Machine learning of viscoelastic properties of 2D porous materials via deep neural network (In Preperation)

Abstract

People are always wondering a clear path that, whether we can design the mechanical properties we desire under some boundary conditions. In this paper, thirty thousand 32 by 32 bits binary microstructure interpreted auxetic samples were generated by a random algorithm and Brownian based algorithm. These samples were then labeled by Finite Element Methods (FEM), with adequate parameters to describe the mechanical properties of the sample material. Our machine learning model is based on a VGG19, an image classification model, but reconstructed as a regression version. We hope to apply this method to the design of viscoelastic auxetic materials, or High damping and high stiffness (HDHS) materials. And with the fast labeling nature of our VGG model, we are able to implement a more efficient optimization model.