Machine Learning Aided Prediction and Design for the Mechanical Properties of Magnesium Alloys

Magnesium (Mg) alloy is an extremely excellent structural material, and the improvement of its mechanical properties has attracted extensive attention. However, there is still lack of a systematical study that considers various types of Mg alloys and versatile mechanical performance. In this paper, extensive and in-depth machine learning (ML) research is carried out on the four mechanical properties of Mg alloys, including ultimate tensile strength (UTS), elongation (EL), yield strength (YS) and hardness (HV). By collecting experimental data in the published literatures, the mechanical properties database of Mg alloy was established. The experimental parameters, basic properties and thermodynamic parameters of alloys are used as descriptors to construct the ML models thus predicting the four mechanical properties of Mg alloys. In the total dataset and the cast and extrusion subsets, the optimum prediction models for UTS, YS, and HV achieved high accuracy, with the values of coefficients of determination above 0.8 and a maximum value of 0.93 in the test set, although the prediction model of EL was not satisfactory. In addition, the machine learning model is interpreted by using the Shapley additive explanations (SHAP) model, and the critical values of important descriptors affecting the mechanical properties are obtained, especially the range of feature values that can improve YS and EL at the same time, which is of great significance for the development of the mechanical properties in the design of high-performance Mg alloys. Graphical Abstract