| Stimulated by the growing demand for improving system performance and reliability, fault tolerant system design has been receiving significant attention. This dissertation investigates the development of fault tolerant control methodologies using adaptive estimation and control approaches based on the on-line learning capabilities of neural networks or fuzzy systems. The objective is to detect the changes of system conditions and the occurrence of faults, and to “reconfigure” (adjust) a controller so as to retain satisfactory system performance in the presence of a variety of system uncertainties such as disturbances, faults, and time-varying dynamics.; To achieve the above goal a hierarchical learning structure is first developed with neural/fuzzy and immunity-based hybrid learning methods for structure and parameter adjustment. These nonlinear system modeling methods serve as a foundation for fault tolerant system design by characterizing the dynamics of the target nonlinear system. Next, on-line approximation based stable adaptive neural/fuzzy control is studied for a class of input-output feedback linearizable time-varying nonlinear systems. This class of systems is large enough so that it is not only of theoretical interest but also of practical applicability. Both indirect and direct adaptive control schemes have been studied and the results have also been extended to a class of interpolated nonlinear systems so as to deal with fast time-varying dynamics by exploiting the inherent system structure. Although the adaptive controller developed with these approaches is capable of serving as a fault tolerant controller, its performance can be further improved by exploiting information estimated from a fault diagnosis unit. This model-based fault estimation system is designed by interfacing multiple models with an expert supervisory scheme and then incorporating this into an adaptive controller to achieve integrated fault tolerant control.; The effectiveness of the proposed integrated fault tolerant control methodologies has been demonstrated by applying the approaches to the component level model simulation of the General Electric XTE46 jet engine. Furthermore, theoretical analysis has been given to study system performance including robustness, fault sensitivity, stability, and convergence. |
