Green-GAN | Barcelona, Spain | Unit 21 | 2024
zczlywu@ucl.ac.uk

My project explores generative design and machine learning techniques to facilitate sustainable architecture. This process revealed the power of data-driven generative methods for design ideation. Seeking applications in architectural design, I researched climate priorities for Barcelona. This inspired me to devise a novel technique for mapping estimated carbon costs onto architectural plan diagrams.

My current approach trains conditional GANs to generate building layouts coupled with carbon cost level resulting from metrics for various materials and domains. The vision is to blend data driven generative building models to give architects an explicit understanding of the relationship between carbon cost and building satellite images on a urban scale. I then integrated color-coded carbon cost mapping into my building design process across multiple scale levels. At the micro level, the system provides real-time visual feedback on the carbon footprint of material choices during early sketching and conceptual stages. At the macro level, more comprehensive lifecycle carbon accounting is linked to 3D model environments and applied to whole building forms and assemblies.