Graph Machine Learning Approaches to Classifying the Building and Ground Relationship

Abstract

Architects struggle to choose the best form of how the building meets the ground and may benefit from a suggestion based on precedents. A precedent suggestion may help architects decide how the building should meet the ground. Machine learning (ML), as a part of artificial intelligence (AI), can play a role in the following scenario to determine the most appropriate relationship from a set of examples provided by trained architects. A key feature of the system involves its classification of three-dimensional (3D) prototypes of architectural precedent models using a topological graph instead of two-dimensional (2D) images to classify the models. This classified model then predicts and retrieves similar architecture precedents to enable the designer to develop or reconsider their design. The research methodology uses mixed methods research. A qualitative interview validates the taxonomy collected in the literature review and image sorting survey to study the similarity of human classification of the building and ground relationship (BGR). Moreover, the researcher leverages the use of two primary technologies in the development of the BGR tool. First, a software library enhances the representation of 3D models by using non-manifold topology (Topologic). The second phase involves an end-to-end deep graph convolutional neural network (DGCNN). This study employs a two-stage experimental workflow. The first step sees a sizable synthetic database of building relationships and ground topologies created by generative simulation for a 3D prototype of architectural precedents. These topologies then undergo conversion into semantically rich topological dual graphs. Second, the prototype architectural graphs are imported to the DGCNN model for graph classification. This experiment's results show that this approach can recognise architectural forms using more semantically relevant and structured data and that using a unique data set prevents direct comparison. Our experiments have shown that the proposed workflow achieves highly accurate results that align with DGCNN’s performance on benchmark graphs. Additionally, the study demonstrates the effectiveness of using different machine learning approaches, such as Deep Graph Library (DGL) and Unsupervised Graph Level Representation Learning (UGLRL). This research demonstrates the potential of AI to help designers identify the topology of architectural solutions and place them within the most relevant architectural canons.