A General Characterization of Model-Based Diagnosis

The Model-Based Diagnosis (MBD) framework developed by Reiter has been a strong theoretical foundation for MBD, yet is limited to models that are described in terms of logical sentences. We propose a more general framework that covers a wide range of modelling languages, ranging from AI-based languages (e.g., logic and Bayesian networks) to FDI-based languages (e.g., linear Gaussian models). We show that a graph-theoretic basis for decomposable system models can be augmented with several languages and corresponding inference algorithms based on valuation algebras.