Data-driven phase-field analysis of static recrystallization in an aluminum alloy

Grain growth during static recrystallization has been modeled by the multi-phase-field method. However, its predictive accuracy is insufficient because grain boundary properties such as grain boundary energy have not been accurately identified, and no methods exist for measuring the stored energy distribution of the deformed matrix. In this study, we developed a Bayesian data assimilation technique based on an ensemble square root filter to estimate the stored energy distribution in each deformed matrix using the in-situ observation of growth of recrystallized grain by scanning electron microscopy–electron backscatter diffraction (EBSD). The results demonstrated that the stored energy distribution, which is difficult to measure experimentally, can be estimated only from the time-series data of the grain distribution. The developed method expresses the multi-phase-field simulation results as a probability density function and calculates its temporal evolution, which allows for evaluating the uncertainty of the estimated results for stored energy distribution and grain boundary position from the standard deviation of the probability density function. The developed technique was proven to estimate the stored energy distribution even in cases where an error of approximately 0.5 μm was included in the location of the grain boundaries measured from the in-situ observations. This study opens new pathways for data-driven phase-field simulation in which both in-situ EBSD observation and phase-field simulation of the static recrystallization are effectively utilized.