Redundant Structures for Fault-Tolerant Control

摘要

An increasing national priority on quality in product design and manufacturing requires new understanding to achieve significant advancement. Fault-tolerant control, a discipline capable of high-level decision making and task execution, is a necessary component for ensuring system reliability in the hierarchy of intelligent control systems. In contrast with current research, redundant control structures provide real-time fault tolerance and error accountability for systems in an untended manufacturing environment without the use of a process model. Fault detection and isolation (FDI) is optimized with respect to a risk or cost function equivalent to the probability of decision error and is generalized to account for both positive and negative faults within any controller. The resultant test compares a significant statistic to a derived threshold which is adjusted over the mission to reflect any change in the reliability of the control structure. The performance of the FDI scheme is found to be proportional to the failure signal-to-noise ratio. The effect of multiple faults on the probability of decision error is found to be negligible, assuming an uniform fault distribution. Analysis of these redundant structures and their associated FDI and reconfiguration schemes emphasizes a probabilistic set of system states which represents all a priori uncertainty inherent within the control system. Information theory defines entropy as a logarithmic measure of system/decision uncertainty. This allows for a comparison of the effective system performance of redundant structures. The optimal redundant structure for fault-tolerance is reached by utilizing a highly reliable control structure at the greatest level of redundancy while maintaining near-perfect FDI at all levels of operation. This allows maximizing the information rate of the discrete FDI decision scheme while minimizing the error variance of the controlled parameter.