Managing large scale knowledge bases

The development of large-scale expert systems (ES) through traditional methods will in most cases present performance problems. The single large knowledge base created with off-the-shelf development tools is not the ideal structure for expert systems based primarily on theoretical knowledge. The research results presented illustrate a segregated knowledge-base structure designed to increase performance in large-scale, highly nonempirical knowledge bases. The case study presented is for a maintenance advisor (MA) for US Navy sonar systems. The goal is to demonstrate the applicability of ES techniques to assist sonar technicians in the isolation of malfunctions to the component level. The candidate MA is selected using the traditional ES implementation guidelines only. The candidate domain area is bounded, static and of sufficient technical challenge to be an asset to sonar maintenance technicians. The methodology for handling the knowledge-base substitution is discussed