Dr.-Ing. Alexander Henkes
Artificial neural networks in continuum micromechanics
Modern composite materials for lightweight design in aerospace, automotive and mechanical engineering consist of multiple constituents, which induce macroscopic properties by means of their microstructure and characteristics. This introduces a multi-scale problem of inferring quantities of interest on the macro-scale by means of microscopic entities. To this end, computational homogenization algorithms aim towards calculation of effective material properties from a given microstructure, which can by utilized by numerical methods like the finite element method (FEM) to resolve effects on structural levels. If complex material behavior or uncertainty are involved, the multi-scale problem is computationally challenging. A complement to conventional numerical methods is represented by artificial neural networks (ANN). ANN can approximate arbitrary functions, including homogenization of three-dimensional microstructures. A trained ANN represents an analytical function, enabling fast evaluation and allowing for efficient deployment in production-scale simulation environments to accelerate multi-scale computations.