Origami-inspired metamaterials are increasingly being applied to the fields of science and engineering owing to their unique mechanical characteristics. Geometric and material arrangements of the panels and creases can lead to different mechanical properties of origami structures, such as energy absorption and multi-stability. The actual folding processes of the origami structures usually include crease rotation and panel deformation. However, the traditional analytical or empirical solutions based on rigid folding assumptions cannot sufficiently reflect the folding process precisely. Finite element analysis can provide detailed origami folding information, but it is a highly time-consuming cycle, including modeling and computing. Typically, conventional data-driven approaches necessitate a considerable amount of data samples to analyze the energy absorption of origami structures. The unavailability of extensive datasets poses a major impediment in employing data-driven methods to explore the authentic behavior of origami structures. Based on the situations above, a new data-driven framework with conditional generative adversarial networks with tabular data (CTGAN) is proposed in this study. To verify the feasibility of the framework, two study cases are presented herein: a Miura-ori structure in-plane quasi-static compression and the square-twist (Type 1) folding process. Energy absorption properties are predicted accurately and efficiently by the framework based on small number of samples, which are highly consistent with predicted results based on a large dataset. The framework not only provides a promising solution for energy absorption analysis of origami structures but also overcomes the bottlenecks (small number of sample datasets and a mixture of data types) associated with machine learning methods in origami structure area.
Read full abstract