Based on the Lyapunov stability theorem, a modified stability analysis as well as a modified observer is proposed in this thesis for a class of uncertain nonlinear systems with an existent high gain observer. By assuming that the first two state variables are indirectly measurable, reanalyzing the stability of the error dynamics is presented first. The advantage of this modified analytic method is that the upper bound of the disturbance distribution functions is not required to be known in advance, and the asymptotic stability is still guaranteed. Next, based on this existent observer, a slightly modified observer is presented for systems with disturbances whose upper bound is unknown. An adaptive mechanism is embedded in the proposed observer, so that the upper bound of perturbations is not required to be known beforehand. The resultant dynamics of estimation errors can be driven into the sliding surface in a finite time, and guarantee asymptotic stability. A numerical example and a practical example are given to demonstrate the feasibility of the proposed observer.

A major motivation for Lie group observers is their application as sensor fusion algorithms for an inertial measurement unit which can be used to estimate the orientation of a rigid-body. In the first part of this thesis we propose several types of nonlinear, deterministic, locally exponentially convergent, state observers for systems with all, or part, of their states evolving on the general linear Lie group of invertible matrices. Our proposed Lie group observer with full-state measurement is applicable to left-invariant systems on linear Lie groups and yields linear estimation error dynamics. We also propose a way to extend our full-state observer, to build observers with partial-state measurement, i.e., only a proper subset of the states are available for measurement. Our proposed Lie group observer with partial-state measurement is applicable when the measured states are evolving on a Lie group and the rest of the states are evolving on the Lie algebra of this Lie group. We illustrate our observer designs on various examples, including rigid-body orientation estimation and dynamic homography estimation. In the second part of this thesis we propose a nonlinear, deterministic state observer, for systems that evolve on real, finite-dimensional vector spaces. This observer uses the property of high-gain observers, that they are approximate differentiators of the output signal of a plant. Our new observer is called a composite high-gain observer because it consists of a chain of two or more sub-observers. The first sub-observer in the chain differentiates the output of the plant. The second sub-observer in the chain differentiates a certain function of the states of the first sub-observer. Effectiveness of the composite observer is demonstrated via simulation.

A major motivation for Lie group observers is their application as sensor fusion algorithms for an inertial measurement unit which can be used to estimate the orientation of a rigid-body. In the first part of this thesis we propose several types of nonlinear, deterministic, locally exponentially convergent, state observers for systems with all, or part, of their states evolving on the general linear Lie group of invertible matrices. Our proposed Lie group observer with full-state measurement is applicable to left-invariant systems on linear Lie groups and yields linear estimation error dynamics. We also propose a way to extend our full-state observer, to build observers with partial-state measurement, i.e., only a proper subset of the states are available for measurement. Our proposed Lie group observer with partial-state measurement is applicable when the measured states are evolving on a Lie group and the rest of the states are evolving on the Lie algebra of this Lie group. We illustrate our observer designs on various examples, including rigid-body orientation estimation and dynamic homography estimation. In the second part of this thesis we propose a nonlinear, deterministic state observer, for systems that evolve on real, finite-dimensional vector spaces. This observer uses the property of high-gain observers, that they are approximate differentiators of the output signal of a plant. Our new observer is called a composite high-gain observer because it consists of a chain of two or more sub-observers. The first sub-observer in the chain differentiates the output of the plant. The second sub-observer in the chain differentiates a certain function of the states of the first sub-observer. Effectiveness of the composite observer is demonstrated via simulation.

