Convolutional Neural Network-Based Approach Through Mixed Reality in Identification of Light-Gauge Steel Framing Structural Elements for Quality Control

Quality control of the light-gauge steel framing (LGSF) system is essential throughout the fabrication and assembly process to deliver tangible intrinsic value to building stakeholders. The study aims to improve upon conventional methods in quality control, which are often inefficient and error-prone, by leveraging digital technologies for enhanced accuracy and efficiency. This study presents a convolutional neural network (CNN) integrated mixed reality (MR) approach in automating the identification of structural elements within an LGSF frame. Results showed that LGSF elements can be reliably identified at 86% accuracy, 85% precision, and 75% recall rate, with an F1-score of 80%. The average error resulting from this process is 7.08, 6.36, and 10.00 mm in the x-, y-, and z-directions, respectively. By identifying each structural element within the LGSF frame, this study demonstrates that this approach can be further developed for a fully automated quality assurance and control system of prefabricated building systems.