ECAI 2004 Conference Paper

[full paper]

Models of Behavior Deviations in Model-based Systems

Peter Struss

Tasks like diagnosis, failure-modes-and-effects analysis (FMEA), and therapy proposal involve reasoning about deviating variables and parameters. In model-based systems, one tries to capture this kind of inferences by models that describe how deviations from some reference state or behavior are created and propagated through a system. Several techniques and systems have been developed that address this issue, in particular in the area of qualitative modeling. In the context of model-based systems, we face the requirement that the system model has to be compositional, i.e. can be generated by aggregating local models of the system constituents (e.g. components) that are stored in a library. However, to our knowledge, a rigorous mathematical foundation and a “recipe” for how to construct such compositional deviation models has not been presented in the literature, despite the widespread use of the idea and the techniques. In this paper, we briefly revisit the concepts and use of deviation models and related techniques. We reveal (sometimes implicit) presumptions and limitations and present a general mathematical formalization of deviation models. Based on this, aspects of constructing libraries of deviation models, their properties, and their application in consistency-based diagnosis and prediction-based FMEA in a component-oriented framework are analyzed.

Keywords: Model-based systems, diagnosis, qualitative modeling, failure-mode-and-effects analysis, deviation models

Citation: Peter Struss: Models of Behavior Deviations in Model-based Systems. In R.López de Mántaras and L.Saitta (eds.): ECAI2004, Proceedings of the 16th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2004, pp.883-887.