Fault Diagnosis Based on Fuzzy C-means Algorithm of the Optimal Number of Clusters and Probabilistic Neural Network

Abstract

Fault diagnosis is essential for the reliable, safe, and efficient operation of the plant and for maintaining quality of the products in industrial system. This paper presents an ensemble fault diagnosis algorithm based on fuzzy c-means algorithm (FCM) with the Optimal Number of Clusters (ONC) and probabilistic neural network (PNN), called FCM-ONC-PNN. In clustering methods, the estimation of the optimal number of clusters is significant for subsequent analysis. As a simple clustering method, FCM has been widely discussed and applied in pattern recognition and machine learning, but FCM could not guarantee unique clustering result because initial cluster number is chosen randomly. As the number of clusters is randomly chosen, the iterative amount is large and the result of the classification is unstable. In this paper, firstly subtractive clustering is proposed to find the optimal number of clusters and the clustering results of the FCM are compared with random initialization method, and then PNN is used to classify the clustering data of FCM. The experiments show that the modified initial cluster number of FCM algorithm can improve the speed, and reduce the iterative amount. At the same time, FCM-ONC-PNN approach can make classification more stable and have higher precision.