Quality assurance for building components through point cloud segmentation leveraging synthetic data

Abstract

Quality Assurance and Quality Control (QA/QC) play a crucial role in the building project life cycle, especially during construction, as discrepancies between as-built structures and as-designed models can lead to cost overruns and schedule delays. Ensuring building quality is of utmost importance, but traditional manual inspections suffer from errors, consume time, and incur significant expenses. This paper describes a deviation detection method for building components using synthetic point clouds and semantic segmentation models. The method entails training a three-dimensional semantic segmentation model using synthetic point clouds generated from BIM to label each point with an object class, resulting in a mean IoU of 41.1% in semantic segmentation. Subsequently, real point clouds collected onsite are segmented using the same model and then compared with the synthetic point cloud to assess the disparities between the building components of the as-designed and as-built structures. This approach can improve the efficiency of QA/QC by reducing the manual workload of field inspectors.