Fault Diagnosis Method Research Of The Integral Reactor Coolant System

The integral reactor coolant system refers an integral reactor pressure vessel, in which all the equipment and components includes. This simplified integral arrangement improves the inherent safety of the reactor. However, this integral structure also brings to inestimable difficulties for the maintenance of equipments. Therefore, it is necessary for the integral nuclear power plant to research typical failure of reactor coolant system.The integral reactor coolant system and its key equipment-the reactor coolant pump is the study object in the paper, research fault diagnosis methods from the two lays of systems and equipment by a variety of diagnostic methods of intelligence analysis. The fault diagnosis platform based on expert system has been builted; it included data acquisition, alarm, diagnosis, knowledge base, reasoning, self-learning functions.Using of internationally recognized thermal hydraulic analysis software relap5/mod3 analysis the integral reactor coolant system typical system failure, such as the leak of take-over, the main pump poweroff, and so on, simulate the main parameters such as pressure, temperature, water level, and other parameters. Rough set theory can make use of a large amount of data extraction, reduction and elimination of redundant characteristics of knowledge, through establishment of an attribute table, get parameters and data of failure characteristics, then extracting rules. And, according to the above steps, program the knowledge acquisition process modules, and integrated with expert systems, Rough Set Theory (RST) used in the acquisition of knowledge to solve the "bottleneck" problem of expert system.The reactor coolant pump plays an important role in the reactor coolant circulation. Through the main pump failure analysis, the typical failure:Crack Open, Misalignment fault, Bending Rotor, Speed Non-match Fault, simulated by the Simulink software, and in the process of accession add to the "white noise" so that the simulation results closer to reality, and then using wavelet analysis-"signal's microscope" to make time-frequency analysis for the fault model in order to extract time-frequency parameters. By comparing with different fault, the corresponding frequency characteristics is extracted, and it is added into expert knowledge database.Using of artificial intelligence methods, research of method of fault diagnosis for integral reactor coolant system and the main pump failure, that can be found: wavelet analysis of vibration signals are very good analysis of the "micro-mathematical" features can effectively extract the vibration signal characteristic of the fault to fault identification, fault diagnosis purpose; through the use of Fourier transform and wavelet packet transform to extract the failure characteristics, it can be seen that through the fault of some Fourier transform and wavelet packet can be extracted characteristics of the fault, but obviously we can see that wavelet packet with a more fine recognition, is more adaptability, especially for those more complex characteristics of the frequency characteristics of the fault; through the use of the reduction algorithm based on discernibility matrix and Boolean logic and the reduction algorithm based on discernibility matrix of attributes characteristics for the integral reactor coolant system's failure, it is successed reduction for the redundancy of the attribute information, which could be used for on-line and off-line self-learning algorithm; Using of the RST can reduce effectively fault information, through the reduction algorithm based on discernibility matrix and Boolean logic can get a more comprehensive rules of fault diagnosis, which is poor real-time, but the reduction algorithm based on discernibility matrix of attributes characteristics to ensure the real-time requirements, so both methods can use for on-line learning and off-line learning, the reduction algorithm based on discernibility matrix and Boolean logic can add and improve for on-line learning in order to provide real-time fault diagnosis and reliability.Due to nuclear power'requirement for safety, the proposed approach has broad applications, will play a huge role in nuclear power fault diagnosis, but nuclear power plant due to the complexity of the system needs to be done a lot of practical efforts.