Utilising Artificial Neural Networks for Assessing Seismic Demands of Buckling Restrained Braces Due to Pulse-like Motions

Abstract

Buckling restrained brace frames (BRBFs) exhibit exceptional lateral stiffness, load-bearing capacity, and energy dissipation properties, rendering them a highly promising choice for regions susceptible to seismic activity. The precise and expeditious prediction of seismic demands on BRBFs is a crucial and challenging task. In this paper, the potential of artificial neural networks (ANNs) to predict the seismic demands of BRBFs is explored. The study presents the characteristics and modelling of prototype BRBFs with different numbers of stories and material properties, utilising the OpenSees software (Version 2.5.0) for numerical simulations. The seismic performance of the BRBFs is evaluated using 91 near-fault pulse-like ground motions, and the maximum inter-storey drift ratio (MIDR) and global drift ratio (GDR) are recorded as a measure of seismic demand. ANNs are then trained to predict the MIDR and GDR of the selected prototypes. The model’s performance is assessed by analysing the residuals and error metrics and then comparing the trend of the results with the real dataset. Feature selection is utilised to decrease the complexity of the problem, with spectral acceleration at the fundamental period (T) of the structure (Sa), peak ground acceleration (PGA), peak ground velocity (PGV), and T being the primary factors impacting seismic demand estimation. The findings demonstrate the effectiveness of the proposed ANN approach in accurately predicting the seismic demands of BRBFs.