Machine-Learning Based Prediction Model for Identifying Torsion-Induced Seismic Response Amplification in Plan-Asymmetric Buildings

Torsion-induced seismic response amplification in plan-asymmetric buildings is of major concern in structural design. Code-based seismic design procedures based on elastic analyses do not address potential seismic risks that are aggravated by torsional actions. Implementing rigorous nonlinear dynamic analysis to guide the design of buildings featuring plan asymmetry is costly and not practical for day-to-day structural engineering practice. This paper presents a machine learning based methodology to identify a building that may experience the stepped increase in the drift demand ratio (i.e. hump) when the yield limit of the lateral load-resisting elements has been exceeded. Parametric studies based on nonlinear dynamic analysis of single-storey buildings with structural walls in varying number, size and position are undertaken to examine the effect of system parameters on hump that may occur in the post yield conditions. Buildings are divided into three categories including no hump, slight hump and large hump by assessing the increase in the inelastic drift demand ratio in comparison to the elastic drift demand ratio. Machine learning based prediction models have been developed to achieve a rapid identification of hump in a building based on dynamic analysis results of various single-storey buildings. The models can be an effective tool for optimising the design of plan asymmetric buildings by identifying the potential seismic risks posed by torsional action at the preliminary design stage.