摘要
提出一种基于核规范变量分析(KCVA)的非线性过程故障诊断方法.该方法使用核函数完成非线性空间到高维线性空间的映射,避免了高维空间中的数据处理和非线性映射函数的使用.在线性空间中使用规范变量分析(CVA)来辨识状态空间模型,从数据中提取状态信息.3个监测量(Tr2,T2s,Q)用来进行故障检测,同时使用贡献图分离故障变量,并判断故障原因.在CSTR系统上的仿真结果表明,KCVA方法比主元分析法(PCA)和CVA方法能更灵敏地检测到故障的发生,更有效地监控过程变化.
A new method based on kernel canonical variate analysis（KCVA） is proposed for nonlinear process fault diagnosis. This method uses the kernel function to map the nonlinear space into a linear high dimension space. The application of the kernel function can avoid nonlinear mapping function and data processing in high dimension spaces. Canonical variate analysis（CVA） is applied to identify a state space model in linear space and state information is extracted. Three monitoring statistics Tr^2, Ts^2 and Q are built for fault detection. Contribution plot is used to isolate faulty variables and locate fault source. The simulation results on CSTR system indicate that KCVA can detect fault more easily than principal component analysis and canonical variate analysis.
出处
《控制与决策》
EI
CSCD
北大核心
2006年第10期1109-1113,共5页
Control and Decision
基金
国家863计划项目(2004AA412050)
关键词
核规范变量分析
故障诊断
贡献图
非线性过程
Kernel canonical variate analysis
Fault diagnosis
Contribution plot
Nonlinear process
