Research and Development Program in Reactor Diagnostics and Monitoring with Neutron Noise Methods. Stage 13. Final report

This report describes the results obtained during Stage 13 of a long-term research and development program concerning the development of diagnostics and monitoring methods for nuclear reactors. A brief proposal for the continuation of this program in Stage 14 is also given at the end of the report. The program executed in Stage 13 consists of three parts and the work performed in each part is summarized below. 1. Study of criticality, neutron kinetics and neutron noise in molten salt reactors (MSR). Although the original goal of the investigations of the MSR in Stage 13 was to calculate the neutron noise induced by the fluctuations of the fuel temperature, the study, solution and interpretation of the static problem, as well as to define an approximate version of the point kinetic approximation was necessary to perform. As it turned out, these tasks in themselves were more involved, and also very edifying, to solve. Hence, in this report, we confine the study of the reactor physics of the MSR to these items. The solution and analysis of the neutron noise induced by the propagating temperature fluctuations will be included in Stage 14. However, having solved the static problem and having obtained a way of defining the point kinetic approximation will make the task of calculating the neutron noise rather straightforward. 2. Non-invasive two-phase flow regime identification employing simple statistical image characteristics and unsupervised artificial neural networks. A novel on-line non-invasive approach to the classification of BWR two-phase flow regimes was investigated in the current work. The proposed approach receives neutron radiography images (frames) of coolant flow recordings (videos) as its input, preprocesses each image via directly computable statistical operators and, subsequently, uses the extracted statistical characteristics as inputs to an ensemble of two SOMs which are capable of accurately classifying each image into the corresponding flow regime. A variety of simple statistical operators were investigated as inputs, each resulting in a scalar value (characteristic) per frame. SOM training results in clustering of the neighbouring nodes such that each neighbourhood determines a type of patterns that are somehow distinct from the other neighbourhoods/types. Following training, the SOM ensemble is capable of identifying novel patterns into one of the existing classes (flow regimes or transitions between them). The study performed in this work shows that the selected statistical operators and SOM configuration (dimension and size) accomplish accurate identification of the different flow regimes (areas of the 2-D input spaces). It has been observed that, after SOM training, a large proportion of the created clusters contain patterns from a single flow regime. For the remaining clusters, which involve patterns from two neighbouring flow regimes, the a priori probabilities of a pattern belonging to each flow regime have been made use of for averaging and comparing the individual SOM decisions and reaching a single reinforced decision of the ensemble. This method appears to be very effective for real-time flow regime classification, including transitions between flow regimes. 3. Simulation of Feynman- and Rossi-alpha-methods with the code MCNP-PoliMi for studying the performance of reactivity determination in realistic systems. In connection with our research in nuclear safeguards, we have installed the Monte- Carlo code MCNP-PoliMi. As it was mentioned in the previous Stages, MCNP-PoliMi can be used also for the simulation of Feynman- and Rossi-alpha measurements in the traditional set-up consisting of a neutron source, a subcritical core and one or more detectors. Unlike the regular MCNP code, MCNP-PoliMi accounts for the full statistical distribution of the neutron multiplicity in a fission event, which in its turn is the basis of the non-trivial correlations between neutrons in the same chain on which the principle of the Feynman- and Rossi-alpha methods is based. It is also possible to simulate the Cf-252 method for measurement of reactivity