À l'aube de la quatrième révolution industrielle, la comanipulation robotique est une technologie clé tant elle allie la dextérité de l'opérateur humain à la puissance de la machine. Ce partage de tache entre humain et la machine, en sus dans un environnement incertain et inconnu à l'avance, apporte un lot de difficultés intrinsèques à la nature de cette interaction. Cette problématique a été intensivement étudiée durant les vingt dernières années, par diverses équipes de recherches, le plus souvent sur des dispositifs à un seul degré de liberté et avec des hypothèses fortes sur la nature du contrôleur. Dans la présente thèse, nous traitons la problématique de la comanipulation robotique à travers la commande Hoo structurée, cadre de travail particulièrement adapté aux systèmes multivariables et pouvant être étendu à une certaine classe de systèmes non linéaires – les robots manipulateurs en font partie – à travers une modélisation linéaire à paramètres variants (LPV). Les exigences de performance et de stabilité propres aux systèmes de comanipulation sont exprimées en termes de contraintes Hoo et de conditions de secteurs à respecter. Les objectifs de commande ainsi formalisés, sont résolus par optimisation non lisse afin de tirer profit des structures particulières des contrôleurs de robots de comanipulation. La validation de la méthodologie est réalisée par des simulations intensives et des expérimentations sur des dispositifs réels. / At the dawn of the fourth industrial revolution, robotic comanipulation is a key technology as it combines the dexterity of the human operator with the power of the machine. This task sharing between human and machine, in an uncertain and previously unknown environment, brings a lot of intrinsic difficulties to the nature of this interaction. This problem has been intensively studied over the last two decades by various research teams, mostly on devices with a single degree of freedom and with strong hypotheses about the controller structure. In this thesis, we deal with the problem of robotic comanipulation through the scope of the structured Hoo control, a framework particularly adapted to multivariable systems and which can be extended to a certain class of non-linear systems – manipulating robots are part of it – through linear parameter varying (LPV) models. The performance and stability requirements specific to comanipulation systems are expressed in terms of Hoo constraints and sector bounds. The control objectives thus formalised are solved by non-smooth optimization in order to take advantage of the particular structures of the comanipulation robot controllers. The validity of the methodology is carried out by intensive simulations and experiments on real devices.

Os problemas de corte e empacotamento são muito comuns nas indústrias e na logística. Dado um conjunto de N itens com diferentes pesos e um conjunto de M contentores com capacidade C, o problema de empacotamento unidimensional consiste em determinar o menor número de contentores a serem utilizados para alocar todos os itens respeitando a restrição de capacidade dos contentores. Nesse estudo pretende-se resolver o problema com instâncias benchmark da literatura, por meio de sessenta heurísticas diferentes, que são comparadas a quatro limitantes inferiores propostos na literatura com o intuito de avaliar a qualidade da solução heurística. Quatro limitantes inferiores e dez heurísticas construtivas diferentes foram programados em C++ num mesmo ambiente computacional, permitindo sua comparação tanto em termos de qualidade das soluções, quanto em termos dos tempos de processamento. Uma heurística simples de troca de itens entre contentores chamada Diferença-de-Quadrados foi proposta para melhorar as soluções iniciais do problema. A metaheurística simulated annealing foi acionada para melhorar a solução inicial quando o limitante inferior não foi atingido. Os parâmetros dos simulated annealing foram determinados com os dados das instâncias de forma diferente da utilizada na literatura. As combinações entre as dez soluções iniciais, a heurística Diferença-de-Quadrados e o simulated annealing geraram um conjunto de sessenta heurísticas diferentes. Os resultados mostraram que o algoritmo proposto é eficiente para resolver o problema com tempos de processamento adequados a tomada de decisão. / Cutting and packing problems are very common in industries and logistics. Given a set of N items with different weights and a set of M bins with full capacity C, the one-dimensional bin packing problem consists of determining the smallest number of bins capable to allocate all items respecting the capacity constraint of the bins. that impose that the sum of the weights of the items allocated to the bin is less than or equal to their capacity. In this study we intend to solve the problem with benchmark instances of the literature, by means of sixty different heuristics, which are compared to four lower bounds proposed in the literature in order to evaluate the quality of the heuristic solution. Four lower bounds and ten different constructive heuristics were programmed in C++ in the same computational environment, allowing their comparison both in terms of the quality of the solutions and in terms of processing times. A simple heuristic of item exchange between bins called Difference-of-Squares was proposed to improve the initial solutions of the problem. The simulated annealing metaheuristic was triggered to improve the initial solution when the lower bounds was not reached. The parameters of the simulated annealing were determined with the data of the instances differently from that used in the literature. The combinations of the ten initial solutions, the Difference-of-Squares heuristic and the simulated annealing generated a set of sixty different heuristics. The results showed that the proposed algorithm is efficient to solve the problem with adequate processing times for decision making.

