Loading...

Vibration-based Structural Damage State Identification by Image-based Two-dimentional Convolutional Neural Network

Daeizadeh, Mohammad javad | 2020

327 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54287 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Mohtasham Dolatshahi, Keyarash
  7. Abstract:
  8. This paper proposes a novel image-based two-dimensional convolutional neural network for identifying damage level of the structures after an earthquake. The acceleration of the structure is the input data that is converted into an image, and the corresponding damage level is the output of the network. The superiority of the proposed method in comparison to the signal-based one-dimensional convolutional neural network method is the incorporation of the high and low frequency of the input data into the kernel of the convolution. Rows of the input image show short period high frequency of the signal and the column represent long duration and low frequency of the response time history acceleration. Damage level detection is conducted by Transfer Learning on a pre-trained Alexnet network. For training the network a single degree of freedom model is developed representing a pier of bridges with a concentrated plastic hinge at the base. The model is subjected to a series of time history ground motions and at each time the acceleration of the top and the plastic rotation of the bottom hinge are recorded as the training and test data. It is shown that the proposed network is capable of predicting the damage levels by a precision level of above 90 percent. This precision level for the image-based method is about 20 percent higher than that of the signal-based approach
  9. Keywords:
  10. Post-disaster Inspection ; Damage State Prediction ; Convolutional Neural Network ; Pre-Trained Alexnet Network ; Image-Based Convolutional Neural Network ; Signal-Based Convolutional Neural Network

 Digital Object List

 Bookmark

No TOC