Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 27
ARTIFICIAL INTELLIGENCE AND OBJECT ORIENTED APPROACHES TO STRUCTURAL ENGINEERING
Edited by: B.H.V. Topping and M. Papadrakakis
Paper II.1

Renumbering Heuristics using Neural Networks

M. Hoit and C. Olsen

Department of Civil Engineering, University of Florida, Gainesville, United States of America

Full Bibliographic Reference for this paper
M. Hoit, C. Olsen, "Renumbering Heuristics using Neural Networks", in B.H.V. Topping, M. Papadrakakis, (Editors), "Artificial Intelligence and Object Oriented Approaches to Structural Engineering", Civil-Comp Press, Edinburgh, UK, pp 69-81, 1994. doi:10.4203/ccp.27.2.1
Abstract
Throughout history, engineers have struggled to devise new and efficient techniques for analyzing structures. Today, the finite element analysis method is by far the most popular and reliable technique. However, this approach does have its limitations. Finite element analysis mandates the use of a computer. Consequently, engineers have encountered problems with lack of storage space and large solution times. One of the most successful solutions to this problem is the use of a renumbering algorithm.

Renumbering algorithms attempt to reduce the size of the matrices used in calculations by changing the ordering of the nodes of a structure. One algorithm, which concentrates on reducing both the profile and wavefront characteristics of a matrix, is profile front minimization-PFM. Since its introduction in 1983, PFM has undergone several changes. PFM uses a set of hueristic weights for determining the optimal nodal ordering. In 1993, the algorithm was modified and parametric studies of these weights were conducted on twenty-nine separate structure files to determine the overall improvement. These studies concluded that there is a significant improvement in the algorithm's performance when the "correct" weights are used.

The parametric studies provided a different heuristic weighting set for each of the given structures. These results had to be generalized to incorporate all twenty-nine structures and allow for the extrapolation to new structures. In order to do this, the results of the individual studies needed to be combined to produce a general prediction method for the heuristic weights based on a structures topology. A neural network with back propagation was used to incorporate the set of optimal weights into a general method for the prediction of weights for any structure. The results of the parametric study along with the use of the neural network is the focus of this paper.

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

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £45 +P&P)