Subspace approach to fault detectability of PCA monitoring model

Principal component analysis (PCA) is an effective multivariate statistical process monitoring approach and its most significant advantage is that no precise process model is needed. Nevertheless, PCA-based process monitoring methods show difficulties in systematically analyzing the issue of fault detectability. Based on the fault description method of fault subspace, sufficient and necessary conditions of fault detectability in the principal component (PC) space and residual space are presented. The concept of critical fault magnitude is introduced and used to analyze the detection behavior of faults. The acquired results are then illustrated and verified by fault detection examples of a double-effective evaporator process.