Inversely optimized design of Al-Mg-Si alloys using machine learning methods

In this study, the previously reported inverse design strategy for simultaneously optimizing two properties of copper alloys was expanded to concurrently optimize three properties of Al-Mg-Si alloys. Following this strategy, 180 input features based on alloy compositions and corresponding physicochemical parameters were constructed. After feature screening, these input features were refined to 5, 6, and 4 key features for machine learning (ML) models of ultimate tensile strength (UTS), yield strength (YS) and elongation (EL), respectively. Utilizing these key features as inputs, SVR ML models were developed for UTS, YS, and EL. Subsequently, the ML models were employed to predict the properties, and these predictions were assessed using a function MOEI, which reflects the combination of all three properties based on Bayesian principles. The combined properties of the optimized alloy evaluated in this study exceeded the Pareto frontier formed by the initially collected alloys. Experimental analysis highlighted the significant contribution of β'' precipitates to the outstanding combined property of the designed alloy. This study showcases the successful extension of the inverse design strategy to concurrently optimize three properties of Al-Mg-Si alloys, offering valuable insights for future alloy design and development.