Informing Architecture with Generated 3D Solid Models

Material data, including interior configurations and faces, is called a 3D solid material. In the past, creating 3D solid materials, especially stochastic ones with a high degree of randomness, was a timeconsuming and expensive process. However, machine learning (ML) advancements have made it easier to create such models. Designers can now use microstructural data of materials to create 3D solid models and incorporate material knowledge into the design process. This paper introduces the Leaf Tower project, which explores ways of informing the structural layout of a tower design through 3D solid material models. Specifically, the project uses an ML-based method called Encoding Deep Convolutional Generative Adversarial Network (EDCGAN) to create a non-existing material model from a close-up image of a leaf. Leaves are typically flat, and their vein structures are essentially anisotropic. EDCGANs trained with a close-up leaf image can generate material blocks resembling leaves, transforming 2D images into 3D blocks. Once a EDCGAN is trained to generate a leaf, it learns how to distribute the paths of veins, which is essential for efficiently dispensing water and other resources throughout the leaf surface, known as leaf venation. These basic leaf functions are repeated layer by layer throughout the synthetic material block. The resulting block demonstrates a gradual change and an adequate leaf pattern on every level. Hence, the generated material block extends the logic of a leaf vein system into the three-dimensional space. The process synthesises a grid originating from the leaf structure, which is gradually changing on the vertical axis. Overall, it creates a geometry that is structurally stable and evenly divided. This project adapts the listed qualities of the leaf block into architectural design, particularly an office tower design, by using the divisions and pockets originating from the leaf veins as structural elements and constraints to planning layout.