Loading...

“Damage Detection and Localization with Wavelet Transformation of Response Accelerations of a Shear Building under Seismic Records Via Deep Learning”

Mirfakhar, Fatemeh | 2022

139 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55482 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Rahimzadeh Rofooei, Fayaz
  7. Abstract:
  8. The main goal of this research is detection, localization, and determining the severity of structural damages. The importance of this objective is to prevent abrupt destruction and severe human and financial losses. Damage detection is divided into local and global detection, and global detection contains static and vibration-based methods. In this research, the vibration-based method is used. Due to the complexity of this method, accurate and powerful tools are required for detecting effective features in damages; Deep Learning (DL), which is part of a broader family of machine learning methods based on artificial neural networks (ANN), could be used. Accordingly, in this project, the combination of Wavelet Transformation of acceleration and DL is utilized to enhance the reliability and productivity of this method. Before using this combination, the acceleration responses of a four-story shear building under seismic records are preserved. Then the Wavelet transformation of these responses is obtained, and their diagrams are depicted because this transformation allows showing the structure’s state and frequency at each moment. For the next step, these diagrams are used to construct Convolutional Neural Network (CNN) to meet the core aim. Results support the effectiveness of using the combination of wavelet transformation and CNN. The outcome of this research is divided into three phases: (1) determining the ability of this combination to detect that the structure is intact or damaged (training accuracy: 99.57 percent, test accuracy: 97.51 percent, and the loss was less than 0.05 in both training and testing data). (2) the ability to detect and localize damages (training accuracy: 87.03 percent, test accuracy: 83.92 percent, and the loss was less than 0.15 in both training and testing data). (3) the ability to meet the main goal (detecting, localizing, and determining the severity of damages), training accuracy: 89.58 percent, test accuracy: 88.44 percent and the loss was approximately 0.3132 in both training and testing data).
  9. Keywords:
  10. Structural Health Monitoring ; Vibration-Based Structural Damage Detection ; Deep Learning ; Wavelet Transform ; Convolutional Neural Network ; Structural Damage Detection (SDD) ; Acceleration Wavelet Transform

 Digital Object List

 Bookmark

No TOC