3D Solid Architectural Block Classification Using Machine Learning

This research investigates the development of a lightweight machine learning (ML)-based workflow for automated solid CAD block classification. The objective is to develop an efficient model specifically designed to analyse and classify solid architectural CAD blocks, enabling their automated conversion into Building Information Modelling (BIM) components. A comparative analysis is conducted, evaluating the performance of various ML models, ranging from numerical methods such as Logistic Regression and Artificial Neural Networks to more complex Graph Neural Networks. To support this investigation, various versions of a surrogate architectural dataset are generated, comprising 3D models derived from 2D floor plans of single and multiple apartment units. This research demonstrates the power of translating CAD data into machine-learnable formats, such as tabular and graph data, for effective classification. Block classification is based on selected geometric and spatial features, optimized through graphical exploratory data analysis. The results demonstrate the effectiveness of machine learning in classifying solid architectural blocks, with certain models achieving high accuracy on the generated datasets.