Model based diagnosis for technical systems

The principles of deep modeling are described. There are two approaches to deep modeling: component oriented description and process oriented description. Both describe a system of the real world hierarchically as composed of simple units on several levels. They differ with respect to the primary aspect for the modeling of a system. In the first approach, this is the structure of the system, in the second one, it is the behavior. In this paper we argue for a combination of both approaches. Diagnosis based on deep models proceeds in three steps: hypothesis generation, hypothesis checking, and hypothesis descrimination. In the first step, a number of hypotheses for the observed symptoms is derived by abduction. In the second step, some of the hypotheses are omitted using only informations at hand, i.e. the symptoms and the deep model, in the third step, further hypotheses are omitted by means of probing and testing. In all steps, extensive use is made of the informations contained in the deep model about the structure and behavior of the modeled system. (IITB)