bound holds, provided that the anisotropic mesh discretizes the problem sufficiently well. A numerical example supports the analysis of our anisotropic error estimator.