Omsktransmash (Russia)

The basis of the self-sustaining chain reaction of nuclear fission lies in the probabilistic nature of fundamental physical processes. Reactivity, as a special physical property of the medium or technical system, is accepted as a measure of the dynamics of these processes (state identifier). This paper proposes to classify reactivity as a class of random functions with zero mathematical expectation in a stationary critical state. Fluctuations relative to the zero value are perceived as the result of external factors causing an instantaneous response of the reactor core (RC) of reactor units (RUs) operating at a set power level for the majority of the operational time. This definition is sufficient to develop methods of technical diagnostics based on perturbations of the RC reactivity. The reactor core is considered as a stochastic object maintained within the normative field of its design and operational parameters. According to archival data of reactivity “measurements” for certain states of the RC, it is possible to correlate fragments of stochastic time series that serve as identifiers of these states. A set of such fragments (tests) constitutes a library of nuclear power plant operational experience. This work addresses the effects of external Poisson perturbations of reactivity and the classification of resulting neutron density modulations within the framework of a one-group kinetics model. The reflection of random static and dynamic operational factors in the probabilistic characteristics of neutron density through reactivity is interpreted as evidence of generation of defectiveness in the RC of RUs. The deterministic component of neutron density is described sufficiently accurately by kinetic equations and corresponds to the trend of the mathematical expectation of reactivity relative to zero. A procedure is proposed for processing data from neutron flux monitoring equipment (NFME) and in-core monitoring systems (ICMS), including forming a library of test perturbations based on archival data and analyzing responses using the Kolmogorov–Smirnov goodness-of-fit criterion between empirical distribution functions of the current empirical data sample and those calculated by kinetic equations based on the perturbation library. It is asserted that the proposed procedure serves as a means of identifying the level of defectiveness of the RU reactor core and can also be used as an event simulator for deep learning of a classifying neural network. The results relate to advances in neutron noise diagnostics.