Time-constrained reasoning and consistency maintenance for real-time knowledge-based systems

Large numbers of knowledge-based systems (KBSs) have been developed that take their input directly from plant sensors rather than via human operators. Although such systems have often been termed real-time KBSs, they have had few architectural features specifically designed for time-critical operation or for dealing with dynamic environments. These issues have been side-stepped by selecting plants with sufficiently long time constants such that reasoning about the plant is not time critical and by selecting plant or environments that do not present KBSs with unexpected events. If the range of KBS applications is to be broadened to encompass truly real-time problems then the following issues must be systematically addressed: time constrained reasoning; and dealing with a dynamic world. The author describes why one of the major features of KBSs, the separation of control information and domain knowledge, poses particular difficulties when the above issues are considered. The trade-off of solution quality against time and use of consistency maintenance mechanisms as ways of dealing with these problems, are also discussed