LMI-based controller synthesis for fuzzy control systems and clustering-based sensor fusion for UXO detection

Traditionally, controller design for a typical nonlinear system or time-varying system is solved through nonlinear optimization. Usually, this nonlinear optimization procedure can be quite complicated. In this dissertation, we are focusing on the class of systems based on the so-called Takagi-Sugeno model. By focusing on this class of systems and applying various mathematical tools, the forementioned nonlinear optimization problem has been successfully converted into an LMI (linear matrix inequality) problem. This is a significant finding in the sense that there exist very efficient algorithms for determining the feasibility of LMIs, so even large scale design problems are computational tractable.; The results in our work cover various aspects of Takagi-Sugeno model, ranging from modeling to controller synthesis, including the following: It has been proven that Takagi-Sugeno model could be a universal approximator of any nonlinear control system. The notion of Dynamic Parallel Distributed Compensation is introduced and a set of LMI conditions is proposed that guarantees the closed-loop stability. Performance-oriented controller synthesis is also discussed and multi-objective controller can be designed based on it. Moreover, a fuzzy suboptimal controller is proposed to minimize certain criterion. Furthermore, the relationship between linear matrix inequality and algebraic riccati equation is investigated in detail. Some of the results are applied to the challenging task of controlling high-rise/high-speed elevators.; The second part of this dissertation describes our work on sensor fusion for UXO (Unexploded Ordnance) detection. This work has centered on the development and application of clustering-based fusion algorithms. Two main areas are covered in our research: development of clustering algorithm, and incorporation of model-based feature extraction algorithms. In the clustering algorithm, several algorithms have been developed or modified to be applicable to the UXO detection problem. We further refine these algorithms by incorporating statistical properties of targets. For the feature extraction, we resort to select more salient feature and find better ways to extract them. A model-based feature extraction is adopted and investigated in detail. It has been shown that our approach performs better than traditional detection algorithms by comparing results on several testing data sets. Furthermore, the benefits of sensor fusion are well demonstrated by the improved performance of fusion approaches over those utilizing only individual sensors.