Physics-informed neural network-based life prediction for copper alloy under low-cycle fatigue and creep-fatigue interaction

Copper alloys exhibit significant fatigue life differences under low-cycle fatigue (LCF) and creep-fatigue interaction (CFI), which poses a significant challenge to the reliability of fatigue life assessments of rocket engine thrust chambers. Consequently, it is imperative to develop an effective methodology for the life assessment of the thrust chambers. A Physics-Informed Neural Network (PINN) based on a modified energy-based method is proposed for the life prediction of copper alloys under both LCF and CFI. The proposed PINN model embeds the modified energy-based model which innovatively separates the plastic strain energy term and the creep strain energy term, and further constrains the network output by establishing a hierarchical penalty mechanism within the loss function. All 88 experimental data samples used were derived from LCF and CFI tests on QCr0.8 copper alloy. Compared to physical models, such as the Plastic Strain Energy Density (PSED) model and the Inelastic Strain Energy Density (ISED) model, the proposed PINN model demonstrates superior prediction accuracy. All its prediction points fall within 2-factor error bands, and approximately 85 % within 1.5-factor error bands. Compared to machine learning models, such as Random Forest (RF), Support Vector Regression (SVR), and Deep Neural Networks (DNN), the proposed PINN model exhibits enhanced prediction stability and generalization capability, as well as superior physical interpretability. This approach offers a novel solution for predicting the fatigue life of copper alloy under LCF and CFI.