Abstract: To improve the accuracy of fault identification,a novel algorithm which could adaptively determine the clustering number of Nonparametric Weighted Feature Extraction (NWFE) and Kernel-based Fuzzy C-means (KFCM) was put forward.The kernel function was combined with fuzzy C-means and the weighted clustering center in NWFE algorithm was adopted to assign different weights for each sample.Besides,clustering evaluation index PBMF was introduced to adaptively determine the optimal clustering number.The performance of algorithm was verified via classical data sets and experiment.The results showed that the disadvantages of classical algorithm were eliminated and the proposed algorithm outperformed the classical algorithm in clustering analysis of rotor failure data sets.

Key words: nonparametric weighted feature extraction, kernel-based fuzzy C-means, adaptive determine the clustering number, rotating machinery, fault diagnosis

摘要： 为提高故障辨识准确率,提出一种专用于故障数据集自适应确定聚类类别数目的非参数加权特征提取(NWFE)和模糊核C-均值(KFCM)相结合的算法。以一个双跨度转子实验台作为实验研究对象,在将核函数与模糊C-均值方法相结合的基础上,采用NWFE算法中加权聚类中心的计算实现了为每个样本分配不同的权值,并引入聚类评价指标PBMF自适应地确定出最佳聚类数目。用Iris经典数据集对算法进行验证表明,所提算法能够克服传统算法中存在的同等对待不同样本特征和完全靠先验知识确定聚类数目的弊端。将该算法应用到转子实验台模拟故障的特征数据集中,进一步表明了其在转子故障数据集聚类分析中的有效性和实用性。

关键词: 非参数特征加权, 模糊核聚类, 自适应聚类数, 旋转机械, 故障诊断