2022 年 94 巻 4 号 p. 187-193
In order to accelerate the calculation of the microstructure formation simulation of multi-component alloys, we have developed a numerical model of the microstructure simulation based on the cellular automaton (CA) method, which applies deep learning to the calculation of the equilibrium concentration at the solid-liquid interface. As a result of verifying the estimation accuracy of the equilibrium concentration calculation by deep learning, it was confirmed that the estimation accuracy is very high and that it is effective as a method of equilibrium concentration calculation. Three-dimensional dendritic growth simulations of Al-5%Si-4%Cu ternary alloy was performed using CA models to which deep learning was applied and in which the CALPHAD method was coupled. As a result, the growth behavior of the dendrite tip was in good agreement between the two models. In addition, the calculation speed per step was about 100 times faster than the CA model in which thermodynamic calculations by the CALPHAD method were coupled. These findings confirm that the alternative to thermodynamic calculation by deep learning is a very effective method for accelerating the calculation of solidification structure simulation of multi-component alloys.