Deep Learning Method for Post-earthquake Damage Assessment of Frame Structures Based on Time–Frequency Analysis and CGAN

This study introduces a new methodology that utilizes time–frequency analysis and deep learning to evaluate the post-earthquake damage analysis of RC frame structures, aiming to enhance assessment efficiency and accuracy. The acceleration signals are subjected to four distinct time–frequency approaches for a six-story RC frame building. To accurately assess the damage condition of the post-earthquake structure, a combination of optimal parameters in a post-earthquake damage assessment model based on a one-dimensional convolutional neural network (1D-CNN) and the Bayesian optimisation (BO) algorithm are employed. The results show that the proposed method achieves a 92.5% accuracy in damage assessment through the wavelet scattering method, which is known for its quick calculation speed. A conditional generative adversarial network (CGAN)-based seismic data generation technique is built to address the issue of inadequate damage sample data sets. By producing high-quality samples that closely resemble actual samples, the combination of wavelet scattering and seismic data generation model increases the accuracy of damage assessment to up to 90.5%. This can be particularly useful in situations when there are limited sample sizes.