Goranchuk V. Monitoring of VVER-1000 core by methods of neuron-noise diagnostic

The thesis is devoted to the solution of an actual scientific and technical task: enhancement of the NPP operation safety by expanding the VVER-1000 core monitor ability, in particular, by implementing of: reactor internals (RI) and fuel assemblies (FA) vibration monitoring, as well as monitoring of important parameters, such as the local coolant velocity, moderator temperature coefficient of reactivity. Before the beginning of these works the systems of VVER-1000 neutron noise monitoring and diagnostics were not applied at NPPs of Ukraine, and additional research must be conducted for these systems implementation. The analysis and systematization of neutron-noise diagnostics methods as well as selection of the most informative ones of them for application at the VVER-1000 monitoring and diagnostics systems are presented. The analysis of the noise component of signals of self-powered neutron detectors (SPND) and ionization chambers (IC) has been carried out. For the first time the analysis of spectral and other signal parameters of 245 SPND of VVER-1000 was performed during several fuel campaigns with sampling rate 436 Hz that is higher essentially than in previous studies. These aspects allowed to determine additional vibration frequencies. The main characteristic frequencies of disturbing forces that influence the vibrations of RIs and FAs are determined and proved. Vibrational diagnostic models and diagnostic thresholds have been developed for the root mean square deviation of neutron flux, as well as for signs of neutron noise. The estimation of the coolant fluctuations transit time by various methods has been carried out. It was shown, which of methods are the most resistant to global neutron noise, and the ways are determined to improve the estimation of the transit time. The selection of frequency range to estimate the transit time from the phase of cross power spectral density was justified. The models of vibration diagnostics were implemented in the in-core noise diagnostic system. Problem of the SPND locality determination was solved, namely, by determining at what distance the signal of SPND is sensitive to the vibration and other processes that take place in the area of the SPND location and that modulate this signal. The contribution of fuel rods to the signal of SPND was determined in accordance with their location in regard to the SPND. This allowed to localize the FAs vibrations. It was shown the possibility to restore the signal of the failed SPND by means of neural networks. It was demonstrated that the SPND signal recovery is possible with an inaccuracy not more than 2 % under condition of neural network training in wide range of data. This allows to provide the control of energy distribution in the FA, where the failed SPND is located. Modeling was carried out for different SPNDs (differences are within a year of these SPND using as well as in their location in the core) for different power units. The input signals that give a minimal inaccuracy were determined, the comparison of training algorithms was presented. The importance of such input signals selection for the neural network that determine the nature of the output signal most was stressed. The best architecture of the neural network was determined. To calculate the correlation and spectral functions, as well as to create artificial neural networks, appropriate software was developed in the C# programming language. Verification of the developed software with test data was carried out. All developed models and algorithms showed their correct work. Validation of software was carried out at signals from neutron detectors located in units of Ukrainian NPPs.