Assessment of Damage Degree in High-Rise Building Structures Based on Continuous Wavelet Transform and Improved ResNet18

Vibration-based Structural Health Monitoring (SHM) has garnered significant attention due to its ability to monitor health state in real-time. Current applications of SHM mainly focus on the occurrence of damage but are unable to assess the severity of the structural damage. Achieving damage severity assessment through multi-channel signal processing has become a key focus of current research. To address this issue, we propose an improved TriAxial-ResNet18 architecture, which is capable of processing acceleration signals in the three directions simultaneously, thereby enabling comprehensive damage assessment. In this study, finite element modelling (FEM) is utilized to emulate various damage states in the investigated high-rise building, generating signals under different health states. Time-frequency feature imaging based on continuous wavelet transform (CWT) is employed to visualize the vibration signals, converting the raw signals into two-dimensional images for training and validating the deep learning model. Experimental results indicate that the TriAxial-ResNet18 model achieves high classification accuracy in distinguishing various damage levels across different monitoring points. By integrating simulation technology with deep learning methods, this research provides an economical and efficient solution for SHM in high-rise buildings.