Reconstructing orientation data from the images of IPF maps and ODF sections extracted from the literature: A data-collection method for machine learning

Crystallographic orientation data are used as a precursor to various microstructural and textural analyses as well as to perform simulations such as crystal plasticity. In the current study, a novel method is developed to reconstruct orientation data from the images of inverse pole figure (IPF) maps and sections of orientation distribution functions (ODFs) extracted from published articles. Crystallographic orientation can be expressed by Miller indices { hkl} (uvw) where the {hkl} plane normal is parallel to the normal direction (ND) and the (uvw) direction is parallel to the rolling direction (RD) of the specimen. The {hkl} values of the orientations were obtained by the image processing of ND-IPF maps while the corresponding (uvw) values were obtained by optimization of the ODF sections via the use of a genetic algorithm (GA). The MTEX software was used to generate ODF sections with optimized data, and these were then compared with real ODF sections to evaluate the error function of the GA. The quality of the reconstruction of orientation data was evaluated by performing the proposed method on images from the ND-IPF map and ODF sections produced through previously available experimental data. The results of both reconstructed and actual data were compared in terms of grain-size distribution, misorientation-angle distribution, texture component fractions, and the overall {hkl} and (uvw) axis distributions. Furthermore, a visco-plastic self-consistent (VPSC) simulation equipped with a Marciniak-Kuczy ' nski (M-K) model along with visco-plastic fast Fourier transform (VPFFT) simulation was applied to reconstructed orientation data in order to obtain the plastic strain ratio (r-value), the forming limit curves (FLCs), and the strain localization distribution, respectively.

Automated summary

A novel method is developed in this study to reconstruct crystallographic orientation data from published articles, and the effectiveness of this method is demonstrated through comparison with real data.