Research On Remaining Useful Life Prediction Method Of Circulating Water Pump Roller Bearing In Nuclear Power Plant

The safe development and peaceful application of nuclear energy is an important force to solve energy crisis,promote energy transformation and achieve environmental protection goals.However,circulating water pumps and other equipment are also faced with safety hazards caused by performance degradation with the long-term operation of nuclear power plants.As the key driving equipment of the final cold source in second-loop system,the reliability of circulating water pump plays an important role in ensuring the continuous and stable operation of the unit.During operation,roller bearing,as a key component of the core part in circulating water pump transmission,plays an important role in carrying equipment load and absorbing system fluctuations.Once degraded,the whole circulating water pump system will fail,thus bringing challenges to the safe operation of nuclear power plant.However,the maintenance strategy of bearings and other components in commercial nuclear power plants are still based on the preventive maintenance of "inservice inspection",which has the disadvantages of "insufficient maintenance" and "over maintenance",which is difficult to meet the application requirements of predictive maintenance.Therefore,it is of great engineering value to effectively evaluate the degradation state of circulating water pump roller bearings and predict the remaining useful life(RUL)by combining artificial intelligence and information technology.Currently,one of the most serious problems is the inaccuracy and incredibility of their core techniques — RUL prediction technology.Therefore,it is difficult to fully meet the practical engineering requirements of continuous health monitoring and predictive maintenance for circulating pumps.In previous studies,RUL prediction methods mainly focus on the analysis of monitoring data,rather than the analysis of causes and mechanisms in equipment degradation process.This makes it impossible to physically parse the mapping relationship between the internal degradation state of the device and the external monitoring form.In addition,the coupled interference of equipment running noise and component modulation,as well as the imperfection of RUL prediction mechanism,make it difficult to identify the running state of equipment in time and predict RUL accurately.In order to solve the above problems,the key theories and methods of the degradation state assessment and RUL prediction of circulating pump roller bearings are paid more attention.And then,the physical model of degradation mechanism,data-driven method and Bayesian theory are organically integrated,and a RUL prediction method based on digital-analog fusion is proposed.Mainly carried out the following work:(1)The function of nuclear power equipment fault prediction and health management system is designed and the core position of RUL prediction technology is defined.On this basis,a set of degradation state assessment and RUL prediction method based on model and data fusion is proposed.At the same time,the rationality of the scheme is elaborated,which improves the passive situation that the prediction results of RUL lack explanation.(2)In view of the lack of research on fatigue degradation mechanism and life cycle data,a finite element explicit dynamic model of roller bearings was established to analyze the damage evolution process of fatigue degradation,internal contact characteristics and vibration response caused by defects.The fatigue degradation mechanism of bearing is also clarified.Then,the whole life cycle experiment of bearing fatigue degradation was completed by setting speed,load and other acceleration forms.On the basis of bearing degradation data,the vibration characteristics analysis of bearing degradation process was realized.Also the impact characteristics,statistical characteristics and spectral characteristics of bearing degradation process were understood,which provided a theoretical basis for the subsequent chapter research.(3)In view of the fact that the early weak degradation characteristics of bearings are susceptible to noise interference and the characteristic information is affected by the modulation of multiple factors such as transmission path,speed and noise,a technical system of bearing monitoring information was constructed.The signal enhancement and demodulation denoising were realized by combining adaptive variable-scale stochastic resonance and improved robust local mean decomposition.Based on the analysis results of bearing vibration characteristics,the NTalaf composite index with definite physical significance was constructed to realize the degradation characterization of roller bearings.Finally,early degradation point detection and degradation pattern recognition of bearings were realized based on sliding window and Hilbert square envelope spectrum analysis in this paper,which provided trigger conditions for subsequent bearing degradation state assessment and RUL prediction.(4)Aiming at the simple physical model of roller bearing degradation and the and poor interpretation of RUL prediction results.A real-time evaluation and RUL prediction method of bearing degradation state was proposed by combining the multi-stage degradation physical model with the nearest similarity particle filter.First of all,the bearing degradation stage is identified through the real-time monitoring of NTalaf,and based on this,the physical model conforming to the current degradation trend is matched.Secondly,the model parameters were updated by the most recent similarity particle filter and the latest measurement index,also the bearing degradation index was tracked and the current degradation damage degree was accurately evaluated.Finally,the recursive prediction of RUL is realized based on the current degradation state of bearings,.(5)Aiming at the problems that the current RUL prediction mechanism of equipment is not perfect and the uncertainty of prediction results is difficult to quantify,this paper organically integrates the Bayesian theory,mechanism model and data-driven method,and a framework is proposed for the degradation state evaluation and RUL prediction of roller bearings.After clarifying the interface relationship between the single RUL prediction model and the joint prediction model,the aleatoric uncertainty and epistemic uncertainty are introduced into the deep learning network.The uncertainty quantification of prediction results of data-driven method is solved by constructing the RUL prediction model based on Bayesian long and short time memory network.Finally,the bayesian long and short time memory network prediction model and the RUL model driven by model and data fusion are effectively verified through the whole life cycle data of roller bearings.