Thermocouple sensor fault detection using Auto-Associative Kernel Regression and Generalized Likelihood Ratio Test

In nuclear power plants, sensor health condition monitoring is necessary to ensure the correctness of measurements. Out-of-calibration sensor data can direct to take inappropriate action of system monitoring and controlling application. 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, an online sensor fault detection technique is proposed using Auto-Associative Kernel Regression (AAKR) and Generalized Likelihood Ratio Test. The AAKR method is used to approximate the data and the GLRT is applied as a metric to detect the faulty sensor on the residual space, the deviation of approximated data from the original. This paper claims that the AAKR-GLRT based fault detection method is better than the PCA-Q-statistic. The method is validated by the real data from the Fast Breeder Test Reactor (FBTR).