Prediction of creep rupture life of ODS steels based on machine learning

Abstract

Creep rupture life is a key material parameter for service life and mechanical properties of ODS steels. Therefore, accurately predicting the creep rupture life of ODS steels holds significant importance. This study aims to build quantitative models to predict the creep rupture life of ODS steels by ML techniques. Before the construction of models, the SFS and RF methods were used for input selection. Seven features were selected as inputs and used to predict creep rupture life. The ML algorithms including LR, SVR, MLP, KRR, RFR, and XGBoost are employed, among which XGBoost performs best in predicting the creep rupture life of ODS steels with high accuracy (RMSE of 0.78 h and 0.88 h, MAE of 0.62 h and 0.71 h, and R2 of 0.96 and 0.91 for training and testing sets, respectively). Feature importance coefficients of the XGBoost model were used to show the magnitude of the effect of descriptors on creep fracture life. For the chemical composition, the parameters that have a great influence on the creep rupture life are Ti, Cr, W and Y2O3 with the values of importance coefficients of 0.154, 0.04, 0.039 and 0.034, respectively. For the heat treatment process, the parameters that have a great influence on the creep rupture life are ST and FT with the values of importance coefficients of 0.127 and 0.025, respectively. For the test conditions, the parameters that have a great influence on the creep rupture life are CR and CS with the values of importance coefficients of 0.342 and 0.041, respectively. The findings of the feature importance of the XGBoost model are similar to those of the RFR model. This study developed an effective ML model to predict the creep rupture life of ODS steels, providing a strategy to predict the creep rupture life of ODS steels and significant insights as additional guidelines for the research and development of novel type ODS alloys with high performance.