Fault Detection and Isolation Based on Abnormal Sub-Regions Using the Improved PCA

Abstract

This paper proposes a novel approach to sensor fault detection and isolation for chemical processes. Based on the improved PCA (principal component analysis), faulty variables are projected onto PRV (principal component related variables) sub-space and OV (other variables) sub-space, respectively according to the correlation between variables and PCs (principal components). Then, by defining the NOR (normal region) and the ANSR (abnormal sub-region), the variables in the two spaces will be classified as normal variables or possibly abnormal variables. This procedure is recursively carried out, and the possibly abnormal variables are finally projected onto the NOR and the ANSR, the only variable retained in the ANSR is considered to be faulty, which completes the fault isolation. This is only used for sensor fault detection and isolation. Application results on a PVC (polyvinylchloride) making process illustrate the effectiveness of the proposed approach.