Quantitative analysis of mechanical properties associated with aging treatment and microstructure in Mg-Al-Zn alloys through machine learning

The present study proposes a methodology for predicting the mechanical properties of AZ61 and AZ91 alloys associated with microstructure, texture and aging parameters and estimating predictor importance. For this, we investigate quantitative correlations between microstructure, texture and mechanical properties of aged AZ61 and AZ91 rods through machine learning. This regression analysis focuses on the precipitation behavior of Mg17Al12 as the main second phase of Mg-Al-Zn alloys with respect to aging conditions. To simplify data generation, only SEM images were used to quantify the features of discontinuous and continuous precipitates. To overcome the lack of data and make the most of the measured data, we devised a method to extend the existing dataset by a factor of 9 using the mean and standard deviation of the measured data. Artificial neural networks predicted tensile and compressive yield strengths and resultant yield asymmetry with a high accuracy of over 98% using 11 predictors for a total of 288 datasets. Decision tree learning quantitatively assessed the importance of predictors in determining the mechanical properties of aged AZ61 and AZ91 rods.