Abstract
The purpose of this study is aimed at improvement of system reliability for the Neuro-Fuzzy Bridge Rating Expert System, by investigating representative problems related to knowledge-base refinement. In this paper, the previous empirical knowledge (teaching data) acquisition method is reviewed, and a reliable method which can extract the intrinsic fuzziness from questionnaire survey results, is proposed. And, in order to represent experts' inference flow for user, a user information system is newly developed, and installs into the expert system. Furthermore, by applying the improved expert system to an actual in-service bridge rating, it is verified that the proposed teaching data acquisition method and the user information system are effective.
