A machine-learning approach to predict creep properties of Cr–Mo steel with time-temperature parameters

Traditional alloy design which requires deep understanding of the Process–Structure–Property relationship, involves numerous tests and iterations. It is not cost-effective and very time-consuming. Current study explored the idea of using machine learning to build quantitative models to predict creep life of Cr–Mo steel with the help of convention creep study wisdom. Three time–temperature parameters, namely Larson-Miller parameter (LMP), Manson-Haferd parameter (MHP) and Manson-Succop parameter (MSP), were used as the target feature instead of creep life. Moreover, Pearson correlation coefficient (PCC) and Spearman correlation coefficient (SCC) were used to discuss the feature importance of creep life. The results demonstrated that the prediction accuracy was effectively improved by transforming creep life into LMP, MHP and MSP. In the case of using MSP as target feature, random forest (RF) had the best accuracy and robustness in all regression models. Feature importance ranking provided new insights for domain experts to analyze the main influencing features to creep property of Cr–Mo steel. This study has successfully proved the feasibility of combining machine learning and time–temperature parameters in predicting creep life and provided a basic idea and workflow for researchers to predict creep life of steel.