摘要
提出一种新的针对KPCA模型的故障识别方法——贡献率图法。该方法是在微分贡献率图和核函数导数的基础上提出来的,它采用统计量T2和SPE对每个变量的偏导数来度量每个变量对统计量T2和SPE的贡献率。和基于数据重构法的KPCA故障识别方法相比,该方法不需要任何迭代近似计算和数据的重构,计算量小且可避免重构产生的误差对识别结果的影响。通过在某型涡扇发动机故障检测与诊断中的应用表明,该方法比基于数据重构法的故障变量识别准确率更高,再结合发动机故障机理分析,便可准确地确诊故障,从而大为缩短故障定位及排故的时间,预防重大事故的发生。
A novel approach of fault identification of KPCA is presented, which is called the contribution plots method. The method is built on the basis of differential contribution plots and the derivative of kernel functions, and measures each variable of contribution to statistics T2 and SPE by theirs partial derivative. Compared with the fault identification method based on data reconstruction, it needs no any approximate computation and avoids the effection of reconstruction errors. The practical applications in monitoring certain type of Turbine-Fan Engine show that the presented method possesses higher accuracy than the method based on data reconstruction in identifying faulty variables, further diagnoses the fault correctly by combining the fault mechanism analysis of aeroengine, and greatly shortens the time of locating faults and elimination.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2008年第3期572-576,共5页
Systems Engineering and Electronics
基金
军队重点科研基金资助课题(2003KJ01705)
关键词
航空发动机
故障检测与故障诊断
故障识别
贡献率图
核主元分析
多元统计分析
aeroengine
fault detection and fault diagnosis
fault identification
contribution plot
kernel principal component analysis
multivariate statistical analysis
