Optimized design for modified auxetic structures based on a neural network approach

Abstract

New technologies and advances in all fields of science increase the demands on our products and materials. Cost efficiency, lightweight, high toughness or a long lifespan are just a few examples. Hence, optimized materials and structures are necessary to develop and investigate. Three dimensional (3D) re-entrant auxetics combine lightweight, high strength, fracture resistance and high impact resistance, making them an ideal choice for crash absorbers and blast protection devices. An optimization work flow was implemented combining finite element (FE) simulations, neural networks and a surrogate model technique to gain the best compromise solution between the mass specific energy absorption capacity and a negative as possible Poisson’s ratio. Parametrized FE simulations were used to generate a training database for the neural network, which then handles the structural optimization via a surrogate model. The resulting structure was produced using additive manufacturing and investigated under compressive loading to validate the simulations. The neural networks were able to predict the stress–strain relation closely and the simulations were a good match with the experiments.