Graph machine learning classification using architectural 3D topological models

Some architects struggle to choose the best form of how building meets ground and may benefit from a suggestion based on precedents. This paper presents novel proof concept workflow that enables machine learning (ML) automatically classify three-dimensional (3D) prototypes with respect formulating most appropriate building/ground relationship. Here, ML, branch artificial intelligence (AI), can ascertain relationship set examples provided by trained architects. Moreover, system classifies 3D architectural precedent models topological graph instead 2D images. The takes advantage two primary technologies. first is software library enhances representation through non-manifold topology (Topologic). second an end-to-end deep convolutional neural network (DGCNN). experimental in this consists stages. First, generative simulation for prototype precedents created large synthetic database relationships numerous variations. geometrical model then underwent conversion into semantically rich dual graphs. Second, graphs were imported DGCNN classification. While using unique data prevents direct comparison, our experiments have shown proposed achieves highly accurate results align DGCNN’s performance benchmark research demonstrates potential AI help designers identify solutions place them within relevant canons.