Machine learning assisted design of shape-programmable 3D kirigami metamaterials

Abstract

Kirigami-engineering has become an avenue for realizing multifunctional metamaterials that tap into the instability landscape of planar surfaces embedded with cuts. Recently, it has been shown that two-dimensional Kirigami motifs can unfurl a rich space of out-of-plane deformations, which are programmable and controllable across spatial scales. Notwithstanding Kirigami’s versatility, arriving at a cut layout that yields the desired functionality remains a challenge. Here, we introduce a comprehensive machine learning framework to shed light on the Kirigami design space and to rationally guide the design and control of Kirigami-based materials from the meta-atom to the metamaterial level. We employ a combination of clustering, tandem neural networks, and symbolic regression analyses to obtain Kirigami that fulfills specific design constraints and inform on their control and deployment. Our systematic approach is experimentally demonstrated by examining a variety of applications at different hierarchical levels, effectively providing a tool for the discovery of shape-shifting Kirigami metamaterials.

Deep Learning in Mechanical Metamaterials: From Prediction and Generation to Inverse Design

“…Deep generative models can not only generate 2D mechanical metamaterials but also 3D ones, taking advantage of either volumetric convolutional layers or building parameters associated with 3D geometries. [73,74,78,173,174,181] Since 3D arrays store more information than 2D matrices, fewer datapoints are required when using 3D arrays to train an ANN. For instance, GANs with a similar architecture have been used to generate both 2D and 3D mechanical metamaterials, and training a 2D GAN requires only 1/10 the number of datapoints required to train a 3D GAN.…”

“…[76,77,[198][199][200] In semi-direct inverse design, DL models map the property space to the modeling parameters, requiring the use of additional modeling processes. [73,74,109,181] In contrast, direct inverse design uses DL models to straightforwardly generate geometries that meet user-defined properties, represented by pixel images or voxel volumes. [75,78,[201][202][203][204] This method takes advantage of convolutional layers or volumetric convolutional layers to generate geometries efficiently.…”

Machine learning-based inverse design methods considering data characteristics and design space size in materials design and manufacturing: a review

“…Consequently, this strategy has found applications across a range of inverse design problems. [66][67][68][69][70][71] However, the dimension of the input variables that correspond to the material performance is usually lower than the dimension of the output variables that correspond to the materials design parameters, thereby limiting the dimension of the recommended optimal values. Such a problem may not be an issue in the case of simple problems in which the dimension of design variables is relatively small, but the strategy may not be applicable to more complex design problems with higher input and output dimensions.…”

Recent Advances and Applications of Machine Learning in Experimental Solid Mechanics: A Review

“…Alderete and Pathak et al [136] have proposed an ML framework that combines the K-mean clustering methods for design space reduction and a tandem NN architecture to inversely design shape-programmable 3D Kirigami metamaterials. The framework was trained on finite element predictions of instabilities triggering 3D out-of-plane shapes validated by full-field experimental measurements using the shadow Moiré method [265].…”