Gibbs Free-energy Prediction Method for Iron-base Alloy Materials Based on Deep Learning*

Abstract

In order to speed up the development of new iron-base alloy materials and reduce the consumption of time and resources caused by a large number of experiments, a prediction method for Gibbs free energy of iron-base alloy materials was proposed based on the theory of material genetic engineering. Firstly, the collected data were preprocessed by splicing, filling, normalization and one-hot coding to adapt to the training of the model. Then, based on the DeepFM model, a fusion model based on Factorization Machine (FM), bitwise self-attention mechanism and Bi-directional Long Short-term Memory Network (Bi-LSTM) was proposed to predict the Gibbs free energy of iron-base alloy materials. It can not only extract the low-order and high-order features of the data effectively, but also the weight coefficients of each data feature can be reasonably optimized and the correlation between the data can be fully considered. The comparative experimental results show that the Gibbs free energy prediction method based on deep learning has a good prediction effect. It provides a new method to predict the Gibbs free energy of iron-base alloys.