Process-oriented Modeling and Diagnosis - Revising and Extending the Theory of Diagnosis from First Principles

Our work in model-based analysis, diagnosis, and therapy of environmental systems reveals principled limitations in standard theories and techniques of model-based diagnosis. It becomes evident that they are (implicitly) tailored for diagnosing artifacts based on component-oriented modeling. When we deal with a natural or technical system that lacks a static structure comprising a fixed set of components and that we prefer to model as a collection of processes, a number of conditions and goals of the diagnosis process change. Processes do not become faulty like components, and often, the goal is not to find the culprit among the known components of a system, but to identify additional objects that were not part of the initial system description, for instance toxic waste. We propose a revision of the traditional theories of diagnosis from first principles. The goal is to make it more general in terms of the class of problems to be addressed and more specific by proposing and exploiting a refined representation of the system description. We also show how standard component-oriented modeling and diagnosis can be reconstructed as a (very) specific instance of the theory and that they benefit from the more general view on the diagnosis problem.