Interpretable machine learning-based analysis of mechanical properties of extruded Mg-Al-Zn-Mn-Ca-Y alloys

Abstract

In this study, the mechanical properties of as-extruded Mg-Al-Zn-Mn-Ca-Y alloys were quantitatively investigated with respect to alloying elements, extrusion temperature, microstructure and texture through interpretable machine learning (IML). To overcome the lack of data, two methods were devised to augment the existing dataset by 39 times using the mean and standard deviation of the measured data. Artificial neural networks predicted room-temperature tensile properties with an accuracy ranging from 0.842 to 0.997 based on R2 using 12 predictors for a total of 1179 data points. Shapley additive explanation identified that Al and Mn are the key determinants for strength and elongation, respectively. Partial dependence plots investigated the interaction of all features to understand the quantitative correlation between features. This IML approach revealed that texture, solid solution and secondary particles are related to the main strengthening mechanism of as-extruded Mg alloys. These results can provide insights into the utilization of IML approach to predict material properties and describe key variables for designing lightweight structural metals.