La chirurgie minimalement invasive est aujourd’hui une thématique de recherche particulièrement active. Un traitement thérapeutique ciblé et la possibilité d’établir un diagnostic précis grâce à l’utilisation de systèmes miniaturisés peuvent considérablement améliorer de nombreuses pratiques médicales. Le recours à des microrobots actionnés à distance et naviguant dans le système cardiovasculaire ouvre de nouvelles perspectives. L’objectif de cette thèse est de proposer un socle théorique solide concernant i) la modélisation d’un microrobot naviguant dans le système cardiovasculaire, ii) l’élaboration de lois de commande et d’observateurs assurant un bon suivi de trajectoire depuis la zone d’injection jusqu’à une zone cible. La modélisation du système fait intervenir de nombreuses forces : forces hydrodynamiques, forces surfaciques (électrostatique, van der Waals, stériques), forces de contact et poids apparent du microrobot. Ce microrobot est contrôlé dans le système cardiovasculaire par l’application de champs ou de gradients de champ magnétique selon le design du microrobot. La prise en compte de l’ensemble des forces aboutit à une représentation d’état sous la forme d’un système non-linéaire affine en la commande avec dérive comportant de nombreux paramètres physiologiques incertains. Une trajectoire de référence optimisée est déduite du modèle. L’approche de commande adoptée est établie à partir de critères de stabilité du système. Le système étant non-linéaire, une commande de type Lyapunov stabilisante est développée suivant une approche de type backstepping. L’estimation de certains paramètres physiologiques est rendue possible par une commande de type backstepping adaptatif. Un observateur grand gain reconstruit l’état complet du système nécessaire au calcul de la commande. La stabilité et la robustesse de l’ensemble sont établies au travers de nombreuses simulations en présence de bruits de mesure et d’erreurs paramétriques. / Minimally invasive medical procedures are currently an active research aera. A drug targeted therapy and the possibility of establishing an accurate diagnosis through the use of miniaturized systems can greatly improve many medical practices. The use of untethered microrobots navigating in the cardiovascular system opens new perspectives. The objective of this PhD work is to provide a theoretical approach on i) the modeling of a microrobot navigating in the cardiovascular system, ii) the development of control laws and observers to ensure a fine tracking from the injection to a target area. Modeling such as system involves many forces : hydrodynamic forces, surface forces (electrostatic, van derWaals, steric), contact forces and apparent weight of the microrobot. This microrobot is controlled in the cardiovascular system by the application of magnetic fields or magnetic field gradients according to the design of the microrobot. The consideration of all the forces leads to a state representation in the form of a nonlinear system with many physiological uncertain parameters, but gives us sufficient informations to plan an optimal trajectory. The control approach is established based on stability consideration. A Lyapunov-stabilizing control is then developed using a backstepping approach. An adaptive backstepping control law estimates some physiological parameters. A high gain observer reconstructs the full state of the system required for implementing the control approach. Robustness and stability of the controller with respect to noise measurement, parameters variations and uncertainties are illustrated by simulations.

Ce travail a pour but la réalisation d'un outil permettant d'estimer la germination et la distribution en taille de cristaux d'un procédé de cristallisation. Pour ce faire un modèle de cristallisation a été développé et nous a permis de réaliser un observateur de dimension finie afin d'estimer la germination. Par la suite, un observateur de dimension infinie a été mis au point dans le but d'estimer la distribution de taille des cristaux à partir de la germination et de la mesure d'une taille de la distribution. Les deux observateurs ont été validés expérimentalement sur un procédé "batch" de cristallisation de l'oxalate d'ammonium. / The goal of this thesis is the development of an observer to estimate the nucleation and the crystal size distribution of a crystallization process. We first developed a model of crystallization in order to realize an observer for nucleation estimation. Then an infinite dimension observer was developed to estimate the crystal size distribution, using the nucleation estimation and one crystal size distribution measurement. An experimental validation was performed using the crystallization of ammonium oxalate.

