A computational building block approach towards multiscale architected materials analysis and design with application to hierarchical metal metamaterials

Abstract

Abstract In this study we report a computational approach towards multiscale architected materials analysis and design. A particular challenge in modeling and simulation of materials, and especially the development of hierarchical design approaches, has been to identify ways by which complex multi-level material structures can be effectively modeled. One way to achieve this is to use coarse-graining approaches, where physical relationships can be effectively described with reduced dimensionality. In this paper we report an integrated deep neural network architecture that first learns coarse-grained representations of complex hierarchical microstructure data via a discrete variational autoencoder and then utilizes an attention-based diffusion model solve both forward and inverse problems, including a capacity to solve degenerate design problems. As an application, we demonstrate the method in the analysis and design of hierarchical highly porous metamaterials within the context of nonlinear stress–strain responses to compressive deformation. We validate the mechanical behavior and mechanisms of deformation using embedded-atom molecular dynamics simulations carried out for copper and nickel, showing good agreement with the design objectives.