Development of Surrogate Models for Braces in Steel Concentrically Braced Frames

Metamodeling refers to the identification of the underlying behaviour of a physical phenomenon by leveraging available data using machine learning methods. In structural and seismic engineering applications, predicting the hysteresis behaviour of structural elements under earthquake loading, i.e., the physical phenomenon, is typically challenging due to the nonlinear material, geometrical response of the element, and the randomness of the earthquake excitation. With current advancements in data-driven modelling and their widespread use in other disciplines, there is ample opportunity to develop and adopt metamodels for seismic simulation of structural systems, which can bypass the daunting task of numerical simulation of nonlinear structural elements. In this research, a deep neural network-based metamodel is developed 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 Long Short-Term Memory (LSTM) algorithm is intended to predict the inelastic cyclic response of HSS braces, namely the brace axial force based on the input axial displacement signal.