基于组合核函数的类均值核主元故障诊断方法

Fault Diagnosis Method with Class Mean Kernel Principal Component Analysis Based on Combined Kernel Function

慢漂移是指系统参数从额定值发生偏移的现象,系统正常的慢漂移影响故障诊断的结果。针对单一核函数对慢漂移系统故障误诊率高的问题,提出了基于组合核函数的类均值核主元分析故障诊断方法。首先,使用组合核函数将原始样本数据映射到高维特征空间,求出各类映射数据的类均值矢量;然后在类均值矢量张成的子空间上对类均值矢量进行主元分析;最后,利用建立的类均值核矩阵构建类均值核主元算法。将该方法应用于田纳西-伊斯曼(TE)和蒸馏塔过程,仿真结果显示,该方法提高了故障的检测率,柔性因子的引入满足故障监控模型对灵敏度、鲁棒性的动态平衡要求,具有较好的过程监控性能。

Slow drift is the phenomenon that the system parameters are shifted from the rated value, the normal slow drift of the system affects the result of fault diagnosis. For the problem that the fault misdiagnosis rate of single kernel function for the slow drift system is high, the paper proposed a fault diagnosis method with class mean kernel principal component analysis based on the combined kernel function. Firstly, the original sample data were mapped to the high-dimensional feature space by using the combined kernel function, and the class mean vector of all kinds of mapping data was obtained. Then, the principal component analysis was carried out for the class mean vector on the subspace of the class mean vector. Finally, the class mean kernel principal component algorithm was constructed based on established class mean kernel matrix. The method was applied to the process of Tennessee-Eastman(TE) and distillation column. The simulation results show that the proposed method improves the detection rate of the fault, and the introduction of the flexible factor satisfies the dynamic balance requirement of the fault monitoring model for the sensitivity and robustness, and the method has good performance of process monitoring.