Research On Vibration Signal Analysis And Fault Diagnosis Of Hