基于多元统计信号处理的间歇过程监控方法研究  

Monitoring method based on multivariate statistical signal processing for batch process

作　　者：陈国金[1] 梁军[1] 刘育明[1] 钱积新[1] 

机构地区：[1]浙江大学工业控制技术国家重点实验室浙江大学系统工程研究所,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2006年第1期5-9,共5页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60574047);教育部高等学校博士学科点专项基金资助项目(20050335018)

摘　　要：针对化工生产中日益增多的间歇过程,提出了一种基于多元统计信号处理的过程监控方法,其主要思想为将过程信息空间划分为由盲源信号描述的信号子空间、过程主元描述的信号子空间和残差信号子空间,随后对各个信号子空间构造过程统计量或分类器提取信号特征进行过程监控,该方法避免了传统多元统计过程控制(mult-ivariate statistical process contro,lMSPC)需假设过程特征信号服从正态分布的前提.将本方法与传统MSPC方法的性能进行了对比,并在仿真中给出了对比研究结果.通过对间歇过程的仿真研究表明,该方法不仅能够有效地检测出故障,而且有利于故障的分离和定位,从而说明该方法不仅是有效的,而且其性能优于仅能检测故障的传统MSPC过程监控方法.A new method based on multivariate statistical signal processing was presented for a large number of batch process in modern chemical industry, which divides the process information space into three subspace： subspace described by blind source signals, subspace described by principal components and subspace of residual information. Then corresponding control limits or operating features classifiers were built to monitor the process. The feasibility of this method was investigated by applying it to a batch process, and the results show that this method can：1）not only detect process faults in time, but also confirm the source of faults exactly; 2）extract or model process features much more simply in each subspace than in whole process information space and 3）overcome the limitation of the assumption in conventional multivariate statistical process control （MSPC） that the separated latent variables must be subjected to normal probability distribution.

关 键 词：盲源信号提取 主元分析 过程监控 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]