Real-time Quality Control for Offsite LGS Frame Manufacturing using Vision-based Deep Learning

• Author / Creator

  Nader, George

• Efficiency and accuracy are crucial in offsite panelized construction, including in the production of light-gauge steel frames. Maintaining consistent screw fastening is an ongoing challenge in the construction industry, and defective screw fastening can lead to quality problems in panelized construction projects. To address this issue, this research introduces a light-gauge steel framing machine that integrates a computer-vision system for precise quality control. This framework is developed and tested on real panelized construction projects. This innovative approach equips the light-gauge steel framing machine with visual perception capabilities, enabling real-time image capture and analysis of the framing process. By employing advanced imaging technology and YOLOv8n machine learning architecture, the computer-vision system provides immediate feedback to the machine’s control system. This process facilitates precise decision-making regarding screw- fastening operations. The implementation of YOLOv8n on the light-gauge steel framing machine uses Python and the OpenCV library to process visual data in real -time, determining optimal methods to mitigate defects. The experimental results demonstrate that the computer-vision system substantially improves the integrity and precision of light-gauge steel frames. The results indicate a reduction in defects and rework, as well as significant enhancements to operational efficiency and material utilization. These enhancements underscore the promising potential of integrating real-time computer vision and artificial intelligence with the manufacturing processes of panelized construction, establishing a foundation for future innovations in autonomous systems within this industrial sector.

• Subjects / Keywords

  • Real-time
  • Quality
  • Offsite
  • Deep
  • Vision

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-9e2a-3428

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.