Science and engineering problems frequently require solving a sequence of single linear systems or a sequence of dual linear systems. We develop algorithms that recycle Krylov subspaces and preconditioners from one system (or pair of systems) in the sequence to the next, leading to efficient solutions. Besides the benefit of only having to store few Lanczos vectors, using BiConjugate Gradients (BiCG) to solve dual linear systems may have application-specific advantages. For example, using BiCG to solve the dual linear systems arising in interpolatory model reduction provides a backward error formulation in the model reduction framework. Using BiCG to evaluate bilinear forms -- for example, in the variational Monte Carlo (VMC) algorithm for electronic structure calculations -- leads to a quadratic error bound. Since one of our focus areas is sequences of dual linear systems, we introduce recycling BiCG, a BiCG method that recycles two Krylov subspaces from one pair of dual linear systems to the next pair. The derivation of recycling BiCG also builds the foundation for developing recycling variants of other bi-Lanczos based methods like CGS, BiCGSTAB, BiCGSTAB2, BiCGSTAB(l), QMR, and TFQMR. We develop a generalized bi-Lanczos algorithm, where the two matrices of the bi-Lanczos procedure are not each other's conjugate transpose but satisfy this relation over the generated Krylov subspaces. This is sufficient for a short term recurrence. Next, we derive an augmented bi-Lanczos algorithm with recycling and show that this algorithm is a special case of generalized bi-Lanczos. The Petrov-Galerkin approximation that includes recycling in the iteration leads to modified two-term recurrences for the solution and residual updates. We generalize and extend the framework of our recycling BiCG to CGS, BiCGSTAB and BiCGSTAB2. We perform extensive numerical experiments and analyze the generated recycle space. We test all of our recycling algorithms on a discretized partial differential equation (PDE) of convection-diffusion type. This PDE problem provides well-known test cases that are easy to analyze further. We use recycling BiCG in the Iterative Rational Krylov Algorithm (IRKA) for interpolatory model reduction and in the VMC algorithm. For a model reduction problem, we show up to 70% savings in iterations, and we also demonstrate that solving the problem without recycling leads to (about) a 50% increase in runtime. Experiments with recycling BiCG for VMC gives promising results. We also present an algorithm that recycles preconditioners, leading to a dramatic reduction in the cost of VMC for large(r) systems. The main cost of the VMC method is in constructing a sequence of Slater matrices and computing the ratios of determinants for successive Slater matrices. Recent work has improved the scaling of constructing Slater matrices for insulators, so that the cost of constructing Slater matrices in these systems is now linear in the number of particles. However, the cost of computing determinant ratios remains cubic in the number of particles. With the long term aim of simulating much larger systems, we improve the scaling of computing determinant ratios in the VMC method for simulating insulators by using preconditioned iterative solvers. The main contribution here is the development of a method to efficiently compute for the Slater matrices a sequence of preconditioners that make the iterative solver converge rapidly. This involves cheap preconditioner updates, an effective reordering strategy, and a cheap method to monitor instability of ILUTP preconditioners. Using the resulting preconditioned iterative solvers to compute determinant ratios of consecutive Slater matrices reduces the scaling of the VMC algorithm from O(n^3) per sweep to roughly O(n^2), where n is the number of particles, and a sweep is a sequence of n steps, each attempting to move a distinct particle. We demonstrate experimentally that we can achieve the improved scaling without increasing statistical errors. / Ph. D.

