Physics‐Informed Neural Network for Fatigue Life Prediction of Filled Natural Rubber Under Varying Hardness and Ambient Temperature

Accurate fatigue life prediction of rubber materials under varying temperatures and hardness levels remains challenging due to the limitations of traditional physical models and the lack of interpretability in purely data‐driven approaches. This study proposes a physics‐informed neural network (PINN) framework that integrates physical constraints—including the rubber fatigue power‐law criterion and monotonic strain‐life and temperature‐life relationships—into the loss function. The model takes ambient temperature, Green–Lagrange strain peak, and material hardness as inputs to predict fatigue life. Comparative results show that the PINN outperforms both a semi‐empirical physical model and a conventional BP neural network, achieving effective prediction performance with predictions within a twofold scatter band and delivering 100% physically consistent predictions. A SHAP‐based sensitivity analysis reveals that the Green–Lagrange strain peak is the dominant factor influencing fatigue life, followed by hardness, with temperature having the weakest effect. The proposed framework offers an effective and interpretable solution for rubber fatigue life prediction under complex service conditions.