Age-hardening behavior guided by the multi-objective evolutionary algorithm and machine learning

In the present work, multi-objective evolutionary (MOE) algorithm and machine learning (ML) techniques were employed to predict the age-hardening behavior of aluminum (Al) alloys in a wide range of processing conditions. For this purpose, data containing hardness, information on alloy compositions, and aging conditions (aging time and temperature) were extracted from previous works that reported the age-hardening of Al-Cu-Mg base alloys. Accordingly, 1591 cases were collected for various alloy compositions and processing conditions. Composition features (140) generated based on the alloy composition and element properties (atomic weight, electronegativity, etc.), and processing features (time and temperature) were subjected to a preprocessing using the MOE algorithm to reduce the number of features and use those which highly influence the hardness. MOE-processed features and counterpart hardness values are then employed in the learning process using various ML algorithms, including decision tree (DT), deep learning (DL), linear general model (GM), gradient boosted trees (GBT), random forest (RF), and support vector machine (SVM). The results show that the MOE algorithm's leveraging with ML learning processes can be successfully used to refine the features and build accurate ML predictive models compared to those created using other feature selection and preprocessing methods. In addition, the learning results showed that the predictive model built using the ensemble GBT algorithm exhibits the best performance among all models built based on other ML algorithms, where a relative error of 3.5% was recorded for the GBT-based model, and it could reproduce the experimental aging behavior of Al alloy.