Metamaterials are artificial structures capable of demonstrating extreme properties and unconventional mechanical behaviour thanks to their involved internal organization. Since internal geometry and topology play a major role in defining the performance of the metamaterial, it opens an avenue for designing deep learning models enabling prediction of the resulting properties for specific geometrical parameters. Auxetics – lattices with negative Poisson’s ratio – are one of the most frequently studied classes of metamaterials. In the classical hexagonal or re-entrant lattices, the sign of the Poisson’s ratio is uniquely defined by the angle between struts in the unit cell.
Here we propose an approach that enables control over Poisson’s ratio and other mechanical properties without inducing changes in the angle and dimensions of a metamaterial by employing curvy beams in the architecture of metamaterial. Using Bezier splines, we demonstrate wide variability of the resulting Poisson’s ratio that can be programmed by just a few input parameters representing the coordinates of the spline’s control points. We embrace the one-to-many nature of the inverse problem, creating the tandem architecture combining two neural networks and introducing a concept of the so-called guide curve, enabling us to not only generate materials with desired properties but also express preferences for a specific shape. As a result, we can generate various metamaterials with orthotropic Poisson’s ratio between -5 and 1.
For elastic metamaterials with similar lattice-like architecture, we demonstrate the tunability of passband-stopband structure via curvy elements. While in metamaterial with straight connectors, waves of all frequencies can be sustained, embedment of curve beams facilitates the opening of the bandgaps. For the extreme cases, the ratio of low frequencies inside bandgap zones reaches 70%, proving metamaterials with embedded curvy beams as viable tools for vibration mitigation.