Structural Plan Schema Generation Through Generative Adversarial Networks

Kösenciğ K. Ö., Okuyucu E. B., Balaban Ö.

Nexus Network Journal, vol.26, no.2, pp.409-427, 2024 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 26 Issue: 2
  • Publication Date: 2024
  • Doi Number: 10.1007/s00004-024-00766-z
  • Journal Name: Nexus Network Journal
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Arts and Humanities Citation Index (AHCI), Scopus, ABI/INFORM, Avery, Index Islamicus, zbMATH, Civil Engineering Abstracts
  • Page Numbers: pp.409-427
  • Keywords: Artificial intelligence (AI), Early design phase, GAN, Plan generator, Structural schema
  • Kocaeli University Affiliated: Yes

Abstract

This paper suggests a workflow that generates floor plans with structural elements. Generating structural layouts in a BIM environment with the implementation of a machine learning method allows a future projection for fast and easy exploration of multiple design options. Pix2Pix, a Generative Adversarial Networks (GAN) model, takes the wall layout as input and generates a structural layout by learning from existing knowledge used to generate a decision support system for structural layout generation. The paper also suggest an additional script as a fine-adjustment model to refine the structural layout based on predetermined structural rules. This script increases the accuracy of the structural layouts generated by the GAN algorithm. Based on the test dataset, the research demonstrates a 64% success rate in providing structural schema assistance. Considering the results, this study seems to have the potential to be a supportive application in the early design phase.