Robust statistical process monitoring

Abstract Principal component analysis (PCA) is a key step to carrying out multivariate statistical process monitoring. Due to the sensitive nature of classical PCA, one or two outliers will cause misleading results. In this paper, a robust PCA via a Hybrid Projection Pursuit (HPP) approach is proposed. Incorporation of this robust PCA into our previously developed data driven strategy, for statistical process monitoring, will mean the whole procedure will be resistant to outliers and thus robust. The performance of the proposed approach is demonstrated by simulation studies on a simple flowsheet example.