Machine learning-enabled creep life prediction model for nickel-based single crystal superalloys with consideration of structural shape effects

Significant discrepancies have been observed in the creep life of SX alloy specimens with diverse structural configurations, manifesting an evident “shape effect”. Therefore, it is of great significance to establish a unified creep life prediction model capable of accommodating a broad spectrum of structural forms. In this study, based on the data sets of creep life of specimens with different structural forms, RBF-ANN, GABP-ANN and XGBoost machine learning paradigms were used to predict the creep life of SX alloy specimens, and the prediction effects of the three models were evaluated. The analysis of mean absolute error and determination coefficient shows that the RBF-ANN model has better fitting performance and generalization ability. The contribution of various shape parameters to creep life is ranked as gauge length > cross-sectional parameters > critical cross-sectional area > SCF > perimeter. Furthermore, verified by conducting creep fracture tests under different structures, the results show that the prediction accuracy of RBF-ANN model can be controlled within the range of the two-times scatter band, which shows the effectiveness of the established model. This method provides a new idea for the life evaluation of specimens with different structural shapes and has the potential to be extended to other structural forms and mechanical properties.