Abnormity Attribution Method and Verification for Mechanical Seal of Reactor Coolant Pump

Mechanical seals are widely used in nuclear power equipment, and the analysis on their abnormities received a lot of attention. The reactor coolant pump focused in the presented study is composed with three seals and four throttles. Linear models for the seals and the throttles were established for efficient computation, the parameters on the designed state were determined according to operating condition and related literature. Discussions on the fluid resistance impact to the seal system show that the monitored information is inadequate for a definite solution of the reactor coolant pump mechanical seal physical model, namely forming an under-defined problem. That is, different characteristic parameters of the components may lead to similar monitored results and thus cannot be distinguished. This is due to the structure of the mechanical seal system of the reactor coolant pumps, and cannot be easily changed for current users of the pumps. Probability-based algorithms were proposed to provide real-time analysis for the health status of seal, thus to alarm and find the cause when abnormity exists. The parameters in the characteristic models of the seals and the throttles were assumed to obey Gaussian distributions which have mean values corresponding to the designed state and standard deviations decided by the flux magnitudes. The results were given in the forms of maximum likelihood state and the abnormity event probability, respectively. The maximum likelihood characteristic parameters were calculated based on the probability density function of the prior distribution, and then the corresponding full description of the system state was generated, providing an intuitive illustration of the multi-stage seal system including the probable abnormity causes and the flux through each of the seals and the throttles. By sampling according to the prior distribution and defining a fuzzy selection rule, the probability of the specified events concerned by the user was obtained through Bayesian principle. Three thresholds of different severity levels were configured for each of the seals in the system and the probability of abnormity beyond the thresholds were analyzed. The methods were tested on an about one-year-duration real operation dataset containing pressure and flow rate data of three pumps in a reactor coolant system, and it is found that the result of the methods is consistent with the conclusion drawn in shutdown maintenance and the monitored total leakage of the three pumps in a reactor coolant, indicating that the method can effectively attributes the abnormity of the multi-stage mechanical seal system of reactor coolant pumps. This shows that the method can effectively carry out health monitoring and fault tracing for the mechanical seal of reactor coolant pump, and it has high popularization value. © 2022 Atomic Energy Press. All rights reserved.