20178

SELF-ORGANIZING BACK PROPAGATION NETWORKS FOR PREDICTING THE MOMENT-ROTATION BEHAVIOR OF BOLTED CONNECTIONS

629

645

17

Evaluating the moment-rotation behavior of bolted connections by finite element method is very intensive task in terms of computational cost. This study proposes an efficient neural system to predict the moment-rotation behavior of the connections. ·The neural system is called self-organizing back propagation (SOBP) networks. The SOBP includes two processing units: classification and approximation. In the classification unit, all the training data are divided into some classes by a self-organizing map network. In the approximation unit, a set of back propagation networks are employed to achieve the approximation task. The numerical results demonstrate the computational advantages of the SOBP.

1204207