Functional Causal Graphs. Yet Another Model for Diagnostic Reasoning

Abstract

The paper addresses the issues of models for diagnostic reasoning based on abductive analysis of causal structures. A basic, core, uniform model for representing causal behaviour of diagnosed systems is proposed; it has the form of a Functional Causal Graph allowing for specification of causality types reflecting arbitrarily selected functions. The causal structure specifies a model defining the search space for diagnostic inference. The diagnoses are found by backward search of the graph. Diagnostic reasoning incorporates failure detection, problem statement by selection of manifestations to be explained, and abductive search for possible explanations which are consistent with observations. The search can interleave with state propagation and consistency verification. The proposed model supports various extensions and modifications, including a multistage approach to diagnostic reasoning, methods for search ordering, using heuristics, tests and probing, consistency constraints and initial observations support. Moreover, the proposed model seems to generalize several approaches to diagnostic process, such as fault isolation based on diagnostic relation, set covering model and simple AND/OR/NOT causal graphs.