Some important problems of neural network applications in civil and structural engineering (C&SE) are presented. Data preprocessing associated with training and testing patterns in the backpropagation networks (BPNs) are considered. Application of dimensional analysis to formulate data representations and transformation of experimental data from time to frequency spaces are discussed as examples of data preprocessing. Multilevel BPNs are described as a neural classificator for parameter identification of steel semi-rigid connections. Improvement of efficiency of BPNs is discussed with respect to application of Radial Basis Functions. Resilient Backpropagation is considered as an adaptive method for faster training of BPNs. Problems corresponding to improvements of Counterpropagation networks are shortly discussed. Great potential possibilities of Hopfield continuous networks are pointed out with respect to the analysis of bilateral and unilateral problems of mechanics. Construction of hybrid neural/numerical codes is shown on an example of elastoplastic beams bending. In the code based on finite differences, BP networks are used as subroutines to invert the constitutive equation of elastoplastic material. Applications of neural networks in expert systems, neuro-fuzzy systems and ART networks are listed as other problems, not discussed in the 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 £82 +P&P)