Aiming at the problems of insufficient accuracy and weak generalization ability of traditional neural networks in fatigue life prediction of bridge cranes, a fatigue life prediction method based on physical-informed neural network (PINN) was proposed. Based on the fatigue crack propagation mechanism of the crane structure, a data model was constructed using bi-directional long short-term memory network, which extracted features from time-series load data and performed equivalent transformation. Subsequently, a physical model was established in combination with fracture mechanics theory to depict the evolution law of fatigue damage. The data model and the physical model were deeply integrated, and the fused dynamic stress data served as the input to the physical neural network, while the fatigue life was set as the output. A penalty term that satisfied the differential equation of the Paris model was used as the physical loss, which was combined with the network data loss to construct a minimized loss function. Precise prediction of the fatigue life of bridge cranes was achieved by optimizing this loss function. Taking the DQ40 kg-1.8 m-1.3 m small general-purpose bridge crane as an example, by comparing the measured data and predicted data of the fatigue life of the crane during normal operation, the feasibility of the proposed method was verified. The results showed that, compared with the convolutional neural network model, support vector regression model, and K-nearest neighbor model, the fatigue life prediction fitting accuracy of the PINN model increased by 19%, 24.9%, and 26% respectively. The research results provide a new strategy for predicting the fatigue life of cranes.