Beyond Traditional H-BIM: AI-Powered Modeling for Heritage Bridges

Abstract. Heritage Building Information Modeling (HBIM) is the methodology that addresses the growing needs in the management and preservation of Cultural Heritage, by integrating three-dimensional digital models with spatial, temporal, organizational, operational, and other types of information. The digital model, derived from a 3D survey, requires significant time and training efforts when transitioning to the HBIM methodology. Moreover, this process often leads to substantial geometric approximations, due to the limited flexibility of converting point clouds into BIM during manual modeling. 
In this context, artificial intelligence plays a fundamental role through the development of algorithms specifically designed to transform project documentation or point clouds into semantic three-dimensional models exportable in IFC format. The proposed case study explores this experimentation applied to an ancient bridge, where infrastructure and culture converge in a perfect dialectical and formal expression of heritage. 
The three main phases of the proposed workflow, photogrammetric acquisition via drone, semantic segmentation of the point cloud using AI models (RandLA-Net), and generation of the parametric model in a BIM environment through tools such as AtlasNet, Rhino, and Revit, have led to a significant reduction in BIM modeling time without compromising the final geometric quality.