Scan-to-FEM: Advanced point cloud segmentation and geometric information extraction methods for finite element model generation of complex steel structures under construction

Three-dimensional point clouds provide an effective solution for reconstructing as-built geometry and finite element models (FEM) of complex steel structures. However, current point cloud segmentation and geometric parameter extraction methods often result in large deviations between the reconstructed model and the actual structure. Therefore, this paper proposes a scan-to-geometric model reconstruction and FEM updating of complex steel structures using laser scanning technology and advanced deep learning algorithms. First, the FastMAC + GICP algorithm is developed to perform coarse alignment between BIM generated point cloud and measured data for component segmentation of complex steel structures, and an improved rolling-ball algorithm with dual-threshold refinement is used to extract center axes and sectional dimensions of steel components for three-dimensional geometric model reconstruction. Second, a finite element model generation framework is developed by calibrating the design model with the reconstructed geometric model to incorporate construction-induced deviations, from which full-field deformation and dynamic properties of the structure are forecasted for construction strategy decision-making. The contribution of this study is the comprehensive investigation of the engineering applicability of the point cloud segmentation, geometric information extraction, and finite element model generation methods for static and dynamic performance forecasting of complex steel structures. The effectiveness of the proposed method was validated using field test data of a large-scale complex steel structure under construction. Results show that the average deviation between the geometric center axis extraction results and the manually annotated ground truth was only 0.86 mm, representing a reduction of 68% and 51% compared with the traditional algorithms. In addition, the reconstructed mechanical simulation model successfully predicted a maximum cantilever displacement of 376.93 mm under curtain wall installation conditions, providing a scientific basis for construction strategy decision-making.