On establishment of novel constitutive model for directionally solidified nickel-based superalloys utilizing machine learning methods

To enhance the accuracy of mechanical simulation in the directional solidification process of turbine blades for heavy-duty gas turbines, a new constitutive model that employs machine learning methods was developed. This model incorporates incremental learning and transfer learning, thus improves the predictive accuracy and generalization performance. To account for the anisotropy of the directionally solidified alloy, a deformation direction parameter is added to the model, enabling prediction of the stress-strain relationship of the alloy under different deformation directions. The predictive capabilities of both models are evaluated using correlation coefficient (R), average relative error (δ), and value of relative error (RE). Compared to the traditional model, the machine learning constitutive model achieves higher prediction accuracy and better generalization performance. This offers a new approach for the establishment of flow constitutive models for other directionally solidified and single-crystal superalloys.