This thesis considers the development of a Hammerstein-bilinear approach to non-linear systems modelling, analysis and control systems design, which builds on and extends the applicability of an existing bilinear approach. The underlying idea of the Hammerstein-bilinear approach is to use the Hammerstein-bilinear system models to capture various physical phenomena of interest and subsequently use these for model based control system designs with the premise being that of achieving enhanced control performance. The advantage of the Hammerstein-bilinear approach is that the well-structured system models allow techniques that have been originally developed for linear systems to be extended and applied, while retaining moderate complexity of the corresponding system identification schemes and nonlinear model based control designs. In recognition of the need to be able to identify the Hammerstein-bilinear models a unified suite of algorithms, being the extensions to the simplified refined instrumental variable method for parameter estimation of linear transfer function models is proposed. These algorithms are able to operate in both the continuous-time and discrete-time domains to reflect the requirements of the intended purposes of the identified models with the emphasis being placed on straightforward applicability of the developed algorithms and recognising the need to be able to operate under realistic practical system identification scenarios. Moreover, the proposed algorithms are also applicable to parameter estimation of Hammerstein and bilinear models, which are special cases of the wider Hammerstein-bilinear model class. The Hammerstein-bilinear approach has been applied to an industrial heating, ventilation and air conditioning (HVAC) system, which has also been the underlying application addressed in this thesis. A unique set of dynamic control design purpose oriented air temperature and humidity Hammerstein-bilinear models of an environmentally controlled clear room manufacturing zone has been identified. The greater insights afforded by the knowledge of the system nonlinearities then allow for enhanced control tuning of the associated commercial HVAC control system leading to an improved overall control performance.

Sistemas Lineares com Saltos Markovianos (SLSMs) são estudados desde a década de 1960 e vêm ganhando visibilidade desde então, com diversas aplicações dentre as quais Finanças, Robótica e Engenharias diversas. Um problema de regulação trata de controlar o SLSM buscando fazer sua trajetória se aproximar de zero. Quando os saltos markovianos são observados, o problema é simples e bem resolvido, muito diferente de quando não se observam os saltos. Neste trabalho é estudado um algoritmo da literatura utilizado para resolver o problema de regulação sem observação dos saltos, chamado Método Variacional (MV). Sendo um dos melhores métodos para o dado problema, enfrenta dificuldades de cunho numérico. Neste trabalho se procura analisar e melhorar o condicionamento dos subproblemas envolvidos, de forma a favorecer a convergência do método. São testadas abordagens diferentes usando precondicionadores e comparados os resultados, permitindo concluir que três das cinco abordagens é que trouxeram os melhores resultados. Por se tratar de sistemas lineares do tipo Ax = b, as abordagens de condicionamento podem ser adaptadas para outros problemas semelhantes. / Markov Jump Linear Systems (MJLSs) have been studied since the decade of 1960 and they are gaining visibility ever since, due to a wide range of applications, such as Finance, Robotics, several Engeneerings among others. The so called regulation problem is to control the MJLS seeking to make its trajectory to approach zero. When markovian jumps are observed, the problem is simple and the solution given as closed formulas, which is quite different from the situation when jumps are not observed. We study an algorithm available in literature called Variational Method (VM). Even though it is one of the best methods to solve the problem, it has some numerical difficulties. We analyse its performance and propose some ideas aiming at the ill-conditioning of the subproblems involved, in order to improve the convergence of the method. Different approaches are tested using preconditioners and the results are compared, indicating that three approaches of the five tested ones are promising for convergence improvement. Because the subproblems are linear systems of type Ax = b, these approaches can be adapted to similar problems.

