False alarm reducing in PCA method for sensor fault detection in a nuclear power plant

Abstract

Principal component analysis (PCA) is applied for fault detection of sensors in a nuclear power plant (NPP) in this paper. In order to reduce the false alarms of T2 and Q statistics during fault detection, two different methods are further proposed in this paper. One is statistics-based method which generates second confidence limits for T2 and Q statistics, and then false alarms are reduced based on the second confidence limit during test. The other is iteration-based method which reduces false alarms during modeling. Measurements beyond the first confidence limit of T2 or Q statistics are successively removed from the training data through iteration process. Finally, sensor measurements from a real NPP are acquired to train and test the proposed methods. On one hand, simulation results show that the proposed PCA model is capable of detecting the faulty sensors no matter with small or major failures. On the other hand, simulations also indicate that the PCA model combined with statistics-based and iteration-based methods simultaneously makes more contribution to the timeliness and effectiveness of sensor fault detection compared with the PCA model only with statistics-based or iteration-based method.