GraphDGM: A Generative Data-Driven Design Approach for Frame and Lattice Structures

Abstract Generative artificial intelligence offers a more efficient solution for the design of structures. However, an inverse generation of structures, which meet multiple design objectives, remains an open problem. This article thus focuses on the inverse design of frame structures and proposes Graph-based Diffusion-Generative Multiobjective design (GraphDGM), a graph-based generative data-driven surrogate model constrained by multiple targets. By integrating the finite element method (FEM), we construct datasets of frame structures subjected to various conditions. We then developed a conditional graph generation model based on the denoising diffusion probabilistic models (DDPM) and the attention mechanism. We show that our method can efficiently accomplish the inverse design of various frame structures, including a vehicle’s skeleton subjected to five simultaneous constraints. Furthermore, we present comparative experiments against baseline methods to demonstrate the effectiveness and superiority of the GraphDGM.