Incorporating biological materials-inspired irregularity into architected materials is potential to unfold a wider property or new function space, while the inverse design of irregular architected materials (IAMs) for targeted properties is still challenging, owing to the intricate structure–property bidirectional relationships. Accordingly, here, a data-driven framework is originally developed to design new irregular architected materials with programmable stiffness. In detail, a robust dataset of IAMs, which are constructed from diverse stretch- and bending-dominated building blocks, is created by a randomized strategy, and an end-to-end deep learning model is established, including a forward and an inverse network. The forward predictions reveal that IAMs exactly achieve a wide space of programmable stiffness, especially, including extremely high or low dominant elastic stiffness C11, C22 and C66. More importantly, the deep learning-driven inverse design flexibly and precisely realizes the generation of IAMs, fulfilling the customized stiffness targets. Especially, various stiffness tensors, including isotropic, orthotropic, and anisotropic types, are successfully achieved. The remarkable advantage of the inverse design is that solely through adjusting the frequency hints of building blocks, without mandatory geometrical symmetry, IAMs present the exclusive ability to hit the targeted stiffness. Overall, this work fills the gap in the inverse design of irregular architected materials with programmable stiffness and paves the way for material design via deep learning.
Read full abstract