Real-Time Expert Systems Computer Architecture

Abstract

Expert systems and real-time systems technology have been developed independently. Expert systems have been successfully implemented in many complex applications traditionally performed by human experts. Real-time systems have been successfully applied in areas requiring interaction with dynamic environments, control and monitoring applications for example. Merging these two technologies will yield intelligent systems capable of interacting with complex dynamic environments, an area in which human operators exhibit poor productivity, due to cognitive overload. The integration of real-time systems and expert systems impose new requirements on computer architectures. A real-time expert system must be able to process critical events in a timely fashion, handle uncertain data, degrade gracefully, and process knowledge efficiently. There are challenges in meeting the requirements for this new class of systems and they are addressed in the design of a computer system, specifically for real-time expert systems. A multiprocessor system is developed to meet the rigorous system requirements. The individual processors, called intelligent control units (ICUs), are organized in a logically reconfigurable topology to efficiently map onto the problem domain. The ICUs are special purpose processors designed to process rules in an extremely efficient fashion. The design encompasses features to support both the real-time environments and the knowledge processing associated with expert systems. The ICUs support dynamic priority processing to facilitate system responsiveness. Uncertainty processing is supported to represent fuzzy data and to perform heuristic search techniques. The system can perform direct environment queries during its knowledge processing and also directly communicate with other ICUs during the processing of a system goal. Time-stamps, temporal operators and a system time base provide the primatives for temporal and progressive reasoning. The processing of knowledge in the system is performed in a dataflow fashion. The data flow methodology allows the processing to proceed based on the availability of data. The real-time expert system architecture is simulated to validate its design and to evaluate its performance. Impressive simulation results confirm the approach taken in this research and encourage further development of this computer architecture.