Generation and evaluation of unimaginable three-dimensional structural joints using generative adversarial networks

Abstract

Compared to the increasingly programmable and automated structural analysis methods, the current joint design suffers from long design time, poor quality and a low degree of automation. Developing an efficient and intelligent method to generate structurally novel and high-performing joints has become an urgent task. This paper combines artificial intelligence, topology optimization and reverse engineering techniques to develop a method named HD-JointGEN, and elaborates the idea and implementation steps of this method through an actual engineering example of a single-layer reticulated shell structure. The joints generated in batch by HD-JointGEN are analyzed and evaluated, revealing that the selected representative joint can reduce self-weight by 29.55%, maximum displacement by 0.99%, and maximum equivalent stress by 1.28% compared to the topology optimization joint. HD-JointGEN not only automatically generates a variety of innovative and previously unimaginable joints but also offers design solutions with superior force performances. It leads the research on intelligent generative design of the connection in complex three-dimensional structures and has broad application prospects.