Cellular materials, widely found in engineered and nature systems, are highly dependent on their geometric arrangement. A non-uniform arrangement could lead to a significant variation of mechanical properties while bringing challenges in material design. Here, this proof-of-concept study demonstrates a machine-learning based framework with the capability of accelerated characterization and pattern generation. Results showed that the proposed framework is capable of predicting the mechanical response curve of any given geometric pattern within the design domain under appropriate neural network architecture and parameters. Additionally, the framework is capable of generating matching geometric patterns for a targeted response through a databank constructed from our machine learning model. The accuracy of the predictions was verified with finite element simulations and the sources of errors were identified. Overall, our machine-learning based framework can boost the design efficiency of cellular materials at unit level, and open new avenues for the programmability of function at system level.

在工程和自然系统中广泛发现的细胞材料高度依赖于它们的几何排列。不均匀的布置可能导致机械性能的显着变化，同时给材料设计带来挑战。在这里，这项概念验证研究展示了一种基于机器学习的框架，该框架具有加速的特征描述和模式生成功能。结果表明，所提出的框架能够在适当的神经网络架构和参数下，预测设计域内任何给定几何图案的机械响应曲线。此外，该框架能够通过从我们的机器学习模型构建的数据库生成针对目标响应的匹配几何图案。通过有限元模拟验证了预测的准确性，并确定了误差来源。总体而言，我们基于机器学习的框架可以提高单元级单元材料的设计效率，并为系统级功能的可编程性开辟新途径。