Abstract: Among the existing fatigue life prediction models for natural rubber (NR), the physical models face challenges such as complex modeling and a narrow scope of application, whereas data-driven models are often constrained by small sample sizes, resulting in low prediction accuracy for fatigue life. To address these issues, a physically-data-driven modeling method for predicting the fatigue life of NR is proposed. The model uses logarithmic strain amplitude and strain ratio as input variables, and the errors between the experimentally measured fatigue life and the life predicted by the physical model as the output variable. Based on uniaxial fatigue test data of NR, a physical model for fatigue life prediction based on the crack initiation method, a data-driven model using support vector machines, and a physically-data-driven model are developed, respectively. By comparing the prediction accuracy of these models against the measured fatigue life of natural rubber, it is found that the physically-data-driven model demonstrated superior prediction accuracy, with predicted lives distributed within 1.5 times the scatter band of the measured lives. The study demonstrates that the physically-data-driven model can more accurately predict the fatigue life of NR, providing a theoretical foundation for the fatigue-resistant design of NR components.

Key words: natural rubber, fatigue life, uniaxial fatigue, physical model, data-driven