Prediction of Microsegregation Behavior in Fe-based Alloys Based on Machine Learning

Abstract

Synopsis :A prediction method for microsegregation in Fe-based alloys was developed based on an approach of machine learning called Deep Learning. A set of model and algorithm of Deep Learning suitable for description of microsegregation was constructed by employing training data obtained by one-dimensional finite difference calculations for interdendritic microsegregation. It is shown that the developed method enables accurate prediction of the microsegregation behavior in Fe-based binary and ternary alloys with the solute atoms of C, Si, Mn, P and S. The present results demonstrate that Deep Learning offers a promising way of constructing an easy-to-use approach for prediction of microsegregation with high accuracy. Importantly, it is expected that the present method can be extended to describe effects of microstructural processes on microsegregation behavior.

Dependence of Eutectic Fraction on Inclination Angle of Columnar Dendrite Structures in Al-3 mass% Cu Alloy Analyzed by Phase-field Simulation

“…The development of a quantitative phase-field model (QPFM) allows for accurate description and prediction of the formation process of solidification microstructures [9][10][11][12][13][14] including PDAS and SDAS, and it has been applied to the analyses of columnar dendrite structures. [15][16][17][18][19][20][21] QPFM for two-sided diffusion can be utilized to investigate microsegregation, 22,23) and it has been demonstrated that microsegregation of columnar and equiaxed structures in Al-Cu and Fe-Mn alloys can be reproduced with high accuracy by QPFM. 24) In the previous study, the microsegregation of columnar dendrites growing in the direction of heat flow was investigated.…”

Estimation of Thermodynamic and Interfacial Parameters of Metallic Materials by Molecular Dynamics Simulations

“…Therefore, the solid-liquid interfacial energy can be estimated from the slope of the relation in eq. (17). From the slope of the fitted line in Fig.…”

“…In addition to conventional MD and MC simulations, recent progress in collaboration with techniques from data science including computational screening, 14,15) neural network computation, 16) machine learning 17) and data assimilation 18,19) is opening a new window of atomistic simulations. As techniques based on the data science permeate computational metallurgy, it becomes increasingly important to ensure the feasibility of simulation results from a metallurgical viewpoint.…”