Journal of Engineering Design
accepted version
Title A deep-learning model for optimising structural topologies Authors (1)
Efthimis Kapsalis, (1,2) Ioannis Daganis
(1)
: AP Engineering, (2): Aristotle University
Abstract This paper presents a novel methodology aimed at accelerating topology optimization in architectural and industrial design form-finding by employing generative adversarial networks (GANs) in place of traditional iterative calculation procedures. The proposed technique efficiently and accurately optimizes two- and three-dimensional structural configurations under specific conditions, despite being trained on a limited dataset generated by a third-party software. By integrating deep learning and topology optimization, the methodology offers substantial benefits to architects and engineers in material conservation and the development of more effective structural arrangements and building enclosures, particularly during the early stages of the design process. The study's findings indicate that the proposed model has considerable potential for real-world implementation in the field of structural engineering, as it demonstrates the ability to produce diverse outcomes related to the enhancement of structural topologies, including trusses. This research paves the way for further exploration and refinement of computational design techniques in the pursuit of more efficient and sustainable design solutions. Keywords: topology optimisation, generative adversarial networks, form-finding, structural engineering, architectural engineering