Ce travail de thèse traite de la problématique d’estimation des défauts et de l’étathybride pour une classe de systèmes linéaires à commutations. L’objectif est de développerune méthode afin de synthétiser un observateur et un estimateur dédiésrespectivement à l’estimation de l’état hybride et des défauts. Après la présentationd’un état de l’art sur les techniques d’estimation, de stabilité et de diagnosticpour les systèmes linéaires à commutations, la thèse est scindée en deux parties.La première partie propose une méthode d’estimation de l’état continu et desdéfauts dans le cas où l’état discret du système est connu. En se basant sur unetransformation de coordonnées qui découple un sous-ensemble de l’état du systèmedes défauts, nous avons synthétisé dans un premier temps un observateur hybridepour estimer l’état continu du système, et dans un second temps, un estimateurpermettant la reconstruction des défauts. L’estimateur de défauts proposé dépendde la dérivée de la sortie du système. Pour cette raison, un différenciateur robusteet exact basé sur des techniques des modes glissants est utilisé. Dans la secondepartie de ce mémoire, l’état discret du système est supposé inconnu. Une approchebasée sur des méthodes algébriques est proposée afin d’estimer les instants decommutation entre les différents sous-systèmes. Par la suite, l’estimation de l’étathybride (état continu et état discret) et des défauts est considérée dans le cas oùl’état discret du système est inconnu. Ce dernier est reconstruit en se basant surles instant de commutation estimé et sur une séquence de commutation connue.L’état continu du système est estimé en se basant sur une méthode de placementde pôles permettant d’améliorer les performances de la phase transitoire. Enfin, enexploitant des résultats trouvés dans la première partie, l’estimation des défautsest considérée en estimant la sortie du système avec un différenciateur algébrique.Ce différenciateur donne des résultats plus intéressants vis-à-vis du bruit par rapportau différenciateur basé sur les techniques des modes glissants utilisé dans lapremière partie. / This work deals with the problem of estimation of fault and hybrid state for a classof switched linear systems. The objective is to develop a method to synthesize anobserver and an estimator dedicated respectively to the estimation of the hybridstate and the faults. After presenting a state of the art for estimation, stabilityand diagnostic techniques for switched linear systems, the report is divided intotwo parts. The first part proposes a method for estimating the continuous stateand the faults in the case where the discrete state of the system is known. Basedon a coordinate transformation which decouples a subset of the state of the systemof faults, we first synthesized a hybrid observer to estimate the continuous stateof the system and, in a second step, an estimator allowing the reconstructionof faults. The proposed fault estimator depends on the derivative of the systemoutput. For this reason, a robust and accurate differentiator based on sliding modetechniques is used. In the second part of this paper, the discrete state of the systemis assumed unknown. An algebraic approach is proposed to estimate the switchingtimes between the different subsystems. Thereafter, the estimation of the hybridstate (continuous and discrete state) and of the faults is considered in the casewhere the discrete state of the system is unknown. The latter is reconstructedfrom the estimated switching times and on a known switching sequence. Thecontinuous state of the system is estimated using a pole placement method allowingimprove the performances of the transient phase. Finally, by exploiting the resultsfound in the first part, the estimation of the faults is considered by estimatingthe output of the system with an algebraic differentiator. This differentiator givesmore interesting results at the noise compared to the differentiator based on thesliding mode techniques used in the first part.

É notório que o emprego de unidades de processamento gráfico (GPU) pode acelerar significativamente métodos numéricos de aplicações científicas. Para suportar essa nova tecnologia, é necessário que programas sejam readaptados, exigindo flexibilidade do código. Nesse trabalho, é apresentada a arquitetura de um programa de elementos finitos (MEF) para a análise de estruturas, com suporte ao processamento em GPU. A orientação a objetos é usada para guiar o desenvolvimento e modelar o código em uma estrutura flexível. A escalabilidade do programa é dada pela extensão de suas funcionalidades através de componentes carregados em tempo de execução. Para ilustrar a robustez do código, o software é aplicado para o estudo de dinâmica estrutural, considerando aspectos complexos de não linearidade de material (plasticidade) e geometria (grandes deslocamentos). A acurácia do código é verificada através da comparação com problemas conhecidos da literatura e com as soluções MEF comerciais (ABAQUS). As comparações mostraram uma boa concordância nos resultados. O speedup da GPU é analisado em relação aos tempos de CPU do próprio programa, sendo observado ganhos de desempenho de até 10 vezes. / It has been recognized that the adoption of graphics processing units (GPU) can significantly boost numerical methods in scientific applications. In order to support such technology, it is necessary to readapt the program, which requires code flexibility. In this work, it is presented the architecture of a finite element (FEM) analysis program for structural analysis with GPU support. Object-oriented design is used to guide development and to build code into a flexible structure. Program scalability is achieved by extensibility of its features, provided by run-time loaded components. In order to demonstrate code robustness, the software is directed to the study of structural dynamics, considering complex non-linear aspects of material (plasticity) and geometry (large displacements). Code accuracy is checked by comparing with known literature problems and with commercial solver packages (ABAQUS). The comparison shows good agreement in the results. The GPU code speedup is analysed against timings of CPU program code, where it is observed performance gain up to 10 times.