Ce manuscrit porte sur la synthèse d'observateurs grand gain pour des systèmes non linéaires à sorties échantillonnées et retardées. Trois contributions sont proposées à la lecture de ce manuscrit. La première contribution, pour une classe de systèmes Multi-entrées / Multi-sorties uniformément observables et dont les sorties sont regroupées en un seul bloc, met en jeu le problème du processus d'acquisition des mesures de sorties (continues, échantillonnées, retardées ou non) et propose un cadre commun pour l'ensemble des cas possibles. La deuxième contribution propose un observateur grand gain filtré sur cette même classe de systèmes dans l'optique de réduire la sensibilité au bruit de mesure, dans le cas où la sortie est continue puis dans le cas où cette dernière est échantillonnée. La dernière contribution vise à étendre la synthèse grand gain standard pour une large classe de systèmes Multi-entrées / Multi-sorties uniformément observables dont les mesures des sorties sont continues. Pour chacune de ces contributions, il a été montré que l'erreur d'observation de chacun des observateurs proposés converge exponentiellement vers zéro en l'absence d'incertitudes sur le système. Toutes ces contributions ont été illustrées par différents exemples issus de plusieurs domaines d'étude. / This manuscript deals with the synthesis of high gain observers for nonlinear systems with sampled and delayed outputs. Three contributions are proposed for consideration in this manuscript. The first contribution, for a class of Multi-input / Multi-output systems whose outputs are grouped into a single block, involves the problem of the acquisition process of output measurements (continuous, sampled, delayed or not) and proposes a common framework for all possible cases. The second contribution proposes a filtered high gain observer on this same class of systems in order to reduce the sensitivity to measurement noise, in the case where the output is continuous and then in the case where the latter is sampled. The last contribution aims to extend the standard high gain synthesis for a large class of uniformly observable Multi-input / Multi-output systems with continuous output measurements. For each of these contributions, it has been shown that the observation error of each of the proposed observers converges exponentially towards zero in the absence of uncertainties in the system. All these contributions have been illustrated through several examples from different fields of study.

