The concept of modular metamaterials and a machine learning-based method are introduced in this Letter. The method starts from selection of the structural bases based on the existing studies and then combines performance evaluation together with structural evolution to construct meta-atoms with specified properties. Both genetic algorithm and neural networks model are adopted to executed the designing process. Mechanical metamaterials with maximized bandgap and tunable bandgaps are demonstrated using the proposed method. This approach offers an effective means to design metamaterials. It is believed that the modular design of metamaterials based on machine learning is capable to construct meta-atoms with specific properties for metamaterials in various fields.

本函介绍了模块化超材料的概念和基于机器学习的方法。该方法从在现有研究的基础上选择结构基础开始，然后将性能评估与结构演变相结合，以构造具有特定性质的亚原子。采用遗传算法和神经网络模型进行设计。使用提出的方法展示了具有最大带隙和可调带隙的机械超材料。这种方法提供了一种设计超材料的有效方法。可以相信，基于机器学习的超材料的模块化设计能够为各个领域的超材料构造具有特定特性的超原子。