Deep learning-based indentation plastometry in anisotropic materials

Indentation plastometry extracting plastic properties of a material from a non-destructive instrumented indentation has emerged as an efficient and practical method beyond the existing destructive tensile test requiring high experimental cost and effort. However, the use of indentation for describing plastic anisotropy has been insufficiently addressed hitherto. Plastic anisotropy greatly influences the formability of engineering materials, thereby accurately determining the parameters representing the anisotropy is one of the utmost scientific and engineering concerns. In this study, we present a general framework for deriving anisotropic plastic flow from indentation responses, via neural networks (NNs) and finite element (FE) analysis. Hyperparameter-tuned NNs were trained using a database created by parametric studies on experimentally verified FE simulations of indentation. The predictive capability of the developed FE-NN model was thoroughly evaluated with uniaxial plastic curves measured in various directions, followed by an in-depth discussion on the influence of each mechanical parameter on the indentation responses. The validation and predictive performance results demonstrated that the proposed approach is robust and effective in capturing reliable anisotropic plastic flow. Furthermore, we propose that an accurate and stable inverse analysis can be achieved without requiring additional deformation information other than the indentation curve by focusing on the geometrically anisotropic indenter, which has not drawn attention in the inverse analysis.