Prediction of phases and mechanical properties of magnesium-based high-entropy alloys using machine learning

Abstract

Objectives

To predict phases and mechanical properties of Mg-Al-Cu-Mn-Zn alloys and to validate the results.

Methods

In this study, 29 predictor features of the alloys were examined based on dataset drawn from relevant publications. The correlation of selected predictor features with mechanical properties of Mg-Al-Cu-Mn-Zn alloys were evaluated. New features specific to vehicle and aerospace applications were used. Feature selection schemes involving four machine learning (ML) classifiers that included artificial neural networks (ANN), linear discriminant analysis (LDA), random forest regression (RF) and k-nearest neighbours (k-NN) were adopted. Tensile test was carried out based on ASTM E8 standard.

Results

Results of correlation of features showed that specific strengths and specific modulus of the alloys were strongly and positively correlated with composition of alloying elements but strongly and negatively correlated with composition of magnesium. The results also revealed that homogenization temperatures and time were weakly correlated with the mechanical properties and phases while electronegativity difference and VEC had significant positive correlation. ANN was the best performing classifier followed by k-NN, LDA, and lastly RF with prediction accuracy on test data of 98.7%, 98.1%, 97.9% and 97.8%, respectively. The validity and applicability of the model was tested with three magnesium-based alloys: Mg-80-Al-10-Cu-5-Mn-5-Zn-0, Mg-80-Al-5-Cu-5-Mn-5-Zn-5 and Mg-91.2-Al-8.3-Cu-0-Mn-0.15-Zn-0.35 and compared with findings in literature. The model had higher prediction accuracies compared to previous ML models used on magnesium alloys. The model was then used to predict phases in the Mg-89.43-Al-8.16-Cu-0.34-Mn-0.25-Zn-1.81 alloy and it accurately predicted presence of Mg₁₇Al₁₂, Mg₂Si, MgZn and MgZn₂. Results of simulation in MatCalc version 6.04 also verified presence of the phases. The phases were further confirmed through SEM/EDS analysis.

Conclusions.

Dominant strengthening phases were Mg₁₇Al₁₂, Mg₂Si, MgZn and MgZn_2. Predicted yield strength, ultimate tensile strength and Young’s modulus were within the range of experimental results. Specific strengths and specific modulus were also within the range.