Machine learning-based prediction and generation model for creep rupture time of Nickel-based alloys

Abstract

Predicting the creep rupture time of Nickel-based alloys has been one of the central topics in materials science. This work is devoted to develop a comprehensive strategy based on machine learning for data generation and prediction. Our objective is to improve significantly the accuracy and effectiveness of the rupture time prediction. The two different prediction models, residual learning combining liner regression model with nonlinear regression model, ensemble learning combining with genetic algorithm, are employed. Both of them are shown to achieve high accuracy on the training dataset and testing dataset. To overcome the limited availability of experimental data, a novel generation model is designed based on generative adversarial network. The new model is capable of generating a batch of high-quality synthetic dataset for expanding the scale of the training dataset. In particular, the best results obtained by the model for three metrics R2, RMSE, MAE are 0.9971, 0.1259, 0.0395. The proportion of the augmented synthetic dataset is also discussed. The promising results demonstrate that incorporating the appropriate proportion of synthetic dataset can significantly improve the accuracy and generalization ability of the prediction models.