Multilevel simultaneous component analysis for fault detection in multicampaign process monitoring: application to on-line high performance liquid chromatography of a continuous process

Abstract

A continuous process is monitored by on-line HPLC in four separate campaigns, ranging in duration from 10 to 104 h. Methods are reported that allow the study of variation over all four campaigns using Multilevel Simultaneous Components Analysis, which separates out the within- and between-campaign variation. In order to obtain control charts, Q- and D-statistics are combined with a within-campaign submodel (Simultaneous Components Analysis) to obtain a single model that is based only on within-campaign variation over all four campaigns.