基于KPCA-LSSVM的发动机PT燃油系统故障诊断

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。

Fault diagnosis of PT fuel system based on KPCA-LSSVM

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.