Contrairement aux systèmes linéaires, il n’existe pas de méthode systématique pour la synthèse d’observateurs pour systèmes non linéaires. Cependant, la synthèse peut être plus ou moins simple suivant les coordonnées choisies pour exprimer la dynamique. Des structures particulières, appelées formes canoniques, ont notamment été identifiées comme permettant la construction facile et directe d’un observateur. Une façon usuelle de résoudre ce problème consiste donc à chercher un changement de coordonnées réversible transformant l’expression de la dynamique dans l’une de ces formes canoniques, puis à synthétiser l’observateur dans ces coordonnées, et enfin à en déduire une estimation de l’état du système dans les coordonnées initiales par inversion de la transformation. Cette thèse contribue à chacune de ces trois étapes. Premièrement, nous montrons l’intérêt d’une nouvelle forme triangulaire avec des non linéarités continues (non Lipschitz). En effet, les systèmes observables pour toutes entrées, mais dont l'ordre d’observabilité différentielle est supérieur à la dimension du système, peuvent ne pas être transformables dans la forme triangulaire Lipschitz standard, mais plutôt dans une forme triangulaire "seulement continue". Le célèbre observateur grand gain n’est alors plus suffisant, et nous proposons d’utiliser plutôt des observateurs homogènes.Une autre forme canonique intéressante est la forme linéaire Hurwitz, qui admet un observateur trivial. La question de la transformation d’un système non linéaire dans une telle forme n’a été étudiée que pour les systèmes autonomes à travers les observateurs de Kazantzis-Kravaris ou de Luenberger. Nous montrons ici comment cette synthèse, consistant à résoudre une EDP, peut être étendue aux systèmes instationnaires/commandés. Quant à l’inversion de la transformation, cette étape est loin d’être triviale en pratique, surtout lorsque les espaces de départ et d’arrivée ont des dimensions différentes. En l’absence d’expression explicite et globale de l’inverse, l’inversion numérique repose souvent sur la résolution d’un problème de minimisation couteux en calcul. C’est pourquoi nous développons une méthode qui permet d’éviter l’inversion explicite de la transformation en ramenant la dynamique de l’observateur (exprimée dans les coordonnées de la forme canonique) dans les coordonnées initiales du système. Ceci passe par l’ajout de nouvelles coordonnées et par l’augmentation d’une immersion injective en un difféomorphisme surjectif. Enfin, dans une partie totalement indépendante, nous proposons aussi des résultats concernant l’estimation de la position du rotor d’un moteur synchrone à aimant permanent en l’absence d’informations mécaniques (sensorless) et lorsque des paramètres tels que la résistance ou le flux de l’aimant sont inconnus. Ceci est illustré par des simulations sur données réelles. / Unlike for linear systems, no systematic method exists for the design of observers for nonlinear systems. However, observer design may be more or less straightforward depending on the coordinates we choose to express the system dynamics. In particular, some specific structures, called canonical forms, have been identified for allowing a direct and easier observer construction. It follows that a common way of addressing the problem consists in looking for a reversible change of coordinates transforming the exression of the system dynamics into one of those canonical forms, design an observer in those coordinates, and finally deduce an estimate of the system state in the initial coordinates via inversion of the transformation. This thesis contributes to each of those three steps.First, we show the interest of a new triangular canonical form with continuous (non-Lipschitz) nonlinearities. Indeed, we have noticed that systems which are observable for any input but with an order of differential observability larger than the system dimension, may not be transformable into the standard Lipschitz triangular form, but rather into an "only continuous" triangular form. In this case, the famous high gain observer no longer is sufficient, and we propose to use homogeneous observers instead.Another canonical form of interest is the Hurwitz linear form which admits a trivial observer. The question of transforming a nonlinear system into such a form has only been addressed for autonomous systems with the so-called Lunberger or Kazantzis-Kravaris observers. This design consists in solving a PDE and we show here how it can be extended to time-varying/controlled systems.As for the inversion of the transformation, this step is far from trivial in practice, in particular when the domain and image spaces have different dimensions. When no explicit expression for a global inverse is available, numerical inversion usually relies on the resolution of a minimization problem with a heavy computational cost. That is why we develop a method to avoid the explicit inversion of the transformation by bringing the observer dynamics (expressed in the canonical form coordinates) back into the initial system coordinates. This is done by dynamic extension, i-e by adding some new coordinates to the system and augmenting an injective immersion into a surjective diffeomorphism.Finally, in a totally independent part, we also provide some results concerning the estimation of the rotor position of a permanent magnet synchronous motors without mechanical information (sensorless) and when some parameters such as the magnet flux or the resistance are unknown. We illustrate this with simulations on real data.

No description available.

No description available.

The work concerns the "observability problem"--the reconstruction of a dynamic process's full state from a partially measured state-- for nonlinear dynamic systems. The Extended Kalman Filter (EKF) is a widely-used observer for such nonlinear systems. However it suffers from a lack of theoretical justifications and displays poor performance when the estimated state is far from the real state, e.g. due to large perturbations, a poor initial state estimate, etc. . . We propose a solution to these problems, the Adaptive High-Gain (EKF). Observability theory reveals the existence of special representations characterizing nonlinear systems having the observability property. Such representations are called observability normal forms. A EKF variant based on the usage of a single scalar parameter, combined with an observability normal form, leads to an observer, the High-Gain EKF, with improved performance when the estimated state is far from the actual state. Its convergence for any initial estimated state is proven. Unfortunately, and contrary to the EKF, this latter observer is very sensitive to measurement noise. Our observer combines the behaviors of the EKF and of the high-gain EKF. Our aim is to take advantage of both efficiency with respect to noise smoothing and reactivity to large estimation errors. In order to achieve this, the parameter that is the heart of the high-gain technique is made adaptive. Voila, the Adaptive High-Gain EKF. A measure of the quality of the estimation is needed in order to drive the adaptation. We propose such an index and prove the relevance of its usage. We provide a proof of convergence for the resulting observer, and the final algorithm is demonstrated via both simulations and a real-time implementation. Finally, extensions to multiple output and to continuous-discrete systems are given.