Using description logic for con guration problems

In order to use a model{based approach for describing connguration problems the following types of knowledge have to be represented: the components of which a connguration can consist. Generally the components are collected in a catalog. the allowed/necessary relationships between the components, that deene a valid set of components in a conngura-tion solution. a speciication of the connguration problem to solve. The connguration problem is solved, when a set of components is selected from the catalog in such a way, that the components satisfy the speciications and do not contradict any relationships. The main diiculty in representing con-guration problems lies not in modeling the components but in describing the relationships. In comparsion to other existing model{based connguration systems like PLAKON RC91], KONWERK G"u95a, G"u95b], or COSMOS Hei93], the term \re-lationship" is sweeping. Several kinds of relationships can be identiied in connguration problems: part{whole taxonomy: components are described as composition from other components functional dependencies: components can be considered as function providers. For providing these functions they need other functions provided by other components and so on. physical laws in a technical domain: such as U = R I or m = r f procedures in the mechanical engineering domain there are procedures verifying and/or calculating values like temperature in crankshaft ... Therefore, to describe components and their relationships a KR system needs a rich and expressive language. In our opinion, ter-minological systems provide such a language. On the inference level of solving conngura-tion problems, one important task is the selection of appropriate components from the catalog. Normally a component is not iden-tiied by its name but by its properties. So selection of components can be seen as a classical classiication task. The diiculty in the classiication task lies in the way how the properties of a component are determined. The relationships between the components impose certain requirements (i.e. their properties) and constraints on the components that have to be obeyed. Computing the relevant requirements for a component have to be taken into account during the classiication task. De-sciption logics provides a powerful mechanism for classiication. The main advantage is that procedures for evaluating the relationships can interact with the classiication mechanism. In our oppinion description logics are useful for solving connguration problems because they provide a rich representation language for modelling the components and their relationships and a powerful classi