Ontologies have been proposed as a specification mechanism to enhance knowledge sharing and reuse across different applications [Neches et al., 1991]. To do so, they must capture the intended meaning of concepts and statements in a domain. Sharing and reuse imply two additional requirements: (i) ontologies must aim at a maximum level of genericity and thus bring out the commonalities within extensive bodies of detailed and specialized knowledge, and (ii) they must be able to explicate tacit and meta-level knowledge, as significant parts of domain expertise are highly implicit and have a background nature.
These aspects are all clearly present in the area we consider in this article: intelligent support for physical systems engineering. Take as a simple example the expression F=ma. Many people will immediately associate this with Newton's law stating that force is the product of mass and acceleration. But this is a highly non-trivial association, because it can only be made by invoking a lot of background knowledge. First, we have to know that we are dealing here with a mathematical expression, and we have to understand the related concepts of equations, parameters and variables. However, this is far from enough: the intended meaning of F=ma may now still be that electrical voltage is the product of resistance and current (which, instead, many people would call Ohm's law but typically write as V=IR). To distinguish between such possible interpretations we need more knowledge about, for example, the concept of physical dimensions of variables. To capture the intended meaning of F=ma in the context of its use in problem solving, we have to additionally invoke a significant body of expert knowledge. For example, we have to understand that in this context of problem-solving use, physical objects are abstracted to and parametrized in terms of a concept called `mass', that this mass acts as a kind of storage place for movement, that this movement does not change when the mass is undisturbed, and that it does change under certain external influences and circumstances which are abstracted to and parametrized in terms of a concept called `force', and so on. That is, in specifying intended meaning we unavoidably have jumped into a background body of specialist knowledge known as classical mechanics.
If ontologies are to enhance knowledge sharing and reuse by capturing intended meaning, the above-mentioned issues have to be confronted. Current information systems supporting complex tasks and domains typically do not possess the body of knowledge necessary for generating adequate interpretations, but instead rely on the fact that the user does. So, they place most of the burden on the user. Intelligent support implies that this burden must be shifted back as much as possible towards the information system. Ontologies are a promising candidate to help achieve this, but to realize this potential we need a better understanding both of their role in complex problem solving and of their construction.
The present work investigates this subject, illustrated by various aspects of physical systems engineering. The paper contains two main lines. Section 2 gives an overview of a general collection of ontologies for physical systems, whereby we attempt to clarify throughout how we can achieve genericity in ontological specifications, what general decomposition and structuring principles play a role, and how we can reuse existing outside ontologies. This ontology work has been extensively applied to the development of a reusable model library for mechatronic design, carried out mainly in the context of the Esprit project OLMECO. In this KAW'96 contribution, which is a shortened version of an article to appear in the International Journal of Human-Computer Studies, these application-specific parts have been left out, however. For these, the reader is referred to the full journal article, as well as to our contribution to KAW'95 (see Akkermans et al., 1995).
We believe that many of our experiences and results are independent of the considered domain, and have a general relevance for the engineering of ontologies. In Sections 3 and 4 we take up the general discussion and discuss the conclusions emerging from the present work in more detail, and compare them with related ongoing work.