Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON THE APPLICATION OF ARTIFICIAL INTELLIGENCE TO CIVIL AND STRUCTURAL ENGINEERING
Edited by: B.H.V. Topping and B. Kumar
Using Neural Networks to Design Cold-Formed Steel Sections
R.I. Mackie, E.M.A. El-Kassas and A.I. El-Sheikh
Department of Civil Engineering, University of Dundee, United Kingdom
R.I. Mackie, E.M.A. El-Kassas, A.I. El-Sheikh, "Using Neural Networks to Design Cold-Formed Steel Sections", in B.H.V. Topping, B. Kumar, (Editors), "Proceedings of the Sixth International Conference on the Application of Artificial Intelligence to Civil and Structural Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 18, 2001. doi:10.4203/ccp.74.18
Keywords: neural network, cold-formed steel, structural design.
Structural steel is used in two main forms: hot rolled and cold-formed sections. Hot rolled sections are by far the most widely used, and are the stronger of the two. However, there are many advantages to cold-formed steel. It is much lighter, more flexible in use, and more environmentally friendly. Cold-formed steel is used extensively in buildings, and its use in frames and trusses is increasing. There is now a growing interest in their use as primary structural members, especially in small span industrial and agricultural building and steel housing systems. This interest has led to use of cold-formed steel remaining buoyant even during the 1990's recession. Currently about 400 000 tonnes of cold-formed steel are used annually in the UK, but there is scope for much greater use. One of the major obstacles to the greater use of cold-formed steel is the flexibility of the ways in which it can be used. For instance, there is an almost limitless range of section profiles that can be used, each offering different advantages. Furthermore, failure is more complex with local buckling often being an important factor. All this makes design with cold-formed steel significantly more complex than design with hot-rolled steel. Most design programmes do not offer any cold-formed sections, mainly because they have been rarely used. The problem is how to identify the "best section", and that the analysis process is far from simple. The work described here describes some of the preliminary work which is part of a project that aims to use neural network technology to overcome many of the difficulties of designing with cold-formed steel.
Neural networks have received increasing attention in the last fifteen years or so. They form one part of the artificial intelligence spectrum, but in many ways can be viewed as pattern recognition systems, or as an extremely powerful and versatile multi-dimensional surface fitting tool. There are many variations of neural networks, but the most common is the multi-layer perceptron network. This consists of a layer of nodes representing the inputs, one or more hidden layers of nodes, and a final layer of nodes representing the outputs from the system.
Neural networks are effective when there is a relationship between the inputs and outputs, but there is no simply delineated rule or set of equations for expressing this relationship. This may be because the relationship is too complex, or some of the inputs or outputs are not easily quantifiable, e.g. ease of construction. This paper reports preliminary investigations into the feasibility of using neural networks in cold-formed steel design. The aim is to use networks to choose the best design given a set of design criteria. This involves choosing the profile type, and the dimensions of the given profile. The work reported here follows on from earlier work by the authors into using neural networks to predict the design load of a given section. The work presented here was restricted to three profile types, hat, lip and plain sections. The design criteria were limited to minimum weight. The input parameters consisted of the wall thickness, design load, and required length. Investigations were also carried out into the size and type of network needed. First of all dedicated networks were created for the three profile types. The networks for the lip and hat sections gave very accurate results. The training set available for the plain section was limited, and this resulted in a less accurate network.
Two strategies were used to create a network to predict both profile type and associated dimensions. The first approach used a single network to choose the profile type and dimensions. The second used a multi-network approach. The first network chose the profile type, and then the corresponding dedicated network was used to choose the dimensions. This second approach was found to be much more successful.
It was concluded that
purchase the full-text of this paper (price £20)