Abstract: Traditional physical models based on crack initiation are limited in predicting rubber fatigue life across multiple temperatures. Data-driven models suffer from small sample sizes, resulting in insufficient prediction accuracy for engineering needs. To solve these problems, ta multi-temperature fatigue life prediction method for rubber materials using a parameter fusion model is proposed. It combines the theoretical benefits of physical models with the efficiency of data-driven models for accurate prediction in different temperature environments. A BP neural network model is built with ambient temperature and engineering strain peak as input variables and physical model predicted fatigue life as the output. The weights between the hidden and output layers of the BP neural network are optimized using actual fatigue life data to create the parameter fusion model. This model is used to predict the fatigue life of rubber materials. Based on fatigue test data for rubber materials under multiple environmental temperatures, the prediction accuracy of the physical model, BP neural network model, and parameter fusion model are compared. It is found that under the multi-environment-temperature conditions, the predicted life from the parameter fusion model is within 1.5 times the dispersion of the actual life. The results show that the parameter fusion model overcomes the limitations of existing models in predicting rubber fatigue life under multiple environmental temperatures. It offers excellent prediction accuracy and an effective method for evaluating the fatigue life of rubber materials.

Key words: ambient temperature, rubber materials, fatigue life, prediction model, fatigue test