Artificial Neural Network-Based Hysteresis Model for Steel Braces in Concentrically Braced Frames

This paper presents an artificial neural network-based metamodel for extracting the inelastic cyclic response of steel hollow structural section braces that are part of concentrically braced frames when subjected to earthquake ground motions. The proposed model constructed using the bidirectional long short-term memory (BiLSTM) algorithm is intended to predict the inelastic cyclic response of steel braces, namely the brace axial force, based on the input axial displacement and out-of-plane displacement signals. The architecture of the proposed metamodel is first described. The validity of the model is then tested when the brace is subjected to a set of displacement signals obtained from numerical nonlinear time-history analyses. The preliminary results confirm that the proposed model has the potential to predict the complex hysteresis response of steel braces, involving tensile yielding and compressive buckling.