Keywords: neural networks, error tolerance, functional mapping, network topology, optimisation, network training, pattern association, pattern classification.

Artificial neural networks (ANNs) are often described as computing devices (or software simulations thereof) that loosely model the operation of the central nervous system. In the biological sciences, the objective of neural network modelling is frequently an attempt to learn more about the biological processes of the central nervous system, and thus may tend to model the physical rather than functional behaviour of the brain. In Engineering, on the other hand, the objective is more usually to emulate or capture the information processing characteristics of the brain that have been found to elude conventional electronic digital computing techniques - characteristics such as an ability to: learn and infer from experience; recognize patterns in information even when that information is presented from a different perspective; and maintain a graceful degradation in performance (as opposed to a catastrophic failure) with an increase in error or noise in the input information.

Examples of the application of ANNs to civil and structural engineering goes back to at least 1989 [1]. In the ensuing period, many successful applications have been developed, from transportation engineering [2] to water resources [3] to construction economics [4]. This chapter provides an introduction to the diverse range of alternative artificial neural networks (ANNs) currently available and the types of application they have been adopted for in civil and structural engineering.

The presentation is made with reference to a classification and decomposition of the main features of ANNs. An introduction is first provided of the essential features of a typical ANN, and its mode of operation. A brief graphical interpretation is presented to illustrate how ANNs model data, and to help gain insight to their scope of application, merits and drawbacks. An identification is then made of the different types of processing that can be performed at the neuron level, the lowest level in an ANN. This is followed by an identification of the different ways in which neurons can be combined into an integral processing device capable of solving non-trivial problems. The range of alternative function types that can be implemented by ANNs are then examined.

Finally, a review is provided of the primary methods of developing/training ANNs to solve specific problems. At each stage in the chapter, relevant neural paradigms are referenced and areas of application in civil and structural engineering are identified.

1

I. Flood, "A neural network approach to the sequencing of construction tasks", Proc. 6th Int. Symp. on Automation and Robotics in Construction, Construction Industry Institute, Austin, Texas, 204-211, 1989.

2

R.L. Cheu, and S.G. Ritchie, "Automated detection of lane blocking freeway incidents using artificial neural networks", Transportation Research C, 3(6), 371-388, 1995. doi:10.1016/0968-090X(95)00016-C

3

N. Karunanithi, W.J. Grenney, D. Whitley, and K. Bovee, "Neural Networks for River Flow Prediction", Journal of Computing in Civil Engineering, 8(2), ASCE, New York, NY, 201-220, 1994. doi:10.1061/(ASCE)0887-3801(1994)8:2(201)

4

S. Karshenas, and F. Xin, "Application of Neural Networks in Earthmoving Equipment Production Estimating", Proceedings of the 8th Conference in Computing in Civil Engineering and Geographic Information Systems, ASCE, New York, NY, 841-847, 1992.

purchase the full-text of this chapter (price £20)

go to the previous chapter
go to the next chapter
return to the table of contents
return to the book description
purchase this book (price £92 +P&P)