Feature Extraction And State Recognition Of Cavitation Acoustic Emission Signals Of Hydraulic Turbines

Hydraulic turbine is the key equipment of the hydropower station to realize energy conversion,whose safe and stable operation determines the safety of the entire hydropower station.Turbine cavitation is the major factor in reducing useful life and operational performance of hydraulic turbine.When the cavitation is serious,the vane,runner,draft tube and other flow component of the turbine often get much damage,which result in the reduction of output and efficiency of the turbine,and lead to the failure of safe and stable operation of unit.Acoustic Emission(AE)detection technology utilizes elastic waves generated by the rapid release of energy from part materials as an excitation source,and plays an important role in nondestructive testing.Therefore,based on AE detection technology,many groups of cavitation tests under multiple working conditions for a Francis turbine model turbine were carried,followed by noise reduction processing and feature extraction for the collected AE signals,so as to clarify the characteristics and rules in the cavitation process of the turbine and seek for the solution to cavitation fault.Based on the cavitation characteristics of hydraulic turbine and the feature of AE signal,the cavitation in different states is identified based on BP neural network.The main research contents of this paper include:(1)A cavitation test of a Francis turbine model was implemented on a domestic close-loop turbine model test bench with an international advanced level and a comprehensive precision less than ±0.2%.The signals were collected by the acoustic emission acquisition system independently developed by our research group.In order to focus on the law of signal variation in the cavitation process and avoid the mutual interference between different cavitation types,6 sets of test conditions that deviate from the design conditions and correspond to different gate opening and unit speeds are selected.(2)A noise reduction method for turbine cavitation AE signal based on lifting wavelet transform with improved threshold were established.The method uses the lifting wavelet transform to decompose the signal to obtain multiple wavelet coefficients,and uses the function of improved threshold to deal with the wavelet coefficients,and finally enhances the wavelet coefficients reconstruction.we conduct noise reduction processes toward the simulated signal and the AE signal of the turbine cavitation test,the result shows that the established noise reduction method can effectively avoid the limitation of the traditional wavelet threshold noise reduction,and effectively reduce the noise while retaining the useful high frequency signal ingredient.(3)AE parameter counting method and the lifting wavelet transform waveform analysis method were used to extract the features of the cavitation AE signal under different conditions of the model turbine after noise reduction.The result shows that in the process of cavitation state from nothing to weak and weak to strong,the parameter characteristics under various working conditions and the absolute energy value of the main frequency band of the improved wavelet transform show a very obvious law,however,the law of relative energy value is not very obvious.(4)A BP neural network-based hydrofoil cavitation state recognition method was established.The extracted eigenvalues are normalized as the input sample eigenvectors of the BP neural network.The cavitation state samples are trained to be different for the cavitation state recognition of turbine.The result shows that the recognition rate of this method is as high as 88%,which can realize the recognition of different cavitation states.