Physics‐Informed Convolutional Neural Network for Predicting Fatigue Life of Rubber Composites Under Different Strain Ratios

ABSTRACT

To improve fatigue life prediction accuracy for rubber composites under different strain ratios, a physics‐informed convolutional neural network (PI‐CNN) model is developed. Based on uniaxial fatigue tests, an equivalent strain amplitude model incorporating a strain ratio adjustment factor is established. This model serves as a damage indicator to construct a physical model. Subsequently, a CNN model is constructed using the strain ratio and engineering strain amplitude as inputs, with the physical model predicted life as the learning target. The weights of the fully connected layer are further calibrated using experimentally measured fatigue life data, yielding the final PI‐CNN model. Comparison with experimental data under different strain ratios shows that the predicted fatigue lives fall within a scatter factor of 1.5 times the measured values, demonstrating superior accuracy compared with the physical model (scatter factor of 3) or the CNN model (scatter factor of 2).