Purification And Feature Extraction Analysis Of The Hydroturbine Swing Signal Based On EEMD

Hydroelectric unit as a kind of rotating machine,its operating state is often closely related to the vibration of the rotor system.Therefore,the swing signal analysis of generating unit is an extremely important part in the study of fault diagnosis.Accurately obtaining the characteristic information about the swing signal is very important for the safe operation and management of generating unit.In this paper,the three aspects among theory,simulation and field test of the swing signal purification and feature extraction were studied.A kind of method for the swing signal purification and feature extraction based on EEMD is proposed in this paper,by the signal time-frequency analysis method as the starting point and the rotor system as the object.The advantages and disadvantages of FFT,STFT,wavelet analysis,EMD analysis and EEMD analysis are compared by theoretical research.Theoretical analysis shows that time-frequency resolution of EEMD has adaptive ability compared with wavelet analysis,and effectively suppresses EMD modal aliasing phenomenon.The time-frequency domain analysis of the simulation signals further shows that EEMD has the strong capability of local time-frequency domain analysis and the adaptive decomposition for different actual signals.This paper proposed an improved linear combination index based on the traditional index.The actual swing signal is difficult to identify,so it needs to be denoised.As for the signal denoising,the evaluation of the denoising effect has been focus of many scholars.However,there are obvious shortcomings in the traditional evaluation index,and there is a possibility of misjudgment in the evaluation process.This index can determine the optimal decomposition layer of wavelet denoising and the optimal denoising layer of EMD(EEMD)denoising,which proven by denoising analysis of the simulated signal.In addition,the denoising method has also been studied in this paper.Through the comparison of wavelet,EMD and EEMD denoising analysis of the simulation signal,it is found that EEMD not only has excellent denoising ability,but also has stronger applicability.Once there is a little deviation in the optimal decomposition layer of actual signal denoising,wavelet denoising will take away more real components,while the EMD(EEMD)denoising won't produce this kind of influence.At last,in the separation and feature extraction of multi-source vibration signal,this paper also proposed a method of EEMD combined with ICA.Firstly,the ICA is used to separate the multi-channel signals.Then,the EEMD is used to denoise the the preliminary separated source and reconstruct signals with same frequency.Finally,the characteristic source is obtained.The simulation signal analysis found that more clear vibration source signals under the condition of as little observation signal as possible can be get through this method.The superiority of EEMD in denoising,rotor orbit purification and feature extraction is further verified in the field test analysis of the swing signal of turbine.According to the above research,the EEMD signal analysis system based on MATLAB GUI platform is constructed,which has realized the time-frequency analysis,the denoising of the swing signal,the purification of rotor orbit and the feature extraction analysis.It provides a convenient research platform for the popularization and application of EEMD algorithm in engineering practice.