Triply periodic minimal surfaces (TPMSs) have attracted great attention due to their distinct advantages such as high strength and light weight compared to traditional lattice structures. Most previous works focus on forward prediction of the mechanical behaviors of TPMSs. Inverse design of the configurations based on customized loading curves would be of great value in engineering applications such as energy absorption. Inspired by TPMSs, we propose the concept of the shell-based mechanical metamaterial (SMM) in this work, which possesses the main geometrical features and mechanical properties of TPMSs. A novel approach, combining machine learning (ML) for high efficiency and genetic algorithm (GA) for global optimization, is put forward to inversely design the configuration of SMM. Two strategies are introduced to develop artificial neural networks (ANNs) for the prediction of their loading curves under compression. GA is then employed to design objective configurations with customized loading curves. The connection between the loading curves and deformation modes is also illustrated to demonstrate the values of such inverse design. SMM with a strain-hardening curve tends to exhibit globally uniform deformation, while SMM with a strain-softening curve tends to present layer-by-layer deformation during compression, which is demonstrated by experiments and simulations. This work fills the blanks of inverse design of SMM with customized loading curves and contributes to the concept of structure design combining ML and traditional optimization approaches.

与传统晶格结构相比，三重周期最小表面（TPMS）由于具有高强度和轻质等独特优势而引起了广泛关注。以前的大多数工作都集中在对 TPMS 的机械行为进行前向预测。基于定制加载曲线的配置逆向设计在能量吸收等工程应用中具有重要价值。受 TPMS 的启发，我们在这项工作中提出了基于壳的机械超材料 (SMM) 的概念，它具有 TPMS 的主要几何特征和机械性能。一种结合机器学习(ML) 用于高效率和遗传算法 (GA) 用于全局优化，提出了逆向设计 SMM 的配置。引入了两种策略来开发人工神经网络 (ANN)，以预测它们在压缩下的加载曲线。然后使用 GA 来设计具有自定义加载曲线的目标配置。载荷曲线与还说明了变形模式以证明这种逆向设计的价值。具有应变硬化曲线的 SMM 倾向于呈现全局均匀变形，而具有应变软化曲线的 SMM 在压缩过程中倾向于呈现逐层变形，这通过实验和模拟得到了证明。这项工作通过自定义加载曲线填补了 SMM 逆向设计的空白，并有助于将 ML 与传统优化方法相结合的结构设计概念。