Neste trabalho estudamos um controlador denominado rastreador linear quadrático (RLQ) com custo médio de longo prazo (CMLP) para sistemas lineares com saltos markovianos (SLSM). Mostramos que o conceito de detetabilidade uniforme, juntamente com a hipótese de que o regulador linear quadrático associado ao RLQ tenha custo uniformemente limitado, são suficientes para que o controle obtido seja estabilizante em um certo sentido. A partir deste resultado, e considerando as mesmas hipóteses, demonstramos a existência do CMLP. Com isto, estendemos os resultados dispostos na literatura desde que consideramos um sistema variante no tempo e uma estrutura mais geral para a cadeia deMarkov. Além disto, avaliamos a aplicação deste controlador no planejamento da operação de um sistema hidrotérmico. Para isto, utilizamos o sistema de usinas do rio São Francisco, em dois casos de estudo, para comparar o desempenho do controlador estudado em relação à solução ótima para o problema, encontrada com o uso da programação dinâmica estocástica, e em relação à solução obtida via programação dinâmica determinística. Os resultados sugerem que o RLQ pode representar uma alternativa interessante para o problema de planejamento hidrotérmico / In the present work we study the reference tracking controller (RTC) for the long run average cost (LRAC) problem for Markov jump linear systems. We show that uniform detectability and an hypothesis that the linear quadratic regulator associated with the RTC has uniformly bounded cost, together, are sufficient conditions for the obtained control be exponentially stabilizing in a certain sense. This result allows us to demonstrate the existence of the LTAC under the same hypotheses. The results can be regarded as an extension of previous works, since we have considered a more general framework with time-varying systems and quite general Markov chains. As an applicatioin, we consider the operational planning of hydrothermal systems. We have considered some power plants of the Sao Francisco river, in two different scenarios, and we have compared the performances of the RTC and standard controls obtained by deterministic and stochastic dynamic programming, indicating that the RTC may be an interesting alternative for the hydrothermal planning problem

Esta dissertação desenvolve filtro de informação, algoritmos array para estimador do erro médio mínimo quadrático para sistemas lineares sujeitos a saltos Markovianos e algoritmos array rápidos para filtragem de sistemas singulares convencionais. Exemplos numéricos serão apresentados para mostrarem as vantagens dos algoritmos array deduzidos. Parte dos resultados obtidos nesta pesquisa serão publicados no seguinte artigo: Terra et al. (2007). Terra, M. H., Ishihara, J. Y. and Jesus, G. Q. (2007). Information filtering and array algorithms for discrete-time Markovian jump linear systems. Proceedings of the American Control Conference ACC07. / This dissertation develops information filter and array algorithms for linear minimum mean square error estimator (LMMSE) of discrete-time Markovian jump linear systems (MJLSs) and fast array algorithms for filtering of standard singular systems. Numerical examples to show the advantage of the array algorithms are presented. Some results obtained in this research are published in the following paper: Terra et al. (2007). Terra, M. H., Ishihara, J. Y. and Jesus, G. Q. (2007). Information filtering and array algorithms for discrete-time Markovian jump linear systems. Proceedings of the American Control Conference ACC07.

Este trabalho propõe uma metodologia de identificação para sistemas lineares sujeitos a saltos markovianos. Dada uma sequência de observações ruidosas da variável de estados, busca-se estimá-la juntamente com os parâmetros (desconhecidos) que descrevem o sistema dinâmico no espaço de estados. Como é bem conhecido, a ltragem ótima nesta classe de sistemas tem requisitos computacionais exponencialmente crescentes em função do tamanho da amostra, e torna-se inviável na prática. Recorre-se, portanto, a um algoritmo sub-ótimo de ltragem, cujos resultados são utilizados na identificação por máxima verossimilhança segundo a metodologia apresentada. Simulações realizadas mostram boa convergência. / This paper proposes a methodology for the identification of Markov-jump linear systems. Given a sequence of noisy observations of the state variable, our objective is to estimate it along with the (unknown) parameters that drive the system in the state-space. As it is well known, the optimal ltering in this class of systems requires exponentially increasing computing power, in proportion to the sample size, and is not feasible in practice. We resort, therefore, to a sub-optimal algorithm, whose results are used for a maximum likelihood identification according to the methodology presented here. Simulations show a good convergence.

