Neural Networks for the Identification of Viscoplastic Material Parameters from Spherical Indentation Experiments

•Edouard Tioulioukovski und Norbert Huber
Forschungszentrum Karlsruhe, Institut für Materialforschung II, Postfach 3640, 76021 Karlsruhe

Viscoplastic behavior, which is typical for metal materials was studied. It includes elasticity, plasticity and viscosity. For a special case of the finite deformation thermo-viscoplasticity model of Jansohn & Tsakmakis, the material parameters are determined from spherical indentation experiments. The force controlled loading history of such experiments consists of loading, creep, and unloading. A neural network has been trained using Finite Element simulations with ideal spherical indenters to determine the material parameters of the constitutive model from experimental data.

However, real indenter tips show deviations from the ideal spherical shape. To avoid systematic errors in the material parameters to be identified, a correction method has been developed. On the basis of the real geometry, the correction transforms the measured curve into a curve which corresponds to the same material but an ideal indenter. Then, the neural network for the identification of the material parameters can be applied to the corrected data. Experiments and identification results for different bulk materials are presented.

This method can be extended towards film/substrate-systems to determine material properties of metal films.