Identification of viscoplastic material parameters from spherical indentation data: Part I. Neural networks

In this paper, a new method for the identification of material parameters is presented. Neural networks, which are trained on the basis of finite element simulations, are used to solve the inverse problem. The material parameters to be identified are part of a viscoplasticity model that has been formulated for finite deformations and implemented in the finite element code ABAQUS. A proper multi-creep loading history was developed in a previous paper using a phenomenological model for viscoplastic spherical indentation. Now, this phenomenological model is replaced by a more realistic finite element model, which provides fast computation and numerical solutions of high accuracy at the same time. As a consequence, existing neural networks developed for the phenomenological model have been extended from a power law hardening with two material parameters to an Armstrong-Frederick hardening rule with three parameters. These are the yield stress, the initial slope of work hardening, and maximum hardening stress of the equilibrium response. In addition, elastic deformation is taken into account. The viscous part is based on a Chaboche-like overstress model, consisting of two material parameters determining velocity dependence and overstress as a function of the strain rate. The method has been verified by additional finite element simulations. Its application for various metals will be presented in Part 11, [J. Mater. Res. 21, 677 (2006)].