Application of Neural networks for the selection of repair methods for steel bridge fatigue damages.




Constraints of money, manpower or equipment make it impossible for public agencies to deal efficiently with all bridges requiring improvement concurrently. Steel bridge assessment and selection of the appropriate repair and retrofit methods require extended in-situ inspections, testing and analysis, which depend on the expertise of engineers and on empirical knowledge collected over several years.

The present research investigates the applicability of the artificial neural networks, in order to classify steel bridges damages and select the appropriate repair method of the structure. At first, a database of cases histories of railroad bridge damages and relevant repair methods is collected. The information is translated to input variables and preprocessed. Characteristic attributes of the dataset include damage location, condition and cause, while the output variable is the corresponding retrofit method. 

A wide range of neural networks architectures is considered, namely SOFM, LVQ2, ART, HAVNET and multiplayer networks. The success or the difficulty with which they classify the input variables is investigated. This research can find practical applications in the field of bridge management, as an attractive alternative in the decision-making process, cost estimation and assessment of unusual bridge damage conditions.




Last updated: 23 September, 2001 07:11:48 PM
Please refer all questions about the site to
Copyright © 2001 UMR's SESL. All Rights reserved.                                 SESL Home