Acquisition of causal knowledge in the CHECK system

In the last few years, approaches have been proposed for the design of diagnostic systems. In particular, 'deep' models and multi-level systems have emerged as the most interesting solutions. However, the critical point in the design of diagnostic architectures still remains to be the knowledge acquisition problem. This paper presents a knowledge acquisition subsystem designed to acquire knowledge in the CHECK system, a two-level diagnostic architecture combining heuristic and causal knowledge (reasoning) developed over the past few years. After a brief sketch of CHECK's architecture, the paper presents NEED, a graphical tool developed to acquire causal knowledge. The adoption of such a tool makes it possible to simplify greatly the interaction with the domain expert: the tool is very natural to use and performs on-line syntactic and semantic checks on the acquired knowledge base which help the user in the design and maintenance of the knowledge base itself.