Abstract: A fault diagnosis model which integrates the expert system into neural network (NN) was established. Through the study of the characteristic about production rule, the binary rule with ‘And’ was discussed for binary coding used in NN with high effectiveness. And the forward NN algorithm was used for reasoning, which avoided the conflict of knowledge. At the same time, the temporal NN results were saved for tracking after reasoning. The results indicate that the model with low misdiagnosis rate can meet the actual needs of fault diagnosis.