Implicit Geometric Descriptor-Enabled ANN Framework for a Unified Structure-Property Relationship in Architected Nanofibrous Materials

Abstract

Hierarchically architected nanofibrous materials, such as the vertically aligned carbon nanotube (VACNT) foams, draw their exceptional mechanical properties from the interplay of nanoscale size effects and inter-nanotube interactions within and across architectures. However, the distinct effects of these mechanisms, amplified by the architecture, on different mechanical properties remain elusive, limiting their independent tunability for targeted property combinations. Reliance on architecture-specific explicit design parameters further inhibits the development of a unified structure-property relationship rooted in those nanoscale mechanisms. Here, we introduce two implicit geometric descriptors -- multi-component shape invariants (MCSI) -- in an artificial neural network (ANN) framework to establish a unified structure-property relationship that governs diverse architectures. The MCSIs effectively capture the key nanoscale mechanisms that give rise to the bulk mechanical properties such as specific-energy absorption, peak stress, and average modulus. Exploiting their ability to predict mechanical properties for designs that are even outside of the training data, we propose generalized design strategies to achieve desired mechanical property combinations in architected VACNT foams. Such implicit descriptor-enabled ANN frameworks can guide the accelerated and tractable design of complex hierarchical materials for applications ranging from shock-absorbing layers in extreme environments to functional components in soft robotics.