Prospecting for Pluripotency in Metamaterial Design

TL;DR

This paper introduces a data-driven method combining neural networks and genetic algorithms to systematically discover rare, pluripotent metamaterials capable of multiple deformations, overcoming traditional optimization challenges.

Contribution

The authors develop a novel approach that efficiently identifies and refines pluripotent metamaterials with multiple target shapes, bypassing the limitations of direct optimization.

Findings

Successfully designed pluripotent metamaterials with multiple deformation modes.

Demonstrated the effectiveness of combining CNNs with genetic algorithms for material discovery.

Generated a library of high-performance, multi-shape metamaterials.

Abstract

From self-assembly and protein folding to combinatorial metamaterials, a key challenge in material design is finding the right combination of interacting building blocks that yield targeted properties. Such structures are fiendishly difficult to find; not only are they rare, but often the design space is so rough that gradients are useless and direct optimization is hopeless. Here, we design ultra rare combinatorial metamaterials capable of multiple desired deformations by introducing a two-fold strategy that avoids the drawbacks of direct optimization. We first combine convolutional neural networks with genetic algorithms to prospect for metamaterial designs with a potential for high performance. In our case, these metamaterials have a high number of spatially extended modes; they are pluripotent. Second, we exploit this library of pluripotent designs to generate metamaterials with multiple target deformations, which we finally refine by strategically placing defects. Our pluripotent, multishape metamaterials would be impossible to design through trial-and-error or standard optimization. Instead, our data-driven approach is systematic and ideally suited to tackling the large and intractable combinatorial problems that are pervasive in material science.