Loading...

Seismic Vulnerability Detection of Shear Buildings Using Wavelet Transformation of Story Responses Via Deep Learning

Doosti, Faezeh | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57537 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Rahimzadeh Rofooei, Fayaz
  7. Abstract:
  8. This research aims to determine the severity of structural damage under earthquakes of varying intensities, taking into account the nonlinear behavior of structures. In recent years, identifying structural damage to prevent sudden collapse and avoid heavy human and financial losses has attracted considerable attention from researchers. Generally, the subject of structural diagnostics can be divided into localized and overall identification methods, with the overall identification approach further categorized into static and dynamic methods. In this study, a dynamic-based overall identification method is employed. Additionally, the identification of damage in large structures requires more powerful processing tools, and in this regard, deep learning, a subset of artificial neural networks, can be utilized. In this research, a combination of coefficients obtained from the wavelet transform of acceleration responses and relative displacements of floors, along with deep learning methods, is used to identify structural damage. Initially, the response parameters of the middle floors and roof of four-story, seven-story, and ten-story shear frames under scaled earthquake records of varying intensities are stored. The wavelet coefficients of the response parameters, which represent the structural behavior over time and the varying frequency content of the time series, are calculated and plotted. Subsequently, a combination of wavelet transform and convolutional neural networks, considering the nonlinear behavior of the structural models, is employed to determine the damage. The damage criterion in this study is considered to be the maximum inter-story drift. Minor damages such as corrosion, loosening, etc., are disregarded in this context. The convolutional network has demonstrated high accuracy in determining the severity of damage in structures with nonlinear behavior, achieving 91% accuracy in determining the level of structural damage. Additionally, to validate the constructed network, wavelet images of other floors from the three existing structures were taken and fed as input into the neural network. The level of damage obtained from the network was then compared with the results from time history analysis. The neural network was able to accurately classify the damage levels of the images with 95% accuracy
  9. Keywords:
  10. Structural Health Monitoring ; Deep Learning ; Wavelet Transform ; Nonlinear Analysis ; Structural Damage Detection (SDD) ; Drift Ratio ; Interstory Drift ; Shearing Buildings

 Digital Object List

 Bookmark

No TOC