Computer Architectures for Real-Time Knowledge-Based Control

Abstract

There has been an increasing interest recently in real-time implementations of knowledge-based control systems. Such systems typically consist of a real-time control system that implements a control algorithm and a supervisory knowledge-based system that monitors the inputs and outputs of the control system to detect any degradation in its performance. If the performance is found to be deteriorating the knowledge-based system attempts to reconfigure the control in order to restore its performance or to maintain safe operation. Implementation of the control algorithm usually involves numerical computing while reasoning performed by the knowledge-based system involves symbolic computing. Real-time implementations of knowledge-based control have been limited by the intensive computational needs required by traditional symbolic computing systems. In this paper we review a number of solutions to the problems of implementing knowledge based contol that use special computer architectures. These approaches include the use of dual-processing; separating symbolic computing from numerical computing. This includes using two different computers; such as a Lisp machine and a Vax, using a Lisp machine with a dual 68020 processor board for numerical computation, and by using an IBM PC-AT with a dual digital signal processing board (Texas Instrument's TMS32010). The paper also examines a number of proposed solutions based on distributed computer architectures for control tasks and/or for the knowledge-based reasoning task. Reported performance and limitations of each approach to implementation and possible future directions for will be discussed.