Temporally distributed symptoms in a diagnostic application

Abstract

For various reasons many diagnostic expert systems do not explicitly represent time. In a number of cases this is a useful abstraction. In other applications — such as MOLTKE's —, however, some symptoms evolve over time in a characteristic manner and require several measurements which have to be carried out in a particular temporal order. In this paper we show how such symptoms can be represented declaratively in the knowledge base and how a general temporal matching algorithm can be smoothly integrated with MOLTKE's otherwise static framework. We emphasize that the extension preserves the general philosophy of the system and does not lead to unacceptable overhead when time-independent symptoms are processed.