Prediction of creep properties of Co–10Al–9W superalloys with machine learning

Prediction of creep in superalloys is challenging due to the complex morphology evolution and high-temperature loading conditions. Machine learning shows great potential in capturing the nonlinear relationship between microstructure and creep mechanical properties. The outputs of microstructure characteristics and creep strain–stress of Co–10Al–9W (at.%) alloy simulated by phase-field are used as the characteristic dataset of machine learning, and the convolutional neural network (CNN) and support vector machine (SVM) are used to classify the creep stage. It shows that the accuracy of CNN in the classification of the creep stage is 96.00%. The individual regression algorithms, including Ridge, Lasso, k-nearest neighbor (Knn), decision tree (DTR) and the integrated algorithm random forest (RF), are used to predict the creep strain. It is observed that the random forest algorithm (RF) reaches a high determination coefficient of 0.98, indicating its superior predictive performance. The developed creep stage classification model and creep strain prediction model can be valuable tools for in-depth studies on creep morphology and properties evaluation of superalloys. Moreover, the CNN creep classification model achieves a remarkable accuracy near 100.00%, while the RF creep strain prediction model maintains a favorable R² value of 0.92. This demonstrates the feasibility and robustness of the machine learning combined with phase-field simulation in studying the creep of superalloys.