Application Research Of Time-frequency Technology In AP1000 Special Monitoring System

Reactor coolant pump vibration monitoring system and digital metal impact monitoring system are the important monitoring systems of AP1000's DCS.The vibration state of RCP can be monitored by the sensors installed,and the metal loose parts can also be alarmed,located and quality estimated by use of its sensors.Now,the early warning of RCP vibration state is by means of the methods of setting vibration amplitude and observing speed harmonics on the RCP vibration monitoring system,and these two methods are achieved by FFT of vibration signal.FFT is the most commonly signal processing method,but it is mainly applied to the linear and stationary signal analysis.In addition,the frequency characteristic of FFT is based on the global information of signal,which can not achieve the local feature analysis.Meanwhile,the vibration signal of RCP rotor on high speed is non-linear,and some early faults of the rotor own very low energy.For identifying the feature of these faults well,the method of signal processing is asked to have better local analysis ability.Digital metal impact monitoring system can achieve alarm,location and quality estimation for loose parts by monitoring the impact between the loose parts and the inner wall of the reactor system,at the same time,the impact signal will be accompanied by a lot of noise.Randomly varying of the noise is likely to cause false alarms of the monitoring system.In addition,some low energy impact also needs to be accurately identified,and alarmed.In order to improve the accuracy of the alarm monitoring system,the impact signal needs to be pre-denoised.In order to identify the local feature of the RCP vibration signal and denoise the metal loose parts' impact signal,a more appropriate time-frequency analysis method in this paper has been taken to identify the common fault feature of the rotor and denoise the impact signal which is accompanied by a lot of noise.Firstly,the common faults on the RCP rotor,such as imbalance,open cracks,rub-impact,loose and so on,have been simply analyzed,and the mathematical models of them have also been established combining the rotor dynamics theory.The rotor's trajectory and amplitude-frequency characteristics at different faults have been researched by using these models.Secondly,a simple experimental platform has been established by researching the collision of metal loose part.The experimental data can be obtained by use of different quality balls hitting the target plate at different height.Analyzing the frequency characteristics of different impact signals,the characteristics of signal which collected by different quality balls and different height are obtained.Based on the simulation signals of different rotor faults and the experimental impact signals,wavelet transform(WT)method has been introduced to identify the fault feature and signal preprocessing.First of all,considering the common limitations of wavelet transform applied in engineering applications,such as frequency confusion and selecting methods of optimal wavelet functions,Gray coding and frequency shift have been introduced in this paper to order the correct frequency band.At the same time,parameterization method of wavelet has been researched to enrich the wavelet function library.The maximum cross-coefficients method and maximum energy-entropy method have been contrasted in this paper to select the optimal wavelet function suited to the rotor vibration signals.Then,using wavelet packet transform(WPT)to identify the faults feature of the RCP rotor,and combing new domain function to denoise the impact signals with a lot of noise at different SNR.The results of simulation show that the WPT method not only identifies the main frequency feature successfully,but also achieves local analysis of frequency feature with lower energy.Moreover,method of wavelet denoising with new domain function can denoise the impact signal at different SNR,and increases the SNR of signal within a certain range.At last,combining the characteristic of Hilbert-huang transform(HHT)and WPT,algorithms of improved HHT and WPT-EMD denoising have been researched in this paper.Improved HHT algorithm can eliminate model aliasing and false components,and with the method of cycle continuation can also remove the endpoint effect.Based on the improved HHT,the faults feature of rotor can be identified perfectly include very lower frequency components.In addition,the algorithm of WPT-EMD achieves the denoising of impact signals involved a lot of noise,and compared with WPT denoising,it can increase SNR effectively.