Fast prediction of the quasi phase equilibrium in phase field model for multicomponent alloys based on machine learning method

Abstract

The phase field model for multicomponent alloys are usually coupled with the thermodynamic database. But it will take much time to solve the quasi phase equilibrium equations. In order to reduce such time and keep the enough accuracy, we develop a new model to predict the quasi phase equilibrium based on the machine learning method. As an example, the quasi phase equilibrium during the isothermal solidification of Al-Cu-Mg alloy is studied in detail. A neural network model with 3 inputs, 4 outputs and a hidden layer of 150 nodes is constructed. The “training data” are prepared by solving the quasi phase equilibrium equations with least square method. The neural network model is trained by different amount of data set, which can fully cover the ranges of all the variables. The accuracy and performance of the neural network model are discussed in detail. Its high accuracy and fast speed demonstrated that this will be a convenient method to acquire the quasi phase equilibrium data in phase field model for multicomponent alloys.