Near-fault pulse seismic ductility spectra for bridge columns based on machine learning

The seismic ductility spectra (SDS) method is a crucial tool to quickly evaluate the ductility demand of bridge columns with varied constitutive models. However, conventional SDS methods usually do not consider the characteristics of pulse-like excitations, which tend to cause severe structural damage. To address this challenge, near-fault pulse seismic ductility spectra (NFPSDS) based on Machine learning (ML) are developed in this study, where two ML models, i.e., the Random Forest (RF) and the artificial neural network (ANN), are utilized to map the relationship between seismic demand and a pulse-structure coupled index alpha 1-p (the structural fundamental period (T1) relative to pulse period (Tp)). Then, the influence of column parameters (such as fundamental period and longitudinal reinforcement ratio) and pulse parameters (such as pulse period and peak pulse velocity) on NFPSDS is quantitatively investigated. Thus, by employing the NFPSDS method, the reasonable design range of longitudinal reinforcement ratio can be obtained under different pulse periods and pulse velocities. Overall, the NFPSDS method significantly benefits the practice for seismic design of structures in near-fault regions.

Automated summary

The study develops a near-fault pulse seismic ductility spectra (NFPSDS) method based on machine learning to quickly evaluate the ductility demand of bridge columns. The method utilizes two machine learning models, random forest (RF) and artificial neural network (ANN), to establish the relationship between seismic demand and the pulse-structure coupled index alpha 1-p. The method also investigates the quantitative influence of column parameters and pulse parameters on NFPSDS. It provides a reasonable design range of longitudinal reinforcement ratio under different pulse periods and pulse velocities, thereby significantly benefiting the seismic design of structures in near-fault regions.