2017 年 103 巻 12 号 p. 720-729
An approach of machine learning called Deep Learning is utilized for construction of a prediction method of microsegregation behavior in Fe-based binary alloys with solute atoms of C, Si, Mn and P. Training data for the machine learning are obtained by quantitative phase-field simulations for directional solidification. Therefore, effects of microstructural evolutions on the microsegregation behavior are taken into account in the present method. Importantly, this method can be coupled with a macrosegregation model. The simulation result of the macrosegregation model is quite different from those obtained by a conventional macrosegregation model with the Scheil model and a model with a prediction method constructed from the training data of one-dimensional finite difference calculations for the microsegregation. This fact highlights the importance of accurate description of microsegregation behavior in prediction of macrosegregation.