Abstract
Abstract
The reactor nuclear measurement system is important in a
nuclear power plant. Its main role is to measure the reactor's core
power distribution using detectors and calibrate and provide data on
the core fuel consumption. This study describes the lack of fault
data and the lack of diagnostic methodology research in the
overhauling process and fault diagnosis of the off-heap nuclear
measurement system core card. This core card provides the detectors
with the necessary working conditions. It also collects signals. In
this study, we propose a methodology for the fault diagnosis of the
card through circuit analysis, simulation of functional module
division, fault data generation, and training of a convolutional
neural network diagnostic model. The proposed methodology can
transform the drawings into convenient diagnostic processes and
algorithms based on expert experience. These drawings are difficult
to use in actual overhauling conditions. The corresponding
experimental equipment was designed for practical testing. The
experimental results show that the accuracy of the obtained
diagnostic model for classifying preset faults can reach 99.5%,
indicating that this model can be applied in actual working
conditions. The accuracy of the trained diagnostic model in
classifying 13 kinds of faults in the training set during the actual
test was tested. Results show that the accuracy rate is close to
100%. Moreover, the correction of the model using the real
maintenance data in applying the actual maintenance conditions was
also analyzed. The intelligent diagnostic system that centers on the
fault diagnosis method investigated in this study has been applied
in the pressurized water reactor off-heap nuclear measurement system
digital transformation and upgrading project of Qinshan No. 2
Plant.