No description available.

Cette thèse porte sur le développement de nouvelles stratégies de commande et d’observation adaptatives par Modes Glissants (MG) et par Modes Glissants d’Ordres Supérieurs (MGOS). En effet, la mise en œuvre des commandes par MG et MGOS classiques nécessite la connaissance des limites supérieures des perturbations ou de leurs dérivées, souvent inconnues. Le premier apport de cette thèse est la synthèse d’une stratégie d’adaptation permettant d'assurer la convergence de la variable de glissement vers un voisinage prédéfini de zéro sans nécessiter d'informations sur les perturbations ou leurs dérivées et sans surestimation du gain. Cette stratégie est ensuite déclinée pour concevoir : deux commandes par MG d’ordre 1 et 2, une commande par mode glissant intégral, ainsi qu’une version du différenciateur de Levant. La deuxième contribution de la thèse est la mise au point de deux commandes adaptatives par MGOS discontinues. Ces deux algorithmes assurent un mode glissant d'ordre n en s’affranchissant de la connaissance de la limite supérieure de la perturbation et de sa dérivée. Enfin, afin de montrer l’efficacité des algorithmes proposés, ils sont appliqués avec succès à travers des simulations pour la commande d’un système de conversion de l’énergie éolienne et la commande d’un moteur à induction linéaire pour la cogénération. / This thesis deals with the development of novel strategies to adapt higher order sliding mode controllers and observers. The implementation of classics first order and higher order sliding mode controllers requires the knowledge of the upper bound of the disturbance or its derivative, which are often not known. The first contribution of this thesis is the design of an adaptive strategy that can ensure the convergence of the sliding variable to a predefined neighborhood of zero without requiring any information of the disturbance or its derivative and without overestimating the adaptive gain. This adaptive strategy is then declined for the design of the first order, second order and integral sliding mode controllers, and for the Levant's differentiator. The second contribution of the thesis is the development of two adaptive strategies for discontinuous higher order sliding mode control. The proposed two algorithms can provide the achievement of n-order sliding mode despite disturbances with unknown upper bounds or with unknown upper bounds of their derivatives. Finally, in order to show the effectiveness of the proposed algorithms, they are successfully applied through simulations to control the wind energy conversion system and the linear induction motor system for cogeneration.

Les principaux défis pour le déploiement de systèmes de conversion de l'énergie éolienne est de maximiser la puissance électrique produite, malgré les variations des conditions météorologiques, tout en minimisant les coûts de fabrication et de maintenance du système. L'efficacité de la turbine éolienne est fortement dépendante des perturbations de l'environnement et des paramètres variables du système, tels que la vitesse du vent et l'angle de tangage. Les incertitudes sur le système sont difficiles à modéliser avec précision alors qu'ils affectent sa stabilité.Afin d'assurer un état de fonctionnement optimal, malgré les perturbations, le commande adaptative peut jouer un rôle déterminant. D'autre part, la synthèse de commandes tolérantes aux défauts, capables de maintenir les éoliennes connectées au réseau après la survenance de certains défauts est indispensable pour le bon fonctionnement du réseau. Le travail de cette thèse porte sur la mise en place de lois de commande adaptatives et tolérantes aux défauts appliqués aux systèmes de conversion de l'énergie éolienne. Après un état de l'art, les contributions de la thèse sont :Dans la première partie de la thèse, un modèle incertain non linéaire du système de conversion d'énergie éolienne avec un générateur à induction à double alimentation est proposé. Une nouvelles approches de commande adaptative par mode glissant est synthétisée et ensuite appliquée pour optimiser l'énergie issue de l'éolienne.Dans la deuxième partie, une nouvelle commande par modes glissants tolérante aux défauts et basée sur les modes glissants intégrales est présentée. Puis, cette méthode est appliquée afin de forcer la vitesse de la turbine éolienne à sa valeur optimale en prenant en compte des défauts qui surviennent sur l'actionneur. / The main challenges for the deployment of wind energy conversion systems (WECS) are to maximize the amount of good quality electrical power extracted from wind energy over a significantly wide range of weather conditions and minimize both manufacturing and maintenance costs. Wind turbine's efficiency is highly dependent on environmental disturbances and varying parameters for operating conditions, such as wind speed, pitch angle, tip-speed ratio, sensitive resistor and inductance. Uncertainties on the system are hard to model exactly while it affects the stability of the system. In order to ensure an optimal operating condition, with unknown perturbations, adaptive control can play an important role. On the other hand, a Fault Tolerant Control (FTC) with control allocation that is able to maintain the WECS connected after the occurrence of certain faults can avoid major economic losses. The thesis work concerns the establishment of an adaptive control and fault diagnosis and tolerant control of WECS. After a literature review, the contributions of the thesis are:In the first part of the thesis, a nonlinear uncertain model of the wind energy conversion system with a doubly fed induction generator (DFIG) is proposed. A novel Lyapunov-based adaptive Sliding Mode (HOSM) controller is designed to optimize the generated power.In the second part, a new output integral sliding mode methodology for fault tolerant control with control allocation of linear time varying systems is presented. Then, this methodology has been applied in order to force the wind turbine speed to its optimal value the presence of faults in the actuator.

