Diverse 3D auxetic unit cell inverse design with deep learning
Xi Fang, Hui‐Shen Shen, Hai Wang
Abstract
The use of metamaterial structures with auxeticity can result in exceptional mechanical properties, such as high energy absorption and fracture resistance. However, traditional design approaches rely heavily on researchers' subjective experiences, while existing inverse design methods limit design possibilities by ignoring generative diversity. In this study, we report a deep-learning-based inverse design approach for 3D auxetic unit cells that overcomes these limitations by providing diverse and accurately conditioned design options. We construct a dataset of symmetric 3D auxetic unit cells and apply an elastic modulus optimization network to generate diversified spatial topological structures with negative Poisson's ratios and optimized stiffness. The resulting 3D unit cells exhibit improved mechanical properties, as confirmed by finite element simulations and experiments. Our approach offers better coverage of the design space and generates optimized 3D unit cells with rich and diverse properties.