Textured water-lubricated bearings in motorized spindles face significant friction and wear during low-speed start-stop processes, making accurate prediction of the friction coefficient crucial. However, traditional experimental methods are limited by time and cost, while purely data-driven machine learning models struggle with limited dataset sizes and poor generalization. To address these issues, this study proposes an artificial neural network model integrated with prior knowledge (ANN-PK), which combines experimental data with prior knowledge to predict the friction coefficient. This approach achieves better generalization and reduces the requirements for dataset size. First, friction tests were conducted on textured UHMWPE-copper ball pairs, generating a small dataset under varying conditions. Then, the relationships between the friction coefficient and individual parameters were incorporated as prior knowledge into the artificial neural network model's loss function. Finally, the effectiveness of the prior knowledge was validated through experiments, ablation studies, and comparisons with purely data-driven models such as support vector regression (SVR), random forests (RF), and Gaussian process regression (GPR). Results show that the proposed ANN-PK outperforms purely data-driven models in predicting friction coefficients and generalization, even with a small dataset.