Generative Lattice Units with 3D Diffusion for Inverse Design: GLU3D

Abstract

Architected materials, exhibiting unique mechanical properties derived from their designs, have seen significant growth due to the design versatility and cost-effectiveness offered by additive manufacturing. While finite element methods accurately evaluate the mechanical response of these structures, identifying new designs exhibiting specific mechanical properties remains challenging, often requiring computationally expensive simulations and design expertise. This underscores the need for a framework that generates structures based on desired mechanical properties without requiring expert input. In this work, a novel denoising diffusion-based model is presented that generates complex lattice unit cell structures based on desired mechanical properties, manufacturable via additive techniques. The proposed framework generates unique lattice unit cell structures in the implicit domain which can be easily converted to mesh structures for fabrication and voxel structures for structural analysis. The proposed model accelerates the design process by generating unique structures with both isotropic and anisotropic stiffness, outperforming traditional unit cells like simple cubic and body-centered-cubic in energy absorption and compression load at lower densities. Additionally, this work explores a new class of hybrid structures, derived by combining multiple configurations of triply periodic minimal surface structures using non-linear transition functions, which may offer equivalent or enhanced strength compared to conventional designs.