Conditional diffusion models for the inverse design of lattice structures

Inverse design, a critical area of mechanical design, focuses on determining the optimal configuration of a structure or material to achieve desired properties or performance. However, the vast array of design possibilities for manufacturable unit cells presents a significant challenge in inverse design: efficiently identifying a complex lattice that meets specific target properties. To address these challenges and, moreover, to offer a solution, we propose a simple yet effective framework that leverages conditional diffusion models, a class of generative models known for their ability to produce high-quality samples conditioned on specific input parameters. Our model, named LatticeOptDiff, enables the efficient exploration of the vast design space, including surface-based, truss-based, and hybrid surface-truss-based lattice structures, by guiding the generation process toward configurations that meet predefined criteria such as Young’s modulus, Poisson’s ratio, and volume fraction. Results indicate that (1) our method can generate various unit cells that satisfy specified material properties with higher accuracy compared to a state-of-the-art conditional generative adversarial network (GAN) and (2) the lattice structures generated through our method exhibit superior mechanical performance when compared to those generated by the GAN. The engineering applications are verified through finite element (FE) simulations and tests on 3D-printed lattice structures. By introducing LatticeOptDiff into the design of lattice structures, we show that conditional diffusion models can outperform GANs in engineering design synthesis, thereby broadening the scope for research and practical applications across diverse engineering fields.