Describing the inelastic deformation behavior of foam structures is challenging due to its complex hardening behavior. Isotropic, kinematic and distortional hardening possibly occur simultaneously and strongly depend on the micro-structure of the foam-like medium as well as on the inelastic behavior of its bulk material.
A common approach to investigate these media is to utilize representative volume elements (RVE), which reduce the overall complexity. Since the effective yield surface, flow rule and corresponding evolution laws are unique for a specific RVE, a homogenized constitutive material model adapting to changes of the RVE micro-structure and bulk properties is needed.
Despite following common thermodynamic theories for elastic-plastic media, the hybrid approach of the current contribution does not define yield and flow potentials a priory. Instead, neural networks which have been trained by data obtained from numerous micro-scale finite element (FE) simulations, are incorporated. This contribution focuses on the general framework --- independent of the underlying RVE --- for describing the three-dimensional inelastic behavior of foam structures. As a model application example, an open-cell Wheire-Phelan foam is considered, whose base material is defined as isotropically elastic plastic, with linear strain hardening. The accuracy of the hybrid approach and the effort required to generate the necessary training data are analyzed and possible extensions to the concept are discussed.