Abstract: In order to improve the accuracy of predicting the degree of the fuel defect, two serial artificial neural networks were established to determine the fuel defect state and its degree. After changing the mass of tramp uranium, increasing data fluctuations, changing single rod power, and taking fewer characteristic nuclides as input vector, performance tests and analysis were performed to evaluate the suitability of the neural network. The same analysis was performed to analyze the second network, which was trained to determine the degree of the fuel defect. Under the condition that the mass of tramp uranium is less than 0.5 g, the data fluctuation is within 30%, and the single rod power is between 77 kW and 120 kW, the first artificial neural network can better determine whether the cladding is damaged. The second neural network has a higher accuracy for the five levels of damage. Compared with the traditional iodine isotope ratio method, the neural network method is faster and more accurate. The results show that the artificial neural network can be used to predict whether the reactor fuel cladding is damaged and the degree of fuel defect.