Temperature dependent strength prediction of nanotwinned nickel-based single crystal superalloys by integrating atomistic simulation and machine learning

Accurate prediction of the mechanical properties is of great importance for the optimal design and application of nanostructured metallic materials. Machine learning (ML) offers an efficient alternative, rapidly acquiring the ability to understand and predict material properties after sufficient training. Herein, this paper proposes a novel strategy by integrating molecular dynamics (MD) simulations, active learning sampling, and back-propagation neural network (BPNN) models. Firstly, the combined impact of twin thickness and ambient temperature on the mechanical behaviours is subjected to comprehensive analysis to guarantee the reliability of MD dataset. Subsequently, the BPNN model is developed and compared with support vector machine (SVM) model and random forest (RF) model. It is demonstrated that the BPNN model is capable of accurately predicting the mechanical properties of nickel-based single crystal superalloys, with a Pearson correlation coefficient exceeding 0.99. Notably, the model effectively captures the anomalous strength softening effect dominated by detwinning at extremely fine twin thicknesses. Overall, this work provides a reliable and efficient framework to facilitate the development of high-performance materials by taking account various nanostructures and service conditions.