Detection of measurement errors in nuclear power plants
Résumé
This paper presents a method for detecting and isolating sensor failures like biases or drifts. The algorithm is based upon analytic redundancy methods and uses a mathematical model of the process to compute estimates of measured quantities. These are compared with the measurements, so that the resulting residuals contain the sensor faults of interest. Statistical tests eliminate wrong data and produce an alarm if a faulty instrument is detected. Some application results are given in this paper for the instrumentation monitoring of a steam generator used in nuclear power plants.