<p>Traditional machine learning models for fatigue life prediction suffer from a lack of physical interpretability. To overcome this limitation, a physics-informed neural network (PINN) framework is developed by integrating multi-source information. In this framework, the continuum damage mechanics (CDM) model associated with additive manufacturing (AM) process parameters (AM-CDM) and the concept of fatigue limit are embedded into the neural network architecture. Specifically, the AM-CDM model is encoded as a regularization term in the loss function, while the fatigue limit is incorporated into the activation function, thereby establishing a data-physics dual-driven fatigue life prediction paradigm. This design ensures that the model parameters remain within physically admissible solution spaces during training, effectively suppressing non-physical solutions caused by sparse data. The findings indicate that the developed PINN achieves superior performance compared with standard neural networks in predictive accuracy and physical consistency. Furthermore, controlled variable analysis confirms that the PINN reliably captures the nonlinear relationships between AM process parameters and fatigue life, offering a physically interpretable approach for evaluating the fatigue performance of AM components under complex service conditions.</p>