Training and using neural networks to represent heuristic design knowledge英文电子书.pdf

Advances in Engineering Software 21 (1996) 117-128 Copyright 0 1996Civil-Comp Liited and Elsevier ScienceLimited Printed in Great Britain. All rights reserved SO965-9978(96)00017-8 0965~9978/96/$15.00 ELSEVIER Training and using neural networks to represent heuristic design knowledge J. D. Biedermann School of Engineering, University of Guelph, Ontario, Canada D. E. Grierson Department of Civil Engineering, University of Waterloo, Ontario, Canada The detailed stageof structural design has benefited considerably from computer automation of the numerically intensive tasks of structural analysis, optimization and conformance checking using the procedural programming approach. Such an approach, however, does not allow for the representation and utilization of heuristic knowledge implicit in previous design solutions, which is often difficult or impossible to represent algorithmically. This paper describes how artificial neural networks can be trained to learn heuristic knowledge from previous design solutions and how this kno