[SY-F13] Symposium F-13

[SY-F13] Design of neural network for thermodynamics data of non-equilibrium multiphase field model

^○Hiroshi Wakameda¹, Sukeharu Nomoto¹, Akinori Yamanaka³, Toshiyuki Koyama² (1.ITOCHU Techno-Solutions Corporation, Japan, 2.Materials Design Innovation Engineering, Nagoya Univ., Japan, 3.Department of Mechanical Systems Engineering, Tokyo University of Agriculture and Technology, Japan)

We construct neural network which estimates Gibbs free energy and chemical potential from temperature and composition of system for non-equilibrium multiphase field model. The mini-batch gradient descent method is selected for training of which data is led from calculation of non-equilibrium multiphase field model using Thermo-Calc thermodynamic database. We achieve highly precision neural network enough to use in non-equilibrium multiphase field model by introducing minimum and maximum data to the mini-batch method.