Detection, isolation and reconstruction of faulty sensors using principal component analysis

Abstract

A strategy based on principal component analysis (PCA) is presented for detection, identification and reconstruction of faulty sensors. In this strategy, sensor fault detection is carried out by using multivariate statistics, faulty sensors are isolated using principal component score contributions and reconstruction of faulty sensors is accomplished through the analysis of fault direction vector. The performance of the strategy is evaluated by applying to a closed-loop controlled CSTR system. The simulation results demonstrate the ability of the strategy for detection, identification and reconstruction of single and multiple faulty sensors.