Design and evaluation of real-time fault-tolerant control systems.
Hagbae Kim · Deep Blue (University of Michigan) · 1994
A real-time control system is generally composed of a synergistic pair, a controlled process and a controller computer. In this dissertation, we derive the timing requirements from a controlled process and/or a controller computer which can be used for designing and evaluating a fault-tolerant controller computer. We derive the temporal inertia of a linear time-invariant control system, called the Control System Deadline (CSD), which formally specifies the need of the controlled process in a form understandable to the controller computer, using the dynamic equation of the controlled process, the information about failure occurrence rates and durations, and well-defined control algorithms. We propose a heuristic algorithm to iteratively compute the control system deadline, which represents system inertia/resilience against a dynamic failure, as a function of the system state and time by testing system stability or checking if the given (or derived) constraints of control inputs and states are met. We also evaluate the Fault-Tolerance Latency (FTL) of the controller computer, defined as the cumulative time taken by all sequential steps in recovering from a controller-computer failure while considering various fault-tolerance features. The FTL must not be larger than the Application Required Latency (ARL), which depends upon the controlled process under consideration and its operating environment. Consequently, we develop a cost-effective design strategy for a fault-tolerant controller computer by optimizing the tradeoff between time and spatial redundancy and by comparing the FTL against the CSD.