Influence of the characteristics of the experimental data set used to identify anisotropy parameters

• Analysis of the influence of the number and type of experimental input data on the yield criteria prediction accuracy.
• Identification of advanced yield criterion parameters using a reduced set of experimental input data.
• Validation of the proposed identification strategies by comparing numerical with experimental results.

This work presents an investigation into the effect of the number and type of experimental input data used in parameter identification of Hill’48, Barlat’91 (Yld91) and Cazacu and Barlat’2001 (CB2001) yield criteria on the accuracy of the finite element simulation results. Different sets of experimental data are used to identify the anisotropy parameters of two metal sheets, exhibiting different anisotropic behaviour and hardening characteristics: a mild steel (DC06) and an aluminium alloy (AA6016-T4). Although it has been shown that the CB2001 yield criterion can lead to an accurate description of anisotropic behaviour of metallic sheets, its calibration requires a large set of experimental input data. A calibration procedure is proposed for CB2001 based on a reduced set of experimental data, i.e. where the results are limited to three uniaxial tensile tests, combined with artificial data obtained using the Barlat’91 yield criterion. Evaluation of the predictive capacity of the studied yield criteria, calibrated using different sets of experimental data, is made by comparing finite element simulation results with experimental results for the deep drawing of a cross-shaped part. A satisfying agreement is observed between experimental and numerical thickness distributions, with a negligible effect of the number and type of experimental data for the Hill’48 and Yld91 yield criteria. On the contrary, CB2001 calibration is quite sensitive to the experimental data available, particularly biaxial values. Nevertheless, CB2001 calibration based on the combination of effective and artificial experimental data achieves satisfying results, which in the worst case are similar to the ones obtained with the Yld91.