Neural Network based Dynamic Threshold Generator for Sensor Failure Detection in a Three Tank Interacting Level Process

摘要

This paper presents neural network based dynamic threshold generator for sensor failure detection in a three tank interacting level process. The Fault Detection and Identification scheme performs the tasks of failure Detection and Identification by continuously monitoring the outputs of the sensors and the estimates of the states. Estimation errors are observed and the decision functions are formed from the estimation errors. Decision function values make fault alarms under sensor failure conditions. If the normal value of this decision function is fixed at zero, even a slight deviation, which is not necessarily because of sensor failure, can also make fault alarm. This false fault alarm is avoided by fixing threshold values for the decision functions instead of zero. The threshold values for these decision functions are to be carefully selected to avoid false fault alarm. A method of dynamic threshold generation using neural network based on set point value is proposed in this paper. The results show that the threshold values are best selected by this method so that no false fault alarm is reported.