The motivation of this thesis is to propose a robust control technique for a laser beam system with target estimation. The laser beam is meant to track and fall on a particular portion of the target until the operation is accomplished. There are many applications of such a system. For example, laser range finder uses laser beam to determine the distance of the target from the source. Recently, unmanned aerial drones have been developed that run on laser power. Drone batteries can be recharged with power sup-ply from laser source on the ground. Laser is also used in high energy laser weapon for defence applications. However, laser beams travelling long distances deviate from the desired location on the target due to continually changing atmospheric parameters (jitter effect) such as pressure, temperature, humidity and wind speed. This deviation error is controlled precisely using a lightweight fast steering mirror (FSM) for fine correction. Furthermore, for a moving target, minimizing the deviation of the beam is not sufficient. Hence, in coarse correction, the target has to be tracked by determining its position and assigning the corresponding azimuth and elevation angles to the laser sources. Once these firing angles are settled within an accuracy of +3 mrad, the effort for minimizing the beam deviation (fine correction) takes place to improve the accu-racy to +10 rad. The beam deviation due to jitter effect is measured by a narrow field of view (NFOV) camera at a high frame rate (1000 frames per second), which takes one frame to com-pute this error information. As a result, controller receives error information witha delay from NFOV. This data cannot be modelled for prediction and hence, a few promising data driven techniques have been implemented for one step ahead prediction of the beam deviation. The predictions are performed over a set of sliding window data online after rejecting the outliers through least square approximated straight line. In time domain, methods like auto-regressive least square, polynomial extrapolation (zeroth, first and second order), Chebyshev polynomial extrapolation, spline curve extrapolation are implemented. Further, a convex combination of zeroth order hold and spline extrapolation is implemented. In frequency domain, Fourier series-Fourier transform and L-point Discrete Fourier Transform stretching are implemented where the frequency component of the signal are analysed properly and propagated for one step ahead prediction. After one step ahead prediction, three nominal controllers (PID, DI and DLQR) are designed such that the output of FSM tracks the predicted beam deviation and the performances of these controllers are compared. Since the FSM is excited by high frequency signals, its performance degrades, which leads to parameter degradation in the mathematical model. Hence, three adaptive controllers have been implemented, namely, model reference adaptive control (MRAC), model reference adaptive sliding mode control (MRASMC) and model following neuro-adaptive control (MFNAC). The parameters of the FSM model are degraded up to 20% and the model is augmented with cross coupling terms because the same mirror is used for horizontal and vertical beam deviation. With this condition, the tracking performance and control rate energy consumption of the implemented adaptive controllers are analysed to choose the best among them. For a moving target, in coarse correction, two tracking radars are placed to measure the position of the target. However, this information is assumed to be noisy, for which an extended Kalman filter is implemented. Once the position of the target is known, the desired firing angles of the laser sources are determined. Given the laser source steering mathematical model, a controller is designed such that it tracks the desired firing angle. Once the residual error of the coarse correction settles inside 3 mrad, fine correction takes part to reduce the residual error to 10 rad. The residual error magnitude of the proposed mechanization was analysed for a moving target by perturbing the FSM model by 20% and zeroth order hold predictor with different combinations of angle tolerance and frame tolerance.