Nuclear power plant sensor fault detection using singular value decomposition-based method

In a nuclear power plant, periodic sensor calibration is necessary to ensure the correctness of measurements. Those sensors which have gone out of calibration can lead to malfunction of the plant, possibly causing a loss in revenue or damage to equipment. Continuous sensor status monitoring is desirable to assure smooth running of the plant and reduce maintenance costs associated with unnecessary manual sensor calibrations. In this paper, a method is proposed to detect and identify any degradation of sensor performance. The validation process consists of two steps: (i) residual generation and (ii) fault detection by residual evaluation. Singular value decomposition (SVD) and Euclidean distance (ED) methods are used to generate the residual and evaluate the fault on the residual space, respectively. This paper claims that SVD-based fault detection method is better than the well-known principal component analysis-based method. The method is validated using data from fast breeder test reactor.