Abstract: This paper introduce data field model to FCM, and proposed an improved FCM algorithm.On one hand, the improved algorithm uses potential values of fault points in the data field to identify noise points and kicks them out;on the other hand, the improved algorithm utilizes the natural aggregation properties presented by fault points in the data field to select initial clustering centers.The simulation experiment results showed that the improved algorithm has a good performance in fault diagnosis process.The improved algorithm could make up the defects of traditional FCM algorithm, and improve the efficiency and accuracy of clustering.