This thesis deals with the design and optimization of robust low-order/fixed-structure controllers. Thereby, two classes are considered, namely integer and fractional order controllers. Three approaches are proposed to tune the parameters of these controllers. Moreover, the obtained controllers are validated in simulation as well as in real environment. In what concerns fractional order controllers, a method is proposed to optimize the parameters of fractional controllers. Thereby, robustness specifications have to be achieved. These are expressed in terms of a desired phase margin, desired crossover frequency and robust performance in case of static gain variations. For this purpose, the H1 norm is used to formulate the control problem. The resulting optimization problem is solved iteratively using our proposed method. Moreover, a Matlab Toolbox has been developed dedicated to the controller optimization, namely Fractional Order FOPID Controller (FOPID)-Toolbox. The second method is dedicated to the optimization of robust PID controllers. Based on Model Predictive Control (MPC), the optimization problem is formulated. The resulting problem is transformed into a matrix inequality problem. Specifically, it is a Bilinear Matrix Inequality (BMI) problem. Using a proposed method, this problem is transformed into a Linear Matrix Inequality (LMI), which is solved in the controller parameters using well known LMI solvers. Simulation examples are presented to show the effectiveness of this approach. The third method is concerned with the optimization of linear controllers based on Recurrent Neural Networks (RNN). Using a novel procedure, the control problem can be formulated as a closed-loop RNN. The plant as well as the controller are structured as a recurrent network. The weights of the network which represent the plant are set fixed. The controller network weights are free parameters, which will be updated during the training stage. Thereby, the goal is to provide the robustness requirements given in terms of the closed-loop step response. Simulation results are also presented to show the effectiveness of this approach. To test the approaches proposed in this work, a test Parallel Hybrid Electrical Vehicle (PHEV) is considered. Based on measured data, a MIMO model is first identified. Then, this model is used to compute the robust controller parameters. After testing the two model-based methods in simulation, the neural network linear controller provides the best performance. This controller is implemented on the real plant using rapid prototyping methods. Experimental results show the achieved performance with this controller.    «

This thesis deals with the design and optimization of robust low-order/fixed-structure controllers. Thereby, two classes are considered, namely integer and fractional order controllers. Three approaches are proposed to tune the parameters of these controllers. Moreover, the obtained controllers are validated in simulation as well as in real environment. In what concerns fractional order controllers, a method is proposed to optimize the parameters of fractional controllers. Thereby, robustness s...    »