Prediction of hardness or yield strength for ODS steels based on machine learning

Oxide dispersion strengthened (ODS) steel has emerged as a highly promising cladding materials for Generation IV nuclear reactors due to its exceptional mechanical properties and remarkable resistance to irradiation, corrosion, and oxidation. In this study, the matrix grain morphology, dispersion morphology, and phases of oxide particles in eight ODS steels were studied by scanning transmission electron microscopy (STEM), transmission electron microscopy (TEM), and high-resolution transmission electron microscopy (HRTEM). The effect of grain refinement in Al-free ODS steels is better than that in Al-added and Zr-added ODS steels. In Al-added ODS steels, the co-addition of Ti and Zr elements could improve the dispersion morphology of nano-sized particles. In this study, more than 500 data from ODS steels were collected, and 420 items were used for machine learning (ML) modeling. Several ML models were developed to evaluate the predictive performance of the dataset of hardness and yield strength. The results indicate that two XGBoost (XGB) models, which show the lowest mean absolute error (MAE) values and the highest R2 values among the six ML models, have the best predictive performance. Therefore, the two XGB models were selected to predict the hardness and yield strength of ODS steels. The independent variables included chemical compositions, test conditions, and microstructural descriptors. A high linear correlation exists between Zr and Ti. Regarding chemical composition, Y2O3 has the most significant effect on hardness and yield strength. The predicted values of hardness & yield strength are in good agreement with the corresponding experimental values. The two generalized ML models show the potential for accurate prediction of hardness & yield strength in ODS steels, thereby providing a valuable theoretical framework for the design and optimization of novel ODS steels.