Fault Detection and Diagnosis of a Nuclear Power Plant Using Artificial Neural Networks

Abstract

Fault detection and diagnosis have always been an important aspect of nuclear power plant system design as early detection of failure can prevent system breakdown or serious disaster. In this article an approach based on neural networks and mathematical models for detecting and diagnosing instrument failures in the pressurized water reactor (PWR) of the H. B. Robinson nuclear plant is presented. Multilayer neural networks are used at the first level for identification of plant parameters; at the second level for distinguishing parameter variations and uncertainties from possible faults; and as a pattern recognizer in the third level for the detection of faulty instruments. The design approach is able to simultaneously classify single and multiple anomalies such as sensor and actuator failures under plant parameter uncertainties. Simulation results presented reveal that it is feasible to use artificial neural networks to improve the operating characteristics of the nuclear power plant.