System that Learns to Design Cable-Stayed Bridges

Abstract

The critical design decisions in bridge design are made at the preliminary design stage. They depend on the expertise of the designer, built up from extensive experience. Experience is difficult to acquire, and may be entirely lacking when new technology is introduced. As a result, there is little shareable and transferable collective design knowledge within the profession. This paper explores how preliminary design knowledge may be generated, updated, and used, employing techniques of machine learning from the field of artificial intelligence. A model of the preliminary design process is first presented as a sequence of five tasks and then specialized to the design of cable-stayed bridges. A computer tool serving as a design support system is described, whose design follows the model of the preliminary design process, and a design example using the tool is presented. The key property of the system is its adaptive nature: it acquires knowledge from information on existing bridges as well as from designs generated with the system, thereby continuously improving its performance. Future enhancements to the tool breadth and depth are offered.