A back-propagation based neural network approach for modelling semi-rigid steel structure connections

Modelling of semi–rigid steel structure connections by appropriate processing of experimental data through back–propagation based neural network systems is considered in this paper. Single back–propagation networks can learn and subsequently reproduce experimentally measured momentrotation curves, after appropriate pre-and post–processing of the data. Multilevel mixtures of such networks can be used as classificator systems in order to extract from the experimental curves best fitted multi–linear laws, from a predefined family of simplified laws, which are used in the sequel for structural analysis tasks. Both approaches are tested on single angle beam–to–column, bolted connections.