Made available in DSpace on 2014-06-11T19:22:18Z (GMT). No. of bitstreams: 0 Previous issue date: 2001-02-17Bitstream added on 2014-06-13T19:07:26Z : No. of bitstreams: 1 fiorotto_dj_me_sjrp.pdf: 485977 bytes, checksum: 8cd2b3ba49a25a9c6a863795f27811c3 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / O problema de dimensionamento de lotes é um problema de otimização da produção, em que o objetivo é planejar a quantidade de itens a ser produzida em várias, ou única, máquinas em cada período ao longo do horizonte de tempo, de modo a tender uma demanda e otimizar uma função objetivo. Este trabalho aborda o problema de dimensionamento de lotes em um único estágio em um ambiente com máquinas paralelas distintas. Cada item pode ser produzido em qualquer máquina, acarretando um tempo de preparação que é gasto antes de começar a produção. O objetivo do trabalho consiste em obter limitantes inferiores de boa qualidade para este problema. Para tanto, é desenvolvido um método de solução baseado numa reformulação do problema a e na relaxação lagrangiana de um conjunto de restrições. Alguns resultados computacionais são apresentados algumas propostas futuras para a continuidade do trabalho. / The lot-sizing problem is a production optimization problem, where the objective is to plan the quantity of items to be produced in multiple, or single, machines in each period over a time horizon, in order to satisfy a demand and optimize an objective function. This work addresses the single stage parallel machine lot-sizing problem. Each item can be produced on any machine, and incur a setup time before to start the production. The objective of this work is to lower bounds of good quality for this problem. A solution method is developed based on a reformulation of the problem and the Lagrangian relaxation of a set of constrainsts. Some computational results are presented comparing the proposed method with a method from the literature, and, some future researches are proposed.

In last few years Value-at-Risk (Var) is a very popular and frequently used risk measure. Risk measure VaR is used in most of the financial institutions. VaR is popular thanks to its simple interpretation and simple valuation. Valuation of VaR is a problem if we assume a few dependent risks. So VaR is estimated in a practice. In presented thesis we study theory of stochastic bounding. Using this theory we obtain bounds for VaR of sum a few dependent risks. In next part of presented thesis we show how we can generalize obtained bounds by theory of copulae. Then we show numerical algorithm, which we can use to evaluate bounds, when exact analytical evaluate isn't possible. In a final part of presented thesis we show our results on practical examples.

We study the relationship between geometric properties of toric varieties and combinatorial properties of the corresponding lattice polytopes. In particular, we give a bound for a very ample lattice polytope to be k-normal. Equivalently, we give a new combinatorial bound for the Castelnuovo-Mumford regularity of normal projective toric varieties. We also give a new combinatorial proof for a special case of Reider's Theorem for smooth toric surfaces.

