Machine learning-based forward and inverse designs for prediction and optimization of fracture toughness of aluminum alloy

Utilization of machine learning framework to design aluminum alloy with high fracture toughness is increasing. Nonetheless, before such model can be applied, the generalizability of the model becomes imperative thus it can give a better performance not only in the present, but also against the future data. In this work, shallow and deep machine learning techniques represented by support vector regression (SVR), k-nearest neighbors (KNN), extreme gradient boosting (XGBoost) and artificial neural network (ANN) was deployed to predict the fracture toughness of various aluminum alloys in the scheme of MLDS framework. Our study reveals that the highest prediction accuracy can be obtained for XGBoost technique with R2 score, RMSE, and MAPE values were found to be 90.6 %, 2.57, and 7.0 %, respectively with robust k-fold validation value of 90.1 +/- 1.5. The superior performance of Xgboost due to its capability handling non-linear regression problems with a small amount of data. The forward properties to compositions (P2C) and reverse compositions to properties (C2P) models exhibited good accuracies in predicting the fracture toughness as illustrated by the error values lower than the machine learning design system (MLDS) error criteria of 5 %. The XGBoost model feasibility to predict fracture toughness for various aluminum alloys of Alcoa 7055-T7751, AA 7055-T7751, 2024-T852, 5083-O 8090-T8151 was demonstrated as well as used to search aluminum alloy compositions with high fracture toughness. The Pearson correlation coefficient (PCC) and feature Importance results were able to reveal the effects of the alloying elements to the fracture toughness of the alloy. These results reveal the potential application of our models as a prescreening tool in selecting material for alloying compounds.