doi:10.3850/978-981-08-5118-7_026

ABSTRACT

In this paper, an approach is introduced which permits a model-free identification and prediction of time-dependent structural behavior. The numerical approach is based on recurrent neural networks for uncertain data. Time-dependent results obtained from measurements or numerical analysis are used to identify the uncertain long-term behavior of engineering structures. Thereby, the uncertainty of time-dependent loadings, environmental influences, and structural responses is modelled by means of fuzzy processes. The identification of uncertain dependencies between structural action processes and structural response processes is realized with recurrent neural networks for fuzzy data. Algorithms for network training and prediction are presented. The new recurrent neural network approach for fuzzy data is verified by a fuzzy fractional rheological material model. The approach is applied to predict the long-term behavior of a textile strengthened reinforced concrete structure.

Keywords: Time-dependent structural behavior, Model-free prediction, Fuzzy process, Recurrent neural network, Textile reinforced concrete.

© 2010 Professional Activities Centre, National University of Singapore