Shell Model Reconstruction of Thin-Walled Structures from Point Clouds for Finite Element Modelling of Existing Steel...

Digital twin models utilising point cloud data have received significant attention for efficient bridge maintenance and performance assessment. There are some studies that show finite element (FE) models from point cloud data. While most of those approaches focus on modelling by solid elements, modelling of some civil structures, such as bridges, requires various uses of beam and shell elements. This study proposes a systematic approach for constructing shell element FE models from point cloud data of thin-walled structural members. The proposed methodology involves k-means clustering for point cloud segmentation into individual plates, principal component analysis for neutral plane estimation, and edge detection based on normal vector variations for geometric structure determination. Validation experiments using point cloud data of a steel corner specimen revealed dimensional errors up to 5 mm and angular errors up to 6°, but static load analysis demonstrated good accuracy with maximum displacement errors within 3.8% and maximum stress errors within 7.7% compared to nominal models. Additionally, the influence of point cloud data quality on FE model geometry and analysis results was evaluated based on geometric accuracy and point cloud density metrics, revealing that significant variations in density within the same surface lead to reduced neutral plane estimation accuracy. Furthermore, toward practical application to actual bridge structures, on-site measurements and quality evaluation of point cloud data from a steel plate girder bridge were conducted. The results showed that thickness errors in the bridge data reached up to 2 mm, while surface deviation RMSE ranged from 3 to 5 mm. This research contributes to establishing practical FE modelling procedures from point cloud data and providing a model validation framework that ensures appropriate abstraction in structural analysis.