next up previous
Next: Modeling Distributed Diagnostic Up:

Deep Interoperation between Previous:

Deep Interoperation between


Introduction

Many efforts have been made to construct knowledge acquisition environments in order to build up expert systems successfully. Early knowledge acquisition environments, such as TEIRESIAS(R.Davis,76) and MORE(G.Kahn, S.Nowlan and J.McDermott ,85), tried to localize bugs in pre-constructed rule bases by meta-level rules, through the interaction of human experts. However, recent work moves to modeling domains and tasks (problem-solving methods) by knowledge, for example PROTEGE II(M.Munsen and S.Tu ,93), CommonKADS (J.Breuker and W.Van de Velde ,94) and EXPECT(W.R.Swartout and G.Yolanda ,96).

Furthermore, the work today is getting into the field of ontologies engineering, such as KIF(M.R.Genesereth and R.Fikes ,92) and Ontolingua(T.R. Gruber ,92). According to (G.Heijist ,95), there are several distinguished ontologies, such as general or generic ontologies for conceptualization across many domains, domain ontologies for conceptualization in specific domains and task ontologies for describing problem-solving methods. The proper use of explicit ontologies contributes greatly to achieving sharable and reusable knowledge bases. Although knowledge acquisition work has changed as explained above, knowledge acquisition environments continue to find good means of cooperation between a human expert and a machine.

On the other hand, as the work on multiagent systems has made progress and available information sources have increased rapidly through the Internet, the work on software agents (M.R.Genesereth, S.P.Ketchpcl ,94) has arose in SIMS(Y.Arens, C,Y,Chee, C-N Hsu and C.A.Knoblock ,93), Softbots(O.Etzioni, N.Lesh, R.Segal ,93), Knowledgeable Community (T. Nishida and H.Takeda ,93). Although finding good ways for cooperation among agents in the field of software agents and other multiagents systems is a key issue, practical coordination (including communication) tools are progressing in the field of software agents. However, software agents systems still take simple and small scale knowledge bases as information sources. So such coordination tools seem to work just at a syntactic level rather than a semantic level. Furthermore, as expert systems have been built up in many real fields over the past decade, the research on Cooperative Distributed Expert Systems (CDES) has emerged (C.Zhang ,92), (M.Zhang and C.Zhang ,94) and (T.Itoh, T.Watanabe and T.Yamaguchi ,95), integrating two kinds of technology from knowledge acquisition and software agents. The work in the field of CDES focuses on the cooperation among distributed expert systems but has not yet been getting into cooperation in real complex domains at a semantic level.

As seen in the fields of software agents and CDES, at present, multiagent and knowledge engineering technologies are becoming integrated. However, in order to develop robust cooperative knowledge systems in real and large scale industrial applications, the approaches from knowledge level analysis and knowledge modeling are few and so we are still in shallow interoperation just at a syntactic level among distributed heterogeneous expert systems. Thus, in this paper, as a first step, we try to present an environment for deep interoperation between two similar diagnostic expert systems at a semantic level, modeling them at a proper level of granularity of knowledge, using the difference arising in the context of the correspondence between inference primitives of an originator and those of a recipient and presenting a wrapper with conversion facilities using a common domain ontology.

In the remainder of this paper, we first describe methods of modeling, cooperating and communicating (wrapping) two similar diagnostic expert systems. Next, we put the methods together into an interoperative environment for them. Furthermore, in a deep interoperation experiment, it has been shown that the trouble-shooting expert system helps the enterprise diagnosis expert system get a refined inference structure and find some way to perform a given task better.



next up previous
Next: Modeling Distributed Diagnostic Up:

Deep Interoperation between Previous:

Deep Interoperation between




Daiki Kishimoto s0011
Sat Sep 28 20:16:55 JST 1996