.png){kind=icon}
.png){kind=icon}
Return
ML Model Optimization-Selection and GFA Prediction for Binary Alloys
.png){kind=icon}
PRE
AI
DOI：10.1021/acs.cgd.1c01519.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Metallic glasses are widely used in many fields because of their excellent physical and chemical properties; however, the absence of a simultaneously accurate and efficient prediction of the glass-forming ability (GFA) has hindered their industrialization. Here, we compile data sets using 31 alloy serials and 179 high-GFA alloys collected from peer-reviewed publications and, in the prediction data set, an absent but necessary feature is completed by a Regression Random Forest. Six machine-learning models are evaluated on training and test data sets by using different dimensional input features. With the best accuracies of 1 and 0.9934 on the two data sets, the Gradient Boosting Decision Tree model, trained with seven-dimensional features, is employed to predict the GFA of binary alloys. It is found that, for the GFA of binary alloys, the atomic fraction of solvent c(1) ranks first, followed by the liquidus temperature of alloys Tliq, and then the ratio of the atomic weight W and radii R. This study suggests that the optimization of hyperparameters, the number and independence of input features, and the proper model play active roles in attaining good predictions of GFA and that further investigation is still necessary for a comprehensive understanding of the GFA of alloys.
Keywords:
GLASS-FORMING ABILITY
BULK METALLIC GLASSES
CLASSIFICATION
DIAGNOSIS