摘要
利用信号的时频域统计指标提取故障特征的方法忽略了故障信号的时变非线性的特性以及局部突变的特点。针对这一问题,提出了小波变换的故障特征提取方法。通过对信号进行小波变换,分析故障信号的局部性质,利用小波系数构造高维特征空间,分别利用等距映射(Isometric Feature Mapping,ISOMAP)、局部线性嵌入(locally linear embedding,LLE)和拉普拉斯特征映射(Laplacian Eigenmap,LE)等算法对高维特征空间进行降维,提取低维流形,表征故障特征。经水基动力无杆抽油机故障模拟试验台进行验证,分析比较三种方法得到的低维流形,可知利用小波系数构造流形高维空间经ISOMAP方法获得的低维流形图形状最易区分,数据也具有较高的可分性,能有效分析识别抽油机故障。
The method of fault signal feature extraction in time domain statistical index ignores the characteristics of fault signals of the nonlinear time-varying characteristics and local mutation characteristics. The signal is transformed through the wavelet; the local properties of fault signals are analyzed; the high dimensional feature space is constructed by wavelet coefficients; the dimensionality reduction of high dimensional feature space is carried out by LLE,LE and ISOMAP; low dimensional manifold is extracted and the fault feature is characterized. The experimental data obtained from the simulation experiment platform of failure for rod-less pumping with water as power verifies that the shape of the low dimensional manifold map obtained by using the wavelet coefficients is the most easily distinguishable in the low dimensional manifold graph obtained by ISOMAP. The data also have high separability and are suitable for analysis and identifying of pumping unit fault by comparing and analyzing the low dimensional manifold graph obtained from the three method.
出处
《北京信息科技大学学报（自然科学版）》
2015年第5期24-29,共6页
Journal of Beijing Information Science and Technology University
基金
北京市教委科研计划重点项目(KZ201311232036)
关键词
小波系数
流形降维
抽油机
特征提取
wavelet coefficients
manifold dimension reduction
pumping unit
feature extraction
