Automated bridge modelling for digital twins: parametric modelling vs. surface modeling techniques for geometric-semantic model creation

Building Information Modeling (BIM) is transforming construction practices, but its application to existing assets remains challenging due to discrepancies between original designs and as-is conditions. Moreover, also Digital Twins require accurate geometric-semantic as-is models, which are traditionally labor-intensive to create. This paper presents a fully-automated approach for generating as-is models of bridges by comparing parametric and surface modeling techniques using semantically segmented 3D point cloud data. The methodology employs Evolutionary Strategy optimization for parametric modeling and RANSAC-based quadratic fitting for surface modeling, focusing on capturing complex geometries like horizontally curved bridges. As expected, our comparative analysis reveals that surface modeling achieves much better accuracy with 60% lower RMSE (0.05m vs. 0.126m) in our case study and requires 90% less processing time (1.2 minutes vs. 12 minutes) than parametric modeling. Both approaches successfully generate Industry Foundation Class (IFC) models, providing standardized geometric-semantic representations suitable for lifecycle BIM and Digital Twin applications.