Solution Approach

In order to achieve the intended functionality we focus on three crucial points of research interest. First, the requirements of the functional level and the knowledge contents of the local expert systems are specified. Second, a high-level problem-specific representation language for constituting the expertise of IMCOD is developed. Third, techniques that realize the design phase under the paradigm of comprehensive problem-solving with central management are expounded.

A local expert-in the form of an existing expert system-makes its contribution to the overall solution by assessing the current state from its isolated point of view. In order to enable IMCOD to incorporate the information in its model some requirements, concerning the represented domain knowledge and the problem-solving capabilities of the local experts, are specified [Bob91]. In consequence, the local experts are regarded as black boxes obeying some standards whose functional behavior can be controlled through a well-defined interface. Its produced output not only contains syntactic items but also meta-knowledge of how the solution was found and on which model it is based. The specification of these standards is an important research topic in the IMCOD project.

The expertise represented in IMCOD includes various facets:

• The system has sometimes a rough knowledge of all fields which are involved in the problem. Hence, it is only capable of understanding the contributions of the local experts partially.

• The system is aware of a model of decision analysis and of decision-making that is domain independent and only relies on functions that can assess a given situation [PH87].

• The system is able of applying the decision model to the actual problem based on the partial understanding of the domain and by assessing the worth of the information that he gains from each local expert.

Consequently, the system that realizes these facets is located anywhere in-between a decision support system, purely based on a model of decision theory, and an expert system with full understanding of the domain [Ric88]. A high-level, problem-specific representation language which closely integrates the various facets of IMCOD's expertise is developed.

In addition to the decision making functionality of IMCOD, regarding it as the central manager for the problem-solving phase, it serves as a

• translator. Each local expert system is supposed to have its own knowledge language. Incorporation of their output requires that IMCOD understands the language and can transform it into its own language under special consideration of a possibly diverging terminology.

• broker. Each local expert system has a specific problem-solving capability which it offers to the overall system. In order to get the requested information and to satisfy information requests of other expert systems IMCOD must be aware of the local experts' offer.

• scheduler. The design phase can be regarded as an iteration sequence of decision steps. Each step must be performed at the right time and therefore scheduled for the local expert systems.

Since these functionalities describe the behavior of IMCOD during the problem-solving phase it is in fact a run-time model of what was previously defined as the expertise of IMCOD. Techniques that realize these tasks are developed such that the overall behavior of the system reflects the design phase in the given domain.

contact: webmaster@dfki.uni-sb.de / last change: 16 Aug. 96