Using neural networks in cold-formed steel design

An important advantage of cold-formed steel is the great flexibility of cro ss-sectional profiles and sizes available to the structural steel designer. However, this flexibility, in addition to the complex rules that govern co ld-formed member design, makes the selection of the most economical section for a particular application difficult, both in optimisation terms. and in terms of practical design. Selecting the best cold-formed solution require s numerous iterations involving analysis of several possible profiles and a spect ratios. This process becomes prohibitively expensive due to the amoun t of computer time required for convergence to an optimum design. This pape r investigates the potential for using neural networks to overcome these de sign problems. By carefully training a neural network with data relating se ction profile, aspect ratio and size to the load carrying capacity, the neu ral network can be prepared to estimate the best section requirements in ne w applications. The paper describes how this process has been carried out f or a limited range of section profiles and presents an assessment of the re sults obtained. (C) 2001 Elsevier Science Ltd. All rights reserved.