基于改进增量LE的压缩机故障特征提取方法  

Fault feature extraction method for compressor based on improved incremental Laplacian eigenmap algorithm

作　　者：许庆诚[1] 胡建中[1] 

机构地区：[1]东南大学机械工程学院,南京211189

出　　处：《仪器仪表学报》2013年第4期791-796,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(51075069;51075070)资助项目

摘　　要：提高离心压缩机故障特征提取精度对于后续故障诊断具有重要意义。针对传统增量LE算法处理精度差的问题,分析了参数t对传统增量LE算法特征提取精度的影响,提出了一种改进的增量LE算法。该方法将传统的增量LE算法与cam加权距离相结合,在新增样本点投影过程中通过cam加权距离选取邻域,采用热核形式计算新增样本的权值,由局部保持特性,通过新增样本的近邻来重构其低维嵌入。S-curve仿真数据以及离心压缩机故障数据分析表明:相比于传统的增量LE方法,改进的增量LE方法能有效提高新增故障样本特征提取的精度。Increasing the accuracy of fault feature extraction for centrifugal compressor has great importance to the subsequent fault diagnosis. Aiming at the problem of insufficient processing precision of traditional incremental LE al- gorithm, the effect of parameter t on the feature extraction accuracy of traditional incremental LE algorithm is ana- lyzed, and an improved incremental LE algorithm is proposed. The method combines the traditional incremental LE algorithm and cam weighted distance. A neighborhood is selected with cam weighted distance in the projection process of the new sample points and the heat kernel is adopted to calculate the weights of the new samples. According to lo- cality preserving characteristic,the low-dimensional embedding is reconstructed with the neighbor of new fault sam- ples. S-curve simulation data and centrifugal compressor failure data analyses show that the improved incremental LE method can effectively improve the accuracy of feature extraction for new fault samples compared with traditional in- cremental LE method.

关 键 词：cam加权距离 拉普拉斯特征影射算法 流形学习 增量 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]