| Evolution-based optimization of fuzzy logic control systems has, in recent years, received a great deal of attention due to the inherent ability of Genetic Algorithms (GA) in efficient and parallel search of complex and multimodal landscapes. This need is vividly realized as knowledge-based fuzzy controllers are applied to more complex systems. A given system is recognized as a complex systems when its mathematical representation is either nonexistent or involves a large number of parameters, nonlinearities, and uncertainties such that classical control techniques cannot easily control the system. In this dissertation, the design of fuzzy controllers for complex systems and optimization of fuzzy knowledge bases is addressed through simulations and experimental investigation. Three significant issues are addressed. The first issue is the fuzzy controller's structure and design. Several new control architectures for distributed and lumped parameter systems are investigated and applied to a flexible robot and a water desalination process. A new bi-level hierarchical structure is proposed, for the flexible robot, which exploits the distributed nature of distributed parameter systems. The second issue is application of GA to optimization of fuzzy control systems. A new method is illustrated which incorporates existing a priori knowledge in developing a highly fit initial population. This method allows such a priori knowledge, commonly available through human intuition and operator experience, to be exploited to significantly increase GA's convergence rate. The third issue is the experimental evaluation of the fuzzy controllers. This issue is divided in two major categories: software and hardware. SoftLabTM is a user friendly and graphical software environment for Dynamic Fuzzy which can evaluate fuzzy rule sets and allow the real-time modification of the controller's knowledge base. The hardware consists of a flexible beam, a brushless DC motor, a TMS320C30 based digital signal processor board, a 486-DX2 personal computer, and a custom-designed data acquisition circuitry. The simulation results indicate that GA is a suitable candidate for optimization of a wide array of fuzzy systems. Simulation results also demonstrate that the proposed method of incorporating a priori expert knowledge significantly improves the convergence time of GA. This is a valueable technique where computational efficiency is of prime significance such as in real-time systems. Furthermore, the experimental results indicate the success of the proposed hierarchical fuzzy control architecture in controlling dynamics of a flexible beam. |
