Sensor Calibration and Monitoring Using Autoassociative Neural Networks

The approach to instrument surveillance and calibration verification (ISCV) through plant wide monitoring proposed in this paper is the use of an autoassociative neural network (AANN) which will utilize digitized data presently available in the Safety Parameter Display computer system from Florida Power Corporations Crystal River #3 nuclear power plant. An autoassociative neural network is one in which the outputs are trained to emulate the inputs over an appropriate dynamic range. The relationships between the different variables are embedded in the weights by the training process. As a result, the output can be a correct version of an input pattern that has been distorted by noise, missing data, or non-linearities. Plant variables that have some degree of coherence with each other constitute the inputs to the network. Once the network has been trained with normal operational data using a robust training procedure, it has been shown to successfully monitor the selected plant variables to detect sensor drift or failure by simply comparing the network inputs with the outputs using a sequential probability ratio test (SPRT). The AANN method of monitoring many variables not only indicates that there is a sensor failure, it clearly indicates the signal channel in which the signal error has occurred.