Inverse machine learning discovered metamaterials with record high recovery stress

Lightweight shape memory polymer (SMP) metamaterials integrated with high strength, high flexibility, and high recovery stress are highly desired in load carrying structures and devices. A grand challenge is that these desired properties have contradictory requirements, for instance between strength and flexibility, and between flexibility and recovery stress. In this study, an inverse design framework using statistical tools and machine learning models is developed to design thin-walled cellular structures with the desired properties. The discovered thin-walled cellular structures are 3D printed using a novel SMP, which exhibited excellent structural properties with record high specific recovery stress. For comparison purpose, lattice structures discovered previously are also 3D printed using the same SMP. The density normalized recovery stress of the validated lattice unit cells is 30% higher than that of the Octet lattice unit cell. The optimal thin-walled unit cells exhibit exponentially higher recovery stress than the honeycomb unit cell in the in-plane orientation and 50% higher recovery stress than other thin-walled structures (both unit cells and 4 x 4 structures). As compared to the solid SMP cylinders, the thin-walled unit cells exhibit 200% higher normalized recovery stress. The inverse design framework can be applied for structural optimization of various other designs and applications.

Automated summary

This study develops an inverse design framework to design thin-walled cellular structures with desired properties using statistical tools and machine learning models. The printed thin-walled cellular structures exhibit excellent structural properties, with record high specific recovery stress.