Loading...

Development of a Machine Vision-Based Tool for Assessing Rebar Installation Quality in Structures

Sahebzadeh, Amir Mohammad | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58251 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Kashani, Hamed
  7. Abstract:
  8. Non-compliance with engineering standards in the installation of reinforcement bars (rebar) in concrete structures leads to a reduction in both load-bearing capacity and seismic performance, potentially resulting in damage in the event of an earthquake. Accurate and rigorous supervision of rebar installation is, therefore, critical to minimizing seismic vulnerability. Traditionally, this supervision and quality control process is carried out manually by experts, which is often time-consuming, costly, and, in some cases, hazardous. Given the limitations of traditional inspection methods, such as their complexity and susceptibility to human error, it is evident that execution errors stemming from contractor mistakes or supervisory negligence can significantly compromise structural integrity and may even lead to disputes between contractors and project owners. Rebar installation quality can be assessed either in real-time or post-installation, the latter being more common in construction projects. However, post-installation inspections often result in rework when non-compliance is detected, leading to additional costs and project delays. Real-time monitoring of rebar installation allows for immediate detection and prevention of errors, thereby avoiding rework and its associated costs. In this study, for the first time in the construction industry, a comprehensive framework is proposed for evaluating rebar and stirrups through instance segmentation combined with monocular depth estimation based on deep learning and geometric feature estimation algorithms. This research utilizes the Mask R-CNN algorithm for three models: Rebar-Stirrup Model, the Rebar Section Model, and Stirrup Angle-estimation Model. These are integrated with a monocular depth estimation algorithm based on deep learning, which estimates the metric depth of objects with a mean accuracy of 93.81%. This approach extracts geometric features relevant to each object and model. The rebar-stirrup model detects objects with an accuracy of 82.5% and estimates diameter, spacing, count, and alignment of rebar and stirrups with a minimum accuracy of 88%. The rebar cross-section model identifies rebar cross-sections with 78% accuracy and estimates their count and diameter in the image with at least 93% accuracy. The stirrup angle estimation model recognizes stirrup bends with 80% accuracy and estimates 90° and 135° bend angles with 94.2% accuracy. The findings of this study demonstrate that the proposed framework facilitates rebar and stirrup evaluation in construction sites, reduces human error, enhances structural durability, and ultimately improves safety in construction environments
  9. Keywords:
  10. Machine Vision ; Seismic Vulnerability ; Image Processing ; Operation ; Quality Control Inspection

 Digital Object List

 Bookmark

No TOC