Study on the evolution of dynamic characteristics and seismic damage of a self-centering concrete structure based on data-driven methods

Abstract

This study concerns on the dynamic characteristics and seismic damage evaluation of a representative precast self-centering concrete frame (PSCCF) structure. A numerical model of the PSCCF was established using OpenSees software, and incremental dynamic analyses (IDA) were carried out. The structural frequency was obtained from displacement time-history responses using a synchrosqueezing continuous wavelet transform (SSQCWT), and the time-frequency distributions were statistically analysed. During seismic excitations, both the medium structural frequency and the post-seismic frequency of the PSCCF decrease with an increase in peak ground acceleration (PGA). The results demonstrated that the PSCCF suffered irreversible damage under significant deformation, also confirmed by the relatively large elongation rates of the post-tensioned (PT) strands and unbonded energy-dissipating bars (EDBs). Furthermore, this research uses convolutional neural networks (CNN) to predict the damage state of PSCCF, and employs gradient-weighted class activation mapping (Grad-CAM) and Gradient-Shap techniques to interpret the convolutional layers. The interpreted results show that the algorithm tends to use lower-frequency signal components to predict the structural damage state, and the data-driven results are consistent with human experience.