Microstructured Geometry Creation Using Deep Neural Networks

Abstract

Microstructured geometries, commonly found in natural materials like cork, honeycombs, or bone structures, can mimic complex material properties on the macroscale that homogeneous materials cannot achieve. However, optimizing these structures is challenging due to the one-to-many mapping from macroscale material properties to microscale geometries and the large difference in scale. Neural networks can improve this optimization process by reducing the complex design space to a low-dimensional and continuous latent space, thus making the design exploration more efficient.