Ultra-stiff and quasi-elastic-isotropic triply periodic minimal surface structures designed by deep learning

Abstract

Multi-functional triply periodic minimal surface (TPMS) structures have been widely studied and proven to have significant importance. Most studies are based on several typical TPMSs. However, some research suggests that the hybrid forms of the TPMSs can exhibit superior mechanical properties. The challenge lies in the unpredictability of the hybrid TPMS property. In this study, an approach is proposed for precisely identify the hybrid TPMS using fully connected neural network (FCNN) and particle swarm optimization (PSO) algorithm. A design space of four TPMSs hybridization is introduced and used to generate over 8000 samples. Stiffness constants obtained using the homogenization method is used in elastic characterization. The hybrid TPMSs is proved to lack common cubic symmetry but possess wide rotational symmetry. Indirect stiffness constant-based way exhibits better predictions for optimal stiffness and elastic isotropy than direct property parameter-based one. The approach successfully identified useful structures: quasi-elastic-isotropic ones and a ultra-stiff one that is ∼ 9 % stronger than the currently strongest TPMS. The method offers comprehensive guidance for stiffness constant-based TPMS design using DL technique.

Keywords

• Triply periodic minimal surface
• Deep learning
• Homogenization method
• Elastic property
• Additive manufacturing