|About this Abstract
||NUMISHEET 2022: The 12th International Conference on Numerical Simulation of 3D Sheet Metal Forming Processes
||Towards Development of a Machine-learning Based Universal Plasticity Model of Sheet Metal under Arbitrary Loadings
||Maysam Gorji, Julian N. Heidenreich, Dirk Mohr
|On-Site Speaker (Planned)
Artificial Neural networks (ANNs) provide a potentially viable alternative to differential equation-based constitutive models. In this study, an ANN model is developed to describe the large deformation response of anisotropic sheet material. Using a conventional return- mapping scheme, virtual experiments are performed to generate stress-strain data for (i) the cyclic uni-dimensional reversal loading to predict the Bauschinger effect, (ii) the two-stage uni- dimensional tensile tests to predict the latent effects, and (iii) monotonic proportional loading and arbitrary loading paths under the biaxial loadings. In the first step, different synthetic stress-strain paths have been generated on a 1×1 mm single shell element in the strain space by using the physical-based Yld2000-2d plasticity model along with the HAH (homogenized anisotropic hardening) approach that is calibrated by real experiments. Subsequently, depending on a problem, a fully connected NN or recurrent NN is trained and validated using the results from virtual experiments. The results of a derived shallow network show remarkably good agreement with all experimental data; they show the great potential of using machine learning to characterize the elastoplastic behavior of a material. An additional byproduct of this study is employing the ANN for the homogenization of the multi-phases materials.
||Definite: At-meeting proceedings