基于KPCA-LSSVM的发动机PT燃油系统故障诊断  

Fault diagnosis of PT fuel system based on KPCA-LSSVM

作　　者：王东[1] 王新晴[1] 闫凤国[2] 杨成松[1] 章青[3] WANG Dong WANG Xinqing YAN Fengguo YANG Chengsong ZHANG Qing(College of Field Engineering, PLA Univ. of Sci. ＆ Tech., Nanjing 210007,China College of Defense Engineering,PLA Univ. of Sci. ＆ Tech., Nanjing 210007, China School of Mechanical Engineering, Tianjin University, Tianjin 300072,China)

机构地区：[1]解放军理工大学野战工程学院,江苏南京210007 [2]解放军理工大学国防工程学院,江苏南京210007 [3]天津大学机械工程学院,天津300072

出　　处：《解放军理工大学学报（自然科学版）》2016年第5期499-504,共6页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(41401518);国家科技重大专项资助项目(2011ZX05056-003-0)

摘　　要：为有效解决PT燃油系统进油油路堵塞、滤清器泄漏、喷油器油路堵塞等多种典型故障诊断问题,提出了基于核主元分析(KPCA)和最小二乘支持向量机(LSSVM)的故障识别方法。首先计算油压信号的时域特征集,然后采用KPCA对原始多维初始特征向量进行特征提取,最后将经过KPCA提取的主特征向量输入经多种群遗传算法(MPGA)优化的LSSVM中实现故障类型的识别。实验结果表明,KPCA提取的主特征向量有效表达了原始故障的特征信息,相比于传统的BP神经网络和未经参数优选的LSSVM等分类模型,基于KPCA-LSSVM的故障识别方法速度更快、分类准确率更高。To solve such problems as plugging of the inlet, leaking of the filter, and plugging of the fuel injector in the fault diagnosis of pressure time（PT） fuel system, a fault diagnosis method based on kernel principal components analysis and least squares support vector machine （KPCA-LSSVM） was proposed. Firstly, the time domain feature set of the oil pressure signal were calculated. Secondly, the original multidimensional feature vectors were extracted by KPCA. Finally, the final features were put into LSSVM optimized by multiple population genetic algorithm（MPGA） to identify faults. The experimental results show that the main characteristics extracted by KPCA effectively express the original information of fault feature set. Compared with the traditional classification model, the proposed fault diagnosis method is faster and has higher classification accuracy.

关 键 词：PT燃油系统 核主元分析 最小二乘支持向量机 多种群遗传算法 故障诊断 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]