摘要
Principal component analysis (PCA) has already been widely applied to process monitoring.However,PCA model is only a special case of probabilistic principal component analysis (PPCA) model and the latter itself is a special case of factor analysis (FA)model.Compared with PCA and PPCA models,FA model has less restriction and can do better to reveal essential features of the data.A FA model was built by the expectation maximum (EM)algorithm,and was introduced into industrial process monitoring.Monitoring indices based on FA were proposed to monitor the process factors space and residual space,respectively.A method was presented to select the number of factors by means of the property that the explanation ratio for the process information was convergent with the increasing number of factors.A contrastive study with PCA and PPCA was carried out in the Tennessee Eastman (TE) process,which showed the FA-based method’s superiority either in missed detection rate or in the sensitivity for fault.
Principal component analysis (PCA) has already been widely applied to process monitoring. However, PCA model is only a special case of probabilistic principal component analysis (PPCA) model and the latter itself is a special case of factor analysis (FA) model. Compared with PCA and PPCA models, FA model has less restriction and can do better to reveal essential features of the data. A FA model was built by the expectation maximum (EM) algorithm, and was introduced into industrial process monitoring. Monitoring indices based on FA were proposed to monitor the process factors space and residual space, respectively. A method was presented to select the number of factors by means of the property that the explanation ratio for the process information was convergent with the increasing number of factors. A contrastive study with PCA and PPCA was carried out in the Tennessee Eastman (TE) process, which showed the FA-based method 's superiority either in missed detection rate or in the sensitivity for fault.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2007年第4期970-974,共5页
CIESC Journal
基金
新世纪优秀人才支持计划项目(NCET-05-0485)~~
关键词
因子分析
监控指标
主元分析
TE过程
factor analysis
monitoring indices
principal component analysis
TE process