The goal of the dissertation is the development of an automated diagnostic system that will promote the increasing of operational reliability and safety of water-cooled reactors of nuclear power units (NPU) of nuclear power plants (NPP) and first of all of mostly responsible and thermally stressed components of reactor, namely fuel elements of nuclear core (NC) due to timely recognition of dangerous heat transfer regimes on fuel element surface by noise characteristics of operation parameters. The existing computer-aided manufacturing control systems (CAMCS) and specialized systems of NPU equipment monitoring in general don't check heat exchange regimes by applying recognition of noise signals of neutron flux gauges and dynamic pressure signals. Although spectral analysis of some NPU process parameters (e.g., the signals of vibroaccelerometers for monitoring a reactor element vibra-tion) is designed in these systems, diagnostic decision-making is laid upon monitoring system operator because the system is disable to insure automated acceptance of such decisions. Computers are used in these diagnostic complexes only to store measurement data making them convenient for an operator-diagnostician. In existing approach to the analysis of extremely permissible levels of monitored signals of tech-nological parameters the probabilistic nature of these signals caused by stochastic character of physical processes (neutronic, heat-and-hydraulic etc.), that lay in the base of NPU technology, is ignored. An at-tempt to use only small number of integrated process features (Raise frequency and others) is a serious shortage as well. The last doesn't give the possibility to get recognition reliability acceptable for practice needs. In a proposed dissertation the creation of an expert system on the base of an intellectual, i.e. got by methods of artificial intellect, software efficient to solve in principal the problem of the recognition of random process, which must be identified in operating reactor equipment diagnosing, is foreseen. Until now in domestic and worldwide practice the means of reliable diagnosing of mentioned damages in fact are absent. Such up-to-date diagnosing and monitoring systems as ALLY, manufactured by Westinghouse Electric Corp., and pwVDN, manufactured by ABB (USA), Siemens's diagnostic sys-tem (Germany), COSS system (Japan), ALARM (Great Britain) and others don't solve the problem. Enumerated shortages of the best existing systems touch these ones as well. The principal idea of the work consists in that noise signals at the outlet of main NPP technologi-cal parameter gauges (neutron flux, dynamic pressure, coolant flow rate) contain important information on technical state of the equipment (actually in existing NPP CAMCS it is lost). As a result efficient algo-rithms of random process identification that after respective spectral transformation are considered as multidimensional random vectors were developed. Automated classification of these vectors in the devel-oped algorithms is realized on the base of probability function, especially of Bayes classifier, and deci-sion-making functions. Application of Bayes classifier is based on construction of multidimensional dis-tribution of probabilities in feature space of corresponding dimension, with the help of which random vectors-realizations of corresponding images that are subjected to automated classification are described. In their turn multidimensional distributions of probabilities represent standard of classes subjected to rec-ognition. In the occasions when the establishing of these standards of classes is complicated by objective factors (impossibility of getting enough a priori statistic information on reactor emergency states) corre-sponding linear decision-making functions, in fact hypersurfaces, are used. These hypersurfaces that in multidimensional feature space divide compactly disposed in it realizations of corresponding classes per-mit to perform classification of a certain realization on the base of the criterion of a distance from corre-sponding hypersurface. The approaches to pattern recognition: of beginning moment of a vapour phase steady in channels of nuclear reactor boiling-type (RBMK) and non-boiling-type (WWER) high-frequency oscillatory heat-carrier instability; thermohydraulics abnormalities, fast burnout (departure of the nucleare boiling-DNB) and slow burnout in based statistical, geometrical, neural networks models..
