Concurrent fault diagnosis of small pressurized water reactors based on long-short term memory networks

Abstract

The control systems of small pressurized water reactors (SPWR) with complex structures, compact layouts, and variable operating environment may be involved in various types of signal and concurrent faults. Concurrent faults can be taken as two or more single faults occurring simultaneously, which usually cause much larger damage to the system and are more difficult to be diagnosed than single faults. However, their fault diagnosis methods are rarely studied because of the numerous fault types and tremendous diagnostic difficulty. This paper explores the concurrent fault diagnosis method for sensors and actuators in SPWR control systems. An intelligent current fault diagnosis model is developed using long short-term memory network with the training and test datasets generated based on a fault simulation platform of the target SPWR. The test results show that both single and concurrent faults of the SPWR can be diagnosed rapidly in an average of 1.06 s after their occurrence with the classification and diagnosis accuracies reaching up to 96.61% and 97.27%, respectively. Moreover, by injecting different noise signals on the faulty dataset for training and validation, it is shown that the proposed LSTM network has strong noise immunity. This demonstrate the excellent diagnosis accuracy and efficiency of the model under both single and concurrent fault conditions. This study provides valuable guidance for the accuracy diagnosis of complex concurrent faults of nuclear power plants.