The Hirsch Conjecture states that for a d-dimensional polytope with n facets, the diameter of the graph of the polytope is at most n-d. This conjecture was disproven in 2010 by Francisco Santos Leal. However, a polynomial bound in n and d on the diameter of a polytope may still exist. Finding a polynomial bound would provide a worst-case scenario runtime for the Simplex Method of Linear Programming. However working only with polytopes in higher dimensions can prove challenging, so other approaches are welcome. There are many equivalent formulations of the Hirsch Conjecture, one of which is the Combinatorial Polynomial Hirsch Conjecture, which turns the problem into a matter of counting sets. This thesis explores the Combinatorial Polynomial Hirsch Conjecture.

We present an implicit a-posteriori finite element procedure to compute bounds for functional outputs of finite element solutions in large strain elasticity. The method proposed relies on the existence of a potential energy functional whose local minima, over a space of suitably chosen continuous functions, corresponds to the problem solution. The output of interest is cast as a constrained minimization problem over an enlarged discontinuous finite element space. A Lagrangian is formed were the multipliers are an adjoint solution, which enforces equilibrium, and hybrid fluxes, which constrain the solution to be continuous. By computing approximate values for the multipliers on a coarse mesh, strict upper and lower bounds for the output of interest on a suitably refined mesh, are obtained. This requires a minimization over a discontinuous space, which can be carried out locally at low cost. The computed bounds are uniformly valid regardless of the size of the underlying coarse discretization. The method is demonstrated with two applications involving large strain plane stress incompressible neo-hookean hyperelasticity. / Singapore-MIT Alliance (SMA)

The optimization, control, and characterization of engineering components or systems require fast, repeated, and accurate evaluation of a partial-differential-equation-induced input-output relationship. We present a technique for the rapid and reliable prediction of linear-functional outputs of elliptic partial differential equations with affine parameter dependence. The method has three components: (i) rapidly convergent reduced{basis approximations; (ii) a posteriori error estimation; and (iii) off-line/on-line computational procedures. These components -- integrated within a special network architecture -- render partial differential equation solutions truly "useful": essentially real{time as regards operation count; "blackbox" as regards reliability; and directly relevant as regards the (limited) input-output data required. / Singapore-MIT Alliance (SMA)

Sigmoid type belief networks, a class of probabilistic neural networks, provide a natural framework for compactly representing probabilistic information in a variety of unsupervised and supervised learning problems. Often the parameters used in these networks need to be learned from examples. Unfortunately, estimating the parameters via exact probabilistic calculations (i.e, the EM-algorithm) is intractable even for networks with fairly small numbers of hidden units. We propose to avoid the infeasibility of the E step by bounding likelihoods instead of computing them exactly. We introduce extended and complementary representations for these networks and show that the estimation of the network parameters can be made fast (reduced to quadratic optimization) by performing the estimation in either of the alternative domains. The complementary networks can be used for continuous density estimation as well.

The thesis treats model reduction for linear time-varying systems. Time-varying models appear in many fields, including power systems, chemical engineering, aeronautics, and computational science. They can also be used for approximation of time-invariant nonlinear models. Model reduction is a topic that deals with simplification of complex models. This is important since it facilitates analysis and synthesis of controllers. The thesis consists of two parts. The first part provides an introduction to the topics of time-varying systems and model reduction. Here, notation, standard results, examples, and some results from the second part of the thesis are presented. The second part of the thesis consists of four papers. In the first paper, we study the balanced truncation method for linear time-varying state-space models. We derive error bounds for the simplified models. These bounds are generalizations of well-known time-invariant results, derived with other methods. In the second paper, we apply balanced truncation to a high-order model of a diesel exhaust catalyst. Furthermore, we discuss practical issues of balanced truncation and approximative discretization. In the third paper, we look at frequency-domain analysis of linear time-periodic impulse-response models. By decomposing the models into Taylor and Fourier series, we can analyze convergence properties of different truncated representations. In the fourth paper, we use the frequency-domain representation developed in the third paper, the harmonic transfer function, to generalize Bode's sensitivity integral. This result quantifies limitations for feedback control of linear time-periodic systems. / QC 20120206