基于改进增量LE的压缩机故障特征提取方法.pdfVIP

学兔兔 第34卷 第4期 仪 器 仪 表 学 报 Vo1．34 No．4 2013年4月 Chinese Journal of Scientific Instrument Apr．2013 基于改进增量LE的压缩机故障特征提取方法水 许庆诚，胡建中 (东南大学机械工程学院 南京 211189) 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-CHIve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。 关键词：cam加权距离；拉普拉斯特征影射算法；流形学习；增量 中图分类号：TP206 文献标识码：A 国家标准学科分类代码：510．40 Fault feature extraction method for compressor based on improved incremental Laplacian eigenmap algorithm Xu Qingcheng，Hu Jianzhong (School ofMechanical Engineering，Southeast University，Nanjing 211189，China) Abstract：Increasing the accuracy of fault feature extraction f0r 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 m