摘要
将粗糙集理论的属性约简与核主元分析方法结合起来,提出一种基于粗糙核主元分析的故障特征提取方法。该方法首先采用粗糙集理论的属性约简删除与分类无关或关系不大的特征,降低输入特征维数,排除干扰特征的影响,减小了特征提取计算量;然后再采用核主元分析方法进一步提取非线性特征;最后将该方法应用于轴承故障特征提取及故障识别中。应用结果表明,所提出的粗糙核主元分析方法(RKPCA)与传统的KPCA、PCA方法相比,使整个样本集的可分性变大,提高了分类正确率;同时还有效地降低了输入特征维数,提高了分类效率;并且对分类器具有较强的适应性和鲁棒性。
A new approach based on rough kernel principal component analysis (RKPCA) is proposed for aeroengine fault feature extraction, in which the rough set is combined with kernel principal component analysis. The method uses rough set to exclude the features which are not related to the fault, and to reduce the dimension of features and the cost of computation. Then kernel principal component analysis is employed to the obtained subset of features to extract the nonlinear features. Experiments of roll bearing are carried out test the performance of the method. Practical results show that compared with the features extracted by KPCA and PCA, the proposed method increases the separability of data set, performs better recognition ability and is adaptive for various classifiers.
出处
《振动与冲击》
EI
CSCD
北大核心
2008年第3期50-54,59,共6页
Journal of Vibration and Shock
基金
国家自然科学基金资助项目(编号:60672179)
军队重点科研基金资助项目(编号:2003KJ01705)
关键词
故障诊断
特征提取
粗糙集
核主元分析
模式分类
fault diagnosis
feature extraction
rough set
kernel principal component analysis
pattern classification
作者简介
胡金海男，博士，讲师，1978年生
