Touat M. DESIGN OF MULTIVARIABLE ROBUST FLIGHT CONTROL SYSTEMS WITH ELEMENTS OF ARTIFICIAL INTELLIGENCE

The thesis is devoted to the design of UAV multivariable robust flight control based on elements from artificial intelligence. The purpose of these methods is to achieve advantages over classical control methods by improving the performance without scarifying the robustness of the controlled system. The first method used is based on classical control in order to reduce the sensors installed on the UAV airborne. This technique uses an estimator to restore the full state vector. Hence to achieve the desired performance the linear quadratic regulator is designed. After attaining the desired performance for the nominal model; it is important to verify if the same results could be accomplished for the parametrically perturbed models. To achieve these requirements the procedure H2/Hinf- multivariable robust optimization using genetic algorithms is applied. The idea behind this method is to find a compromise between the performance and robustness of the closed loop system. The utilization of the genetic algorithms instead of traditional optimization procedure is to have more options to reach the global minimum. Moreover, the formed fitness function comports several and contradictory objectives, thus the traditional optimization methods could converge to the local minimum. Another important perspective control method used in this thesis is the combination of "crisp" and fuzzy logic control. This is essential, especially in the area of unmanned aerial vehicle, where it is hard to compute the exact mathematical model. The structure of the controller is divided into two loops: inner loop and outer loop. The inner loop controller is designed using the aforementioned method (H2/H?-robust optimization). The outer loop controller is designed using fuzzy inference system. The design of fuzzy controller is based only on the expert knowledge about the motion of the UAV. If this knowledge is not available one can add an adaptive mechanism to adjust the parameters of the controller in order to avoid the inaccuracies which could arise during the flight mission. It is proved that utilizing combined fuzzy logic control theory and robust 'crisp' control to increase the robustness of control law design is tangible from the view point of theoretical background and practical one. Sometimes, the required expert knowledge to design a UAV fuzzy control law may be inaccurate or insecure. In order to provide UAV flight safety in these conditions, it is necessary to improve the learning capabilities of the control system. This could be achieved by application of neural networks to adjust the fuzzy control parameters during the flight mission. From the simulation results, it could be seen that the used methods show a promising results from the viewpoint of robustness and performance. Moreover, it is shown that the robustness of the designed control laws is preserved if structured and/or unstructured perturbations occur during the flight mission.