All times are based on UTC +2 resp. CEST.
Applications of a hybrid approach to describe the elastic-plastic deformation behavior of highly porous media by neural networksWednesday (07.10.2020) 16:40 - 17:00 Room 2 Part of:
The constitutive description of inelastic deformation behavior of porous media is a challenging task. The complex hardening behavior (isotropic, kinematic and distortional hardening at the same time) and anisotropic yielding depend strongly on the microstructure of the porous media and the inelastic behavior of its bulk material. To investigate the behavior of porous media representative volume elements (RVE) are used to reduce the overall complexity. In general each structure has its own special yield surface, flow rule and corresponding evolutions, which is why a homogenized constitutive material model is needed that is adaptable to changes of the RVE micro structure and bulk material behavior. The hybrid approach follows common thermodynamic theories for elastic-plastic media, but yield and flow potentials are not given a priory. Instead neural networks are used which are trained with data obtained by micro-scale simulations. This contribution concentrates on application aspects, as the amount of necessary training data, data sampling strategies, neural network size and complexity. The hybrid approach is applied for a three-dimensional open cell Wheire-Phelan-Foam. The approximation accuracy is as well discussed as the necessary numerical effort.