Digital twin-based bridge geometric quality inspection using knowledge mapping and data-driven method

Accurate geometric quality inspection is vital for detecting bridge defects during construction. To address the challenges of limited point cloud segmentation accuracy at bridge component connections and insufficient detection efficiency, a digital twin bridge geometric quality inspection method based on knowledge mapping and data-driven is proposed. In the digital space, a reusable bridge geometric quality inspection knowledge model is established to enhance the scalability of knowledge. In the twin data processing space, a large-scale point cloud segmentation network based on hybrid feature aggregation and neighbor feature enhancement (HANE-Net) is proposed to improve the segmentation accuracy. The network achieves superior performance in the S3DIS dataset and real bridge point cloud, with mean intersection over union of 66.8 % and 95.44 %, respectively, surpassing the baseline method RandLANet by 3.2 % and 0.79 %, respectively. Finally, a prototype system is designed based on Revit to prove the feasibility of the proposed method.