Denoising and Voxelization Algorithms for Finite Element Analysis of Cultural Heritage

Abstract. Reality-based surveys deliver 3D point clouds that can be converted in volumes for FEA. The paper aims at analysing the geometric accuracy for the conversion of 3D point clouds: first step is the denoising of 3D data, then the automatic voxelization for the volumetric models that can be used for structural analyses through Finite element analysis (FEA) process. This work presents an automated pipeline for point cloud denoising and voxelization using Open3D and SciKit-Image. The approach leverages Marching Cubes for surface extraction and exports the results in STL format, ensuring compatibility with CAD and manufacturing systems. This process has been compared to an established one using retopology for the direct conversion of 3D meshes to volumes. The test objects selected for this study are the statue of Moses by Michelangelo, a stool in the Royal Palace of Caserta and a part of a suspension of a Porsche Cayenne. The choice was made considering the increase in geometric complexity of the object, to test the algorithm on objects of different shapes and dimensions. The portion of the suspension was chosen because laboratory data on traction and compression are available for comparison. In this stage, geometrical comparison between the different models have been done to set up the best pipeline for volumetric models for FEA.