Application of neural networks to the design of collective housing Automation and combinatorial generative processes using artificial intelligence.

Abstract

The following article presents a research work oriented towards the use of artificial intelligence for the design of architecture and automation processes. The proposed workflow makes use of the capabilities of neural networks in combination with the design of automation algorithms to avoid the repetition of routine tasks. As a test of the capabilities of neural networks for use in architecture, a conditional adversarial neural network (Pix2Pix) is used, trained by the author for the generation of two-dimensional plans for collective housing. Through this workflow, the disaggregation of the complex into individual housing units, processing by the neural network and re-aggregation into the housing complex are carried out. Due to the large number of housing plans required for training the network, it is necessary to automate the processes, perform labeling and store them in data lists. The plans generated by the neural network are exported to Grasshopper for further processing, and different approaches can be defined through automation processes.