Synthetic Data-Driven Predictions of Dislocation Density from Polycrystalline Ta Neutron Diffraction Line Profiles

A synthetic data-driven model of dislocation broadening of diffraction peaks is applied to experimentally measured neutron time-of-flight diffraction of Ta in the prediction of dislocation density. Diffraction profiles of polycrystalline Ta of various dislocation densities are synthesized using discrete dislocation dynamics and two strain-based virtual diffraction algorithms, with and without the Stokes-Wilson approximation. Heterogeneity in dislocation density between grains is included in the synthetic database. Instrumental broadening is applied to synthetic profiles to mimic experiments. The data-driven model is a Gaussian process regression which relies on a principal component analysis of normalized profiles. The data-driven predictions of dislocation density are presented in comparison to traditional line profile analysis predictions which assume homogeneous dislocation density. The data-driven model is used to predict the grain-wise variance in dislocation density. The degree of heterogeneity expected due to plastic deformation is also discussed.