For the problem that multi-fault diagnosis can not be solved by the extension neural network, a problem model is built, and the multi-fault diagnosis problem is transformed into the clustering problem for multi-attribute samples. ENN2 is improved from two faces of the model structure and learning algorithm, and the multi-fault diagnosis method based on the improved ENN2 clustering algorithm is proposed with the analysis of parameters and time complexity. The proposed method is verified by an engineering instance. The results show that the method can resolve the offline multi-fault diagnosis problem, and the obtained diagnosis model can also be applied to online fault monitoring, so it has a wide application prospect.