Microstructural feature-based physics-informed neural network for creep residual life prediction of P91 steel

Abstract

Creep residual life prediction of materials at elevated temperature is an important topic in the field of structural integrity. Traditional creep residual life prediction methods only consider mechanical parameters (e.g. stress, strain, temperature), while the microstructural features are rarely mentioned, reducing the prediction accuracy. In this work, taking the P91 steel as an example, a microstructural feature-based physics-informed neural network (PINN) for predicting creep residual life was developed by integrating the microstructural characteristics and mechanical parameters. The influence of microstructural features on the prediction results was discussed, and the prediction results of the proposed model and some conventional machine learning methods were compared. The effect of the strain data on creep residual life prediction results was included. The results indicated that the introduction of the microstructural evolution mechanisms (i.e. coarsening of precipitations and subgrain growth) could enhance the creep residual life prediction capacity of the proposed PINN model. The proposed PINN model outperforms the aforementioned traditional machine learning methods in predicting the creep residual life of materials. This model also exhibits excellent prediction performance without incorporating the creep strain data as an input feature, demonstrating the generalization ability and robustness of the proposed model.