Nesta tese de doutorado são apresentados sistemas globais de estimativa baseados em modelos Markovianos aplicados na área de reabilitação robótica. Os sistemas propostos foram desenvolvidos para estimar as posições angulares dos elos de exoesqueletos para membros inferiores, desenvolvidos para reabilitação motora em pacientes que sofreram Acidente Vascular Cerebral (AVC) ou lesão medular. Filtros baseados no filtro de Kalman, um nominal e outro considerando incertezas no modelo, foram utilizados em estratégias de fusão de dados de sensores provenientes de sensores inerciais, possibilitando estimativas de posicionamentos angulares. Algoritmos genéticos são utilizados na otimização dos filtros, ajustando as matrizes de peso destes. Em oposição as modelagens tradicionais, via estimativa local, utilizando somente uma unidade inercial para cada modelo, propõe-se um sistema global de estimativa, obtendo-se a melhor informação de cada sensor combinando-os em um modelo Markoviano. Resultados experimentais com um exoesqueleto foram utilizados para comparar a abordagem Markoviana às convencionais. / In this thesis are presented global estimation systems based on Markov models applied in robotic rehabilitation area. The proposed systems have been developed to estimate the angular positions of the exoskeletons for lower limbs, designed to provide motor rehabilitation of stroke and spinal cord injured people. Filters based on the Kalman filter, one nominal and other considering uncertainties in the model, were used in sensor data fusion strategies from inertial sensors, to estimate angular positions. Genetic algorithms are used to the optimization of filters, tuning the weighting matrices. In opposition to these modelling via local estimation, using only one inertial unit, we also chose a global modelling getting the best information from each sensor, combining them in a Markov model. Experimental results with an exoskeleton were used to compare the Markovian approach to conventional.

Neste trabalho, é apresentada uma análise de estabilidade linear para um sistema flowline-riser. O modelo utiliza as equações de continuidade para as fases líquidas e gasosas e uma equação de momento para a mistura onde são considerados os efeitos de atrito. A mistura água-óleo é modelada de forma homogênea. Para a determinação da fração de vazio, adota-se o modelo de fluxo de deriva, baseado em várias correlações drift. Se o escoamento é estratificado a fração de vazio é modelada através do modelo de equilíbrio local de Taitel e Dukler (1976). Para a caracterização dos fluidos é adotado um modelo de equilíbrio de fase black-oil onde a transferência de massa é considerada entre as correntes óleo e gás em função das condições locais de pressão e temperatura. É considerada a abordagem de parâmetros distribuídos, onde diversas geometrias com discretização variável podem ser consideradas tais como riser em catenária e riser lazy wave. Para realizar a análise de estabilidade linear, as equações do modelo são linearizadas em torno do estado estacionário e discretizadas pelo método das diferenças finitas, onde foi utilizado um programa escrito em Matlab. A partir do sistema linearizado, é avaliada a estabilidade do estado estacionário pelas raízes do polinômio característico, que são solução do problema de autovalores e autovetores. É avaliada a convergência numérica do modelo e mapas de estabilidade são apresentados para vários sistemas de produção de petróleo. Os resultados obtidos numericamente são comparados com pontos operacionais de sistemas reais de produção de petróleo. Observou-se que o modelo convergiu para todos os casos avaliados e apresentou ótima concordância com os dados de campo. / This work presents a linear stability analysis for a flowline-riser system. The model considers continuity equations for the liquid and gas phases, and a momentum equation for the mixture that accounts for friction effects. The water-oil mixture is modeled as being homogeneous. The void fraction is determined by using the drift-flux model based on several drift correlations. For stratified flow, the void fraction is expressed by the local equilibrium model of Taitel and Dukler (1976). Fluid characterization uses a black-oil model that considers mass transfer between the oil and gas flows as a function of the local pressure and temperature conditions. In the proposed approach with distributed parameters, several geometries with variable discretization can be considered, such as catenary and lazy wave risers. To perform the linear stability analysis, the equations of the model are linearized around the stationary state and discretized using the finite difference method, implemented using custom-written code in Matlab. From the linearized system, the stability of the steady state is evaluated by computing the roots of the characteristic polynomial equation of the eigenvalues and eigenvectors problem. Convergence of the numerical model is evaluated and stability maps for several oil production systems were presented. Numerical results were compared with the actual measurements of oil production systems. The model converged in all cases tested and presented an excellent agreement with field data.