Article:

In this paper, we are aiming to present a methodology for generation,manipulation and form finding of structural typologies using variationalautoencoders, a machine learning model based on neural networks. We aregiving a detailed description of the neural network architecture used as well asthe data representation based on the concept of a 3D-canvas with voxelizedwireframes. In this 3D-canvas, the input geometry of the building typologies isrepresented through their connectivity map and subsequently augmented toincrease the size of the training set. Our variational autoencoder model thenlearns a continuous latent distribution of the input data from which we cansample to generate new geometry instances, essentially hybrids of the initial inputgeometries. Finally, we present the results of these computational experimentsand lay out the conclusions as well as outlook for future research in this field.