Physically-Motivated Elasto-Visco-Plastic Model for the Large Strain-Rate Behavior of Steels

Abstract

A physically based elasto-visco-plastic constitutive model is presented and compared to experimental results for three different mild steels. The experiments consist of tensile tests at strain rates up to 103 s-1 and reverse shear tests. The model requires significantly fewer material parameters compared to other visco-plasticity models from the literature while exhibiting very good accuracy. Accordingly, the parameter identification is simple and intuitive, requiring a relatively small set of experiments. The strain-rate sensitivity modeling is not restricted to a particular hardening law and thus provides a general framework in which advanced hardening equations can be adopted. The model was eventually used as the basis for a homogenization approach at the phase scale; preliminary investigations showed the benefit of such an approach, where microstructure-relevant data can explicitly enter the model and may be used for material design simulations.

Unsupervised learning of history-dependent constitutive material laws with thermodynamically-consistent neural networks in the modified Constitutive Relation Error framework

“…In contrast, the form of the dissipation pseudo-potential is released with a neural network. This neural network is initialized to represent a classical power law [52], whereas the model used to generate the data is a hyperbolic cosine law [71], as summarized in Table 3. Figure 18 shows the discrepancy between the initial and the true models.…”