Physics-Informed Neural Network Model for Predicting the Fatigue Life of Natural Rubber Under Ambient Temperature Effects

Abstract

This study develops a physics-informed neural network (PINN) model combining physical principles and data-driven efficiency to predict natural rubber (NR) fatigue life under varying temperatures. The proposed model utilizes fatigue damage parameters and environmental temperature as input variables, while the relative error between the measured fatigue life and the fatigue life predicted by the physical model serves as the output variable. Using experimental fatigue test data under varying environmental temperatures, the predictive performance of the physical model, BP neural network model, and PINN model was evaluated. The results demonstrate that the PINN model outperforms existing predictive approaches, with its predictions consistently falling within 1.5 times the dispersion band of the measured values. Furthermore, a partial sensitivity analysis was conducted based on the connection weights of the PINN model and the Garson equation, quantifying the relative influence of the input variables on the predicted fatigue life.