Loading...

Detection and Evaluation of Damage in Concrete Structures Using Data-Driven Methods

Ataei, Saeed | 2023

45 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55832 (45)
  4. University: Sharif University of Technology
  5. Department: Aerospace Engineering
  6. Advisor(s): Adibnazari, Saeed
  7. Abstract:
  8. Infrastructure maintenance is one of the important tasks that require the detection and classification of the condition of their components. One of the main challenges in identifying damage in concrete structures is the speed and accuracy of object detection methods. In this research, vision-based deep learning methods have been implemented to detect and analyze damages in concrete structures. To better evaluate this method, the dataset used in this research is a combination of the three datasets used in researches, which includes 400 pictures of cracks and spalls. This dataset is first augmented using different techniques, then with YOLOv7 instance segmentation and Mask R-CNN algorithms, surface damages of concrete, such as cracks and spalls, are discovered and compared. For this purpose, the target dataset is divided into three sections: training, test, and validation, in the ratio of 90-5-5, respectively. Subsequently, the performance of the two instance segmentation algorithms is evaluated in terms of detection accuracy and speed. In addition, the performance of these two algorithms is compared with other similar works in the literature review. Finally, the effectiveness of these two algorithms was evaluated using the test set, and it was found that the YOLOv7 instance segmentation algorithm, which can be used to identify and detect damages on the surface of structures, has higher accuracy (96.1 %) and speed (44 frames per second) than other algorithms worked on. By optimizing the hyperparameters of these two methods, higher accuracy can also be obtained. As a result, the YOLOv7 instance segmentation algorithm is recommended for automatic and real-time damage detection in concrete structures
  9. Keywords:
  10. Concrete Structures ; Damage Identification ; Mask RCNN Model ; Deep Learning ; Data Driven Method ; You Only Look Once (YOLOv7)

 Digital Object List

 Bookmark

No TOC