Automated Defect Inspection in Building Construction with Multi-Sensor Fusion and Deep Learning

Abstract

The occurrence of defects during the building construction process significantly impacts housing quality. One such defect, the distortion of a building's framework, affects both sustainability and aesthetics. This study presents an automated technique for inspecting framework distortion in building construction by measuring the angles between walls. The proposed method employs a portable data acquisition system that allows for dynamic data collection. The system's accuracy is enhanced through calibration based on terrestrial laser scanning (TLS) data as a reference. Point cloud data are registered to form a map of the interior space, leveraging a deep learning algorithm to visualize framework distortions. When tested in an apartment construction environment, the method reduces data acquisition time compared to the TLS-based approach, while maintaining precision with an average angular error of 0.28 degrees. This study demonstrates a cost-effective and accurate solution for defect inspection in the construction industry.