A generative architectural and urban design method through artificial neural networks
Hao Zheng, Philip F. Yuan
Abstract
Machine learning, as a computational tool for finding mappings between the input and output data, has been widely used in engineering fields. Researchers have applied machine learning models to generate 2D drawings with pixels or 3D models with voxels, but the pixelization reduces the precision of the geometries. Therefore, in order to learn and generate 3D geometries as vectorized models with higher precision and faster computation speed, we develop a specific artificial neural network, learning and generating design features for the forms of buildings. A customized data structure with feature parameters is constructed, meeting the requirements of the neural network by rebuilding surfaces with controlling points and appending additional input neurons as quantified vectors to describe the properties of the design. The neural network is first trained with generated design data and then tested by adjusting the feature parameters. The prediction of the generated data shows the basic generative ability of the neural network. Furthermore, trained with design data collected from existing buildings, the neural network learns and infers the geometric design features of architectural design with different feature parameters, providing a data-driven method for designers to generate and analyze architectural forms.