Aplicación de redes neuronales al diseño de vivienda colectiva. Procesos generativos de combinatoria y automatización mediante inteligencia artificial

Abstract

The following paper presents a research project aiming to use artificial intelligence for architectural design and automation processes. The proposed workflow uses the capabilities of neural networks combined with the design of automation algorithms to avoid the repetition of routine tasks. In order to prove the potential appliance of neural networks in architecture, a conditional adversarial neural network (Pix2Pix) is used and trained by the author for the generation of two-dimensional collective housing floor plans. The workflow includes the dissemination of the collective housing complex into individual units, the processing with the neural network and the re-aggregation back into the assembled group. Due to the large quantity of housing floor plans that are needed for the correct training of the network, it has been necessary to automate its process, tagging and storage in data lists. The plans outputted by the neural network are then exported to Grasshopper, where different approximations can be defined through automation processes.