A Physics-Informed Neural Network (PINN) model base on deep learning has been proposed for predicting low-cycle fatigue (LCF) and thermo-mechanical fatigue (TMF) life. By analyzing the LCF and TMF data of compacted graphite iron (CGI), characteristic parameters were identified that can simultaneously represent both types of fatigue, achieving a unification of the parameters for the two fatigue life models. The incorporation of fatigue life physical information as a constraint in the loss function of the deep neural network enabled accurate predictions of LCF and TMF for CGI under small-sample conditions. Comparative analysis results indicated that the deep learning-based PINN model outperformed traditional machine learning models in terms of prediction accuracy. Additionally, comparisons with traditional LCF and TMF life prediction models showed that the deep learning-based PINN model achieves high prediction accuracy while possessing generalization and extrapolation capabilities unattainable by traditional models. These results demonstrate that the PINN model exhibits high accuracy and versatility.