Data-driven prediction of tensile strength for friction stir welded butt joints

Objective The aim is to explore predictive effect of data-driven methods on tensile strength of aluminum alloy friction stir welded butt joints. Methods In this paper, machine learning algorithms such as decision tree, random forest, support vector machine and artificial neural network of Python programming language were utilized to predict tensile strength of friction stir welded butt joints based on test data. After conducting feature analysis on the dataset, 12 key parameters including mass fraction of each element in the material, welding parameters, and plate properties were selected as input features to predict tensile strength. Results The results show that R² of test set reaches 0.92, and R² of five-fold cross-validation test set reaches 0.9. It indicates that the adopted machine learning algorithm performs well in predicting tensile strength of friction stir welded butt joints. Among them, random forest algorithm has the best prediction effect. R² of test set is 0.955, and R² of five-fold cross-validation reaches 0.935. The unbiasedness of this method is verified by extrapolating validation set of test data through test results of 5.82 mm thick 6082-T6 aluminum alloy plates. Conclusion This indicates that for tensile strength prediction of friction stir welded butt joints, data-driven solutions can enhance and optimize performance and efficiency of traditional solutions.