الفهرس Only 14 pages are availabe for public view
 |  | from | 101 |  |  |
| | | from | 101 | | |
Abstract
The problem of designing a controller for nonlinear MIMO systems is crucial for many applications in the industry. Researchers have targeted this problem through different control schemes and have achieved important results toward this subject. During the last decade, proposed control algorithms based on learning has proved efficiency in targeting the problem especially in the presence of uncertainties. Fuzzy logic and neural networks has played a pivotal role in most of these learning based control schemes.
The recent different methods generally suffers either from the complexity due to the input/output coupling effect, the complexity induced through heavy optimization techniques, inflexibility in dealing with nonlinearities or uncertainties, or from the lack in accurately computing the control law necessary to produce satisfactory results while imposing a time limit. These problems were a major factor between the different methods but at different degrees. Also, problem specific solutions were introduced to satisfy industry needs.
The thesis presents a method that inherits from fuzzy logic ideas but with a new geometrical perspective. Fuzzy neural networks have been largely used as a form of integrating the fuzzy inference model in the neural structure. The Delaunay triangulation structure is proposed to be used as a function approximation tool. The idea is ’based on limiting the complexity of the control algoritl.un through limiting the point set used for inference during any single time step. The flexibility to query the Delaunay triangulation is used to collect the point set with relatively low overhead. Also, the method alleviates the need to hold different structures for each output through using a single Delaunay triangulation N-tree structure. This structure has been used as the base for the proposed reference model. A new model reference adaptive controller is then presented based on the reference model described.
Several numerical simulations have been conducted to analyze the ability of the proposed method to track or control SISO systems and MIMO systems. A simulation program was developed using C++ for this purpose. The simulations proved the efficiency of the proposed controller. Also, the Delaunay triangulation size growth and system performance were investigated with the different sampling periods and the different error limits imposed