Loading...

Condition Assessment and Damage Detection in Concrete Structures Using Computer Vision-Based Deep Learning Techniques

Younesian, Ali | 2022

252 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54807 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Khaloo, Alireza
  7. Abstract:
  8. The ultimate goal of this study is to evaluate the condition of concrete structures using computer vision methods and using powerful tools such as machine learning methods based on visual information. This assessment is performed by detecting damage in these structures. With the aging of structures such as dams, bridges and tall buildings, structural health monitoring is an important task in ensuring their safety and stability. Therefore, rapid assessment of the health of structures and diagnosis of damage after destructive events is of great value in terms of providing resilience of structures. Visual inspection of structures by experts is one of the basic methods of evaluating structures. These methods are not possible for structures with inaccessible geographical locations and it is also quite expensive.Today, computer vision and machine learning techniques are widely used to assess the condition of structures, including crack and damage detection. The old methods lacked accuracy and precision. Therefore, in this research, an attempt is made to use a deep learning tool based on semantic segmentation. In this research, with the help of this method, in the existing images of concrete structures, damage detection, including identification and location, as well as dimensional examination of cracks are performed.
    Using deep learning computer vision methods, powerful models for automatic crack detection can be provided. The ultimate goal of such a model is to extract the geometric properties of cracks in an image and convert them into analyzable information. The output of this model is a split image in which cracked neighborhoods are visible on concrete surfaces and the crack pattern is recognizable in the input image. Using advanced knowledge in convolutional neural network programming in Python, this research offers two models for fast and accurate crack detection at the pixel level of the input image. Specifically, the authors have generated a dataset designed to validate and train the model and use a open-source dataset from up-to-date research in this field. The results show that the models designed in this study can detect crack pixels with an approximate accuracy of more than 99%
  9. Keywords:
  10. Structural Health Monitoring ; Damage Detection ; Machine Learning ; Deep Learning ; Semantic Segmentation ; Crack ; Concrete Structures ; Computer Vision ; Vision Based Inspection ; Automated Inspection

 Digital Object List

 Bookmark

No TOC