Harsh operating conditions subject the water-lubricated stern bearings of a ship to a non-uniform stress field, leading to frequent boundary lubrication between the bearing and shaft. Poor lubrication results in local temperature rise due to friction heat, leading to heat aging or even carbonization of the bearings. This study proposes a novel universal monitoring approach based on the internal temperature of water-lubricated bearings (WLBs). Radial basis function neural networks (RBFNN) were established to evaluate the surface temperature and cooling statuses of the bearing surface during rubbing, in which simulation results obtained from a temperature model were used as a training set. The accuracy of the simulation results and prediction model was verified through experimental data. The results showed that the maximum error of predicted contact surface temperatures is only 0.84 °C, 0.02 °C, and 1.00 °C in MAE, MAPE, and RMSE, respectively. The accuracy of cooling conditions identification can be achieved at 91.7 % with 600 s of continuous temperature data. The scheme proposed in this study can effectively predict thermal failures of water-lubricated bearings without requiring extensive experimentation, which promotes the development of intelligent water-lubricated bearings.