Improved EMD And Correlation Dimension Method And Its Application In Feature Extraction Of Acoustic Emission Signals Under Hydraulic Turbine Cavitation

Hydraulic turbine is the core component of hydropower generation,but cavitation can reduce the efficiency of water turbine and seriously endanger the safety operation of hydraulic turbine.Therefore,it is of great significance to monitor and study the cavitation states of hydraulic turbines.At present,the use of acoustic emission technology to detect cavitation signals is getting more and more attention from researchers.The acoustic emission(AE)signals generated by cavitation are a non-stationary and non-periodic mutation signal,and it is difficult to extract the characteristic by conventional methods.This paper proposes a feature extraction method of cavitation signals of hydraulic turbine based on improved Empirical Mode Decomposition(EMD)and correlation dimension,and studies the change rule of the correlation dimension of acoustic emission signals in the cavitation states of a hydraulic turbine with cavitation coefficients.The main contents include:(1)A series of cavitation tests of a Francis turbine model were implemented on a closed-loop hydraulic model test bench with an advanced international level and comprehensive precision less than±0.2%,and the acoustic emission signals were acquired during various cavitation courses,which could collect and save the relevant physical parameters,calculate the efficiency and cavitation coefficient of the turbine mode.Cavitation acoustic emission signals under different working conditions were obtained by changing the guide vane opening,rotational speed and the pressure of the tail water tank.When the working condition was stable,The data acquisition system was used to collect the acoustic emission signals of cavitation for subsequent signal feature extraction.(2)Aiming at the problem of endpoint effect in EMD algorithm,improved EMD method based on mirror extension and variable cosine window is proposed.Firstly,the signals are extended by mirror extension.Then the extended signals are added with a variable cosine window function.Finally,the signals are decomposed by EMD,and the data of Intrinsic Mode Functions(IMF)are obtained by discarding the extension part of each IMF at the same time.Simulation results show that this method can effectively suppress the endpoint effect of EMD.A false mode identification method based on energy spectrum is used to solve the EMD false mode problem.The method calculates the energy spectrum of each IMF and compares it with the threshold of the energy spectrum.Simulation results show that the proposed method can effectively eliminate spurious modal components.(3)A feature extraction method based on improved EMD and correlation dimension for acoustic emission signals of hydraulic turbine cavitation is proposed.Firstly,a series of IMFs are obtained by processing AE signals with improved EMD,and the false IMF components are eliminated.Secondly,the correlation dimension of each IMF is extracted for reflecting the cavitation feature.(4)Based on the improved EMD and correlation dimension,the acoustic emission signal features of hydraulic turbine cavitation under different rotational speeds and different flow rates are extracted.It is shown that with the cavitation state from non-cavitation,and the incipient cavitation to the critical cavitation,cavitation coefficient gradually decreases,and the correlation dimension of each order of IMF components gradually increases,with obvious regularity.The increase of the correlation dimension of the acoustic emission signal directly reflects the cavitation evolving process of the hydraulic turbine,where the flow state becomes more turbulent.The study may provide a reference for the feature extraction of cavitation signals of hydraulic machinery including hydraulic turbines.