Interpretable Predicting Creep Rupture Life of Superalloys: Enhanced by Domain-Specific Knowledge

Evaluating and understanding the effect of manufacturing processes on the creep performance in superalloys poses a significant challenge due to the intricate composition involved. This study presents a machine-learning strategy capable of evaluating the effect of the heat treatment process on the creep performance of superalloys and predicting creep rupture life with high accuracy. This approach integrates classification and regression models with domain-specific knowledge. The physical constraints lead to significantly enhanced prediction accuracy of the classification and regression models. Moreover, the heat treatment process is evaluated as the most important descriptor by integrating machine learning with superalloy creep theory. The heat treatment design of Waspaloy alloy is used as the experimental validation. The improved heat treatment leads to a significant enhancement in creep performance (5.5 times higher than the previous study). The research provides novel insights for enhancing the precision of predicting creep rupture life in superalloys, with the potential to broaden its applicability to the study of the effects of heat treatment processes on other properties. Furthermore, it offers auxiliary support for the utilization of machine learning in the design of heat treatment processes of superalloys. The workflow in the above figure is used to show the framework and how to use it to achieve HT optimization with enhanced creep rupture life. Figure (a) presents the model process, i.e., label HT, evaluate HT, and creep rupture prediction. Figure (b) shows the experimental validation for HT optimization based on the three processes.image