Designing Aperiodic Metamaterials Using Mechanical Neural Networks

Abstract

Designing aperiodic metamaterials that accurately achieve multiple behaviours is often not possible due to the enormity of the complex factors that need to be considered. Here we introduce a design approach that uses mechanical neural networks (MNNs) to determine the arrangement of different passive-beam designs within aperiodic metamaterials such that they achieve desired behaviours while considering all of realitity's factors. The MNNs physically learn how to arrange the passive beams by tuning the stiffness of their mechatronically controlled active beams using the same discrete states of stiffness as the passive beams until the desired behaviours are achieved. In this work, two aperiodic metamaterials that achieve different shape-morphing behaviours are designed using a MNN.