Knowledge Acquisition and Engineering for Steel Bridge Fabrication

Abstract

A knowledge-based system for resolving fabrication errors in steel highway bridges is in operation at the Kansas Department of Transportation (KDOT). The Bridge Fabrication Error Solution Expert System (BFX) addresses errors encountered with relative frequency, but not remedied with standard actions prescribed by specifications or KDOT guidelines. Example cases were gathered directly from fabrication shops, state inspectors' field notes, and bridge-project documents. Panel meetings and interviews were conducted with multiple-domain experts. These cases formed the basis for building the knowledge representation space and testing of the system. An attempt was made to use an ID3-based learning tool for knowledge acquisition in selected categories, but because of the small number of examples available, this investigation was not very successful. Therefore, knowledge acquisition was primarily based on explicit domain extraction. Test results showed that BFX provided correct repair solutions two-thirds of the time while correctly identifying that no solution could be properly generated using the existing knowledge base for the remaining one-third of the test cases.