关键词: 密度峰值聚类, 马氏距离, 聚类中心, 故障诊断

Abstract: Aiming at the problem of density center selection inconvenient and low clustering accuracy in Clustering by Fast Search and Find of Density Peaks (CFSFDP) algorithm,a clustering algorithm based on Mahalanobis distance was proposed to find density peaks by automatically searching,which had Introduced Mahalanobis distance into the distance measurement to improve the clustering accuracy.The determination parameter γ of clustering center was proposed to achieve acquisition of clustering center automatically.Traditional CFSFDP algorithm,improved CFSFDP algorithm,K-means clustering and fuzzy C-means clustering analysis were carried out respectively by experimental data of aero-engine rotor simulation vibration signal.The results showed that the improved CFSFDP algorithm could improve the clustering accuracy well.Compared with K-means clustering and fuzzy C-means clustering,the improved CFSFDP had great advantages on clustering accuracy,classification and recognition of fault features.

Key words: density peak clustering, Mahalanobis distance, cluster center, fault diagnosis