Data driven computational analysis of open foam RVEs

An automated approach, that relies on the use of distance and level set functions as explained in [1], has been described in [2] to build computationally Representative Volume Elements (RVE) of open foam materials, enabling the study of the effects of the microstructural features on the macroscopic behavior. These models have been compared with real foam samples from existing literature to verify statistically the morphological properties like face-to-cell ratio, edge-to-face ratio and strut length distribution along with the variations in the strut morphology like the shape of cross-sections of the struts and their variation along the axis of the struts. The responses obtained from a uniaxial compression test of the sample RVEs have been validated against the experimental observations and the results have showed close similarity with respect to the variations in the foam density. This approach enables us to generate multiple Stochastic volume elements (SVE) and get their material response in a short period of time. In [3], the authors have taken inspiration from artificial neural network concepts and used linear elastic RVE data to train a material network to describe complex material behavior. They have also validated the extrapolations of the trained network to a wide range of problems, including non-linear history-dependent plasticity and finite-strain hyper-elasticity under large deformations. In the current work, the goal is to utilize the material responses obtained from the SVEs as the prespecified material data set ( Figure 1 ) and investigate the performance of various data-driven solvers on these data sets in order to eliminate the experimental testing altogether with the knowledge that the material response is in close agreement to that of the RVEs.Enlighteni