Study On Fault Feature Extraction Method From Vibration And Noise Of Hydroelectric Unit

As an important part of the power system for peak and frequency modulation and reducing the impact of new energy on the grid,the safe and stable operation of hydropower station is very important.Although the traditional scheme of regular shutdown and maintenance of hydropower units can effectively discover and solve various safety problems in operation,it affects economic benefits to a certain extent.About 80% of the failures of hydropower units are related to vibration abnormalities due to their rotating characteristics.Therefore,the operation and maintenance level of hydropower stations can be improved by changing the "regular maintenance" of hydropower units into "on-demand maintenance" based on vibration state monitoring.However,in practice,the traditional contact vibration sensor has the problems of low measuring frequency range and unsuitable for noncontact monitoring scenarios.At the same time,it is difficult to accurately extract the early fault features of the unit because of the weak amplitude and the strong background noise of the early fault signal.In addition,due to the complexity of the unit structure and the influence of hydraulic,mechanical and electrical coupling factors during operation,the vibration signal of the unit has strong nonlinearity and non-stationarity.Therefore,fault feature extraction based on vibration signals of hydropower units is always a difficult problem for state monitoring of hydropower units.In order to solve the above problems,this thesis designed a condition monitoring scheme of hydroelectric generator units based on vibration and noise signals,and proposed a fault feature extraction method of hydropower units based on singular value decomposition and multiple matching synchrosqueezing transform.The main research of this thesis is shown as follows:1.Aiming at the weak amplitude and strong modulation of the early fault signal of hydroelectric generator units,an improved singular value signal denoising method was proposed based on Difference Ratio(DR)index and Periodic Modulation Intensity(PMI)index.Firstly,the appropriate number of lines of Hankel matrix is determined by DR index to ensure that the single component signal in the complex multi-component signal to be analyzed can be decomposed into two similar singular components with half the amplitude and the same phase as the original signal component.Then,with prior knowledge of mechanical faults,PMI index of each singular component is calculated by autocorrelation function.Finally,according to the PMI index of each singular component,the reconstruction weight of the corresponding component is calculated,and the noise of vibration signal is reduced by the weighted sum of the singular component.The simulation results show that the singular value denoising method based on DR and PMI index can accurately extract the fault components with periodic modulation characteristics related to mechanical faults from the monitoring signals containing complex background noise.2.Aiming at the problem of low time-frequency resolution of synchrosqueezing transform in processing strongly non-stationary timevarying signals,a time-frequency post-processing algorithm based on multiple matching synchrosqueezing transform is proposed in this paper.Based on the framework of short-time Fourier transform,this paper constructs matched instantaneous frequency estimator based on linear frequency modulation signal,and introduces iterative operation to reduce the instantaneous frequency estimation error of matched instantaneous frequency estimator in processing strong non-stationary signal.In the realization of the algorithm,the quadratic rounding operation of matched instantaneous frequency estimation operator is carried out to ensure that the scattered time-frequency coefficients can be allocated to the correct position.When processing non-stationary signals with SNR(Signal Noise Raito)of30d B,compared with synchrosqueezing transform and matching synchrosqueezing transform,the Renyi entropy of multiple matching synchrosqueezing transform is reduced by 1.928 and 0.918,and the mean square error of instantaneous frequency extraction is reduced by 0.1707 and0.0037.At the same time,under the condition of different SNR of 0～30d B,the above two indexes of the multiple matching synchrosqueezing transform are also smaller than those of the synchrosqueezing transform and matching synchrosqueezing transform.Experimental results show that the proposed method can obviously improve the time-frequency resolution and has strong robustness in processing strong non-stationary signals.3.The vibration and noise signal acquisition system of hydropower unit was designed and built.The improved SVD signal denoising method and multiple matching synchrosqueezing transform were used to extract features,and the load adjustment discrimination,simulated knock fault and sharp fault characteristic frequency extraction of hydropower unit were realized.In dealing with the vibration noise signal of stable load and variable load,the root mean square value of the time-frequency coefficient of the corresponding frequency band is extracted based on the time-frequency diagram obtained by the multiple matching synchrosqueezing transform to realize the distinction between the stable load and the load regulation condition of the hydroelectric unit.In dealing with the noise signal of simulated knocking fault,the singular value noise reduction method based on DR index and PMI index realizes the extraction of knocking fault signal components,and the corresponding fault frequency 2Hz is obtained by combining the envelope analysis.In the signal,the instantaneous frequency of the fault component near 1700 Hz is extracted on the basis of the results of the multiple matching synchrosqueezing transform,the signal components related to sharp faults are reconstructed according to the instantaneous frequency,and the fault frequency 2Hz can be accurately extracted by combining with Fourier analysis.