A thermo-hydraulic model for the water-cooled stator bars in large generators is presented. By calculation and deduction， the formula for reflecting the dynamic model is obtained， and the neural network of radial basis function（RBF） is used to identify the model parameters and calculate the transient time of the temperature occurring in water-cooled stator bars from load-1 to load-2. The transient time is used as a tool to judge whether the generator is working in normal condition or in thermo-fault condition. An example based on a 600 MW synchronous generator is used to prove the correctness of this technique. The distinct advantage of this technique is that the judgment result is not affected by the delay time due to the temperature signal.