基于潜隐变量自相关性子空间划分的故障检测策略  

Fault Detection Strategy Based on Dividing Autocorrelation of Latent Variables

作　　者：张成 郭青秀 李元 ZHANG Chong;GUO Qing-xiu;LI Yuan(Shenyang University of Chemical Technology,Shenyang 110142 China)

机构地区：[1]沈阳化工大学技术过程故障诊断与安全性研究中心,辽宁沈阳110142

出　　处：《沈阳化工大学学报》2020年第4期369-376,共8页Journal of Shenyang University of Chemical Technology

基　　金：国家自然科学基金(61673279);国家自然科学基金重大项目(61490701);辽宁省科学事业公益研究基金(2016001006)。

摘　　要：针对主元分析(principal component analysis,PCA)中潜隐变量自相关性对故障检测的影响,提出一种基于潜隐变量自相关性子空间划分的故障检测策略(fault detection strategy based on dividing autocorrelation of latent variables,FDDA).首先,应用PCA将输入空间通过线性变换分解为主元子空间(principal component subspace,PCS)和残差子空间(residual subspace,RS).其次,依据潜隐变量自相关性的强弱,将上述两个子空间分别进行二次划分.综上,原始输入空间依据方差和自相关性信息被划分成4个子空间,并利用不同的监控指标进行故障检测.子空间划分方法既可以提取输入变量间的相关性,又可以捕获潜隐变量自相关性.通过4个子空间的联合监控,可以有效地解决动态过程的故障检测问题,具有较高的故障检测率.将FDDA方法应用在TE过程和半导体蚀刻过程,并与PCA、动态PCA(dynamic PCA,DPCA)、kNN等进行对比分析,仿真实验结果验证了FDDA方法的有效性.Aiming at the influence of latent variable autocorrelation in fault detection is principal component analysis( PCA),f ault detection strategy based on dividing autocorrelation of scores( FDDA) is proposed.Firstly, input space is decomposed into principal component subspace( PCS) and residual subspace( RS) through linear transformation using principal component analysis. Then, the above subspaces are divided again separately according to the autocorrelation of score variables. In summary, the original input space is divided into four subspaces based on variance and autocorrelation information and different monitoring indicators are used to detect faults. The FDDA method can extract the correlation between the input variables and capture the autocorrelation of the score variables. Through the joint monitoring of the four subspaces, the fault detection problem of dynamic process can be effectively solved and the fault detection rate is high. FDDA method is applied to TE process and semiconductor etching process,and compared with PCA,DPCA,kNN and so on. The simulation results verify the effectiveness of FDDA method.

关 键 词：自相关性 主元分析 k近邻规则 TE过程 半导体蚀刻过程 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]