Model-Based Knowledge Acquisition from Structure Descriptions in a Technical Diagnosis Domain

Abstract

To be able to quickly build expert systems for technical diagnosis causal knowledge is needed. In this paper we propose a specific way of integration between "shallow" (heuristic) knowledge and "deep" (causal) knowledge (i.e. a model of the device under investigation), namely the translation (compilation) from the model into causal rules. We present a way to separate the behavior acquisition of the typical components of a device (given by some domain experts) from the construction of a concrete representation (model) which can be built without experts. This model together with domain specific "technical common sense" integrated into our system is used to construct rules for a rule-based expert system that represent the causal knowledge . Our system is also able to modularize the resulting knowledge base into contexts (rule sets) similar to those given by experts themselves. These techniques are used in MOLTKE, an expert system for the diagnosis of CNC machining centers at the University of Kaiserslautern.