Diagnosing core local flow blockages in a VVER-1000/446 reactor using ex-core detectors and neural networks

Abstract

According to the Defense in Depth (DiD) concept, it is crucially important to detect and mitigate the consequences of accidents in Nuclear Power Plants (NPPs) in the early stages. Due to the nature of local blockages, the timely and suitable detection of melting accidents is a challenging task. Therefore, developing early detection and accurate mechanisms to prevent nuclear accidents as well as mitigation of their consequences are an essential step to be taken into account. In the present study, an add-on system for detecting and monitoring local melting accidents which in the early stages do not lead to SCRAM, is developed using neural networks and neutron ex-core detectors. The results of the core MCNPX model in terms of local flow blockage location in different sub-channels are used as the network training inputs. The core is divided into four, eight, and twenty-four angular segmentations, along with seven radial and ten axial partitions. More than 50,000 local blockage accidents are simulated in the MCNPX code, and 500 networks are trained for achieving an optimal network topology with 90.1% accuracy. The results of trained network shows that the proposed monitoring system is capable of diagnosing local flow blockages in less than one-tenth of a fuel rod height. The sensitivity and noise analysis results can represent the reliability of the designed system considering blockage distances from the detectors. The obtained results demonstrate that by implementing the proposed add-on system, local blockage accidents can be diagnosed with high accuracy.