Research On The Techniques Of Sensor Condition Monitoring For A Nuclear Power Plant

Safety and economy are the prerequisites for the sustainable development of nuclear power.For safety,sensors are the important part of instrument and control system,which is used to obtain the operation information of the plant.On one hand,on-line monitoring of systems and equipment can be realized,on the other hand,sensors can provide data for the safe and reliable operation of the nuclear power plant(NPP).Due to the harsh working conditions of sensors in a NPP,sensors may appear aging or mechanical faults as time goes on.Thus,it is necessary to implement on-line condition monitoring of sensors during operation to guarantee the safety of the plant.For economy,usually quantities of sensors are applied in a NPP,condition monitoring of sensors not only can enhance the safety of the plant,but also can contribute to the maintenance of sensors which is improved as condition-based maintenance(CBM)schedule from the original time-based periodic schedule.In this way,the economy of the plant is greatly improved.In this context,principal component analysis(PCA)is applied for condition monitoring of sensors,with the aim of realizing on-line monitoring of sensors and detecting the common mode failure which the traditional cross calibration technique is incapable,to increase the safety of the plant;estimating the operating conditions of sensors for the CBM schedule,to increase the economy of the plant.When PCA is applied,many factors may influence the model performance,including the quality of modeling data,the suitability of modeling structure parameters,the accuracy and stability of fault detection and identification,and so on.Thus,various optimization algorithms are proposed to deal with the foregoing factors in this paper,which is described as follows.(1)Preprocessing of modeling data:Since the quality of modeling data has direct influence on the model performance,thus data preprocessing is applied for the sensors measurements from a real NPP in section 3.Firstly,singular points are eliminated with statistics-based methods;then random fluctuations are further reduced with sliding window and wavelet transformation methods.Through simulations with measurements from a real NPP,the proposed data preprocessing methods is proved to be effective on the improvement of the PCA model performance.(2)Selecting of structure parameters:Modeling parameter is one of the main structure parameters of PCA model,which directly influences the model performance.Therefore,the selection criterion of modeling parameters is discussed in section 4.Based on traditional random selection criterion,another four criteria are proposed:the criterion based on type of sensors,the variance,the volatility degree and the correlation of sensor measurements.Through the comparison of PCA models based on various modeling parameter selection criteria,the optimal modeling criterion for PCA method among the five criteria is determined.(3)Fault detecting:For fault detection,two different optimizations are proposed in this paper.(1)The?discriminant is applied to calculate the non-detection zones of sensors in the PCA model in section 2.Through simulations with sensor measurements from a real NPP,it is proved that the proposed?discriminant is effective,that is the fault detectability of the developed PCA model is accurately evaluated.(2)In section 5,iteration-based method is applied to eliminate the false alarm of T~2 and Q statistics during model training,and statistics-based method is applied to reduce the false alarms during fault detection.Through simulations with sensor measurements from a real NPP,it is proved that the proposed false alarm reducing method can improve the condition monitoring performance of the PCA model.(4)Fault identifying:For fault identification,based on traditional contribution analysis,improved weighted contribution analysis and iterative reconstruction analysis methods are proposed in section 6.Through simulations with sensor measurements from a real NPP,the proposed fault identification methods are proved to be reasonable and effective.Through the foregoing optimizations,a comprehensive PCA-based condition monitoring framework is developed for condition monitoring of sensors in a NPP in this thesis.To test the influence of the comprehensive use of the proposed optimizations on the model performance,comprehensive simulation tests are carry out with sensor measurements from a real NPP.Through the simulation tests,it is proved that the condition monitoring performance of the PCA model with all the proposed optimizations is better than that with any one optimization in the foregoing 2~6 sections.