Shixiang Zhao

Determining the parameters of constitutive models is typically a material-specific process that requires recalibration when the material changes. This procedure becomes notably time-consuming with an increased number of parameters and integral terms in the constitutive equations. Recently, a novel approach utilizing neural networks has emerged as an alternative. By training neural networks to replace constitutive equations, constitutive substitution method maps the extensive parameter space to computational results equivalent to those derived from original equations. Comparing these results with experimental data allows for efficient determination of optimal parameter vectors. The powerful fitting capabilities and rapid computational speed of neural networks significantly expedite the parameter determination process. Consequently, the task of solving for parameters transitions from a complex optimization problem to a straightforward computational search. In this work, we introduce an iterative parameter generation (IPG) algorithm to enhance this substitution method, thereby improving its ability to fit a diverse range of materials. Additionally, we explore the benefits of constructing a general parameter space applicable to various material classes.

The paper proposes an incremental relaxation plasticity (IRP) model for predicting possible instabilities and overall behavior of flow curves under dynamic loads. Compared to its original non-incremental version, the IRP model allows one to predict the behavior of stress-strain curves over a longer time after the start of yielding and to more accurately describe their instabilities such as sharp yield points (yield drops) and further nonmonotonic or oscillatory effects. The efficiency of the IRP model is demonstrated by comparing its predictions with those of the original non-incremental version and of the widely known Johnson–Cook model on the example of experimental flow curves for dual-phase high-strength steel DP800 and aluminum alloy 2519A. The major feature of the proposed IRP model is that its parameters are invariant with the loading history and strain rate of a material and are related only to the evolution of its defect structure on the micro- and mesoscales. With such a set of IRP model parameters, one can obtain a variety of flow curves of the same material at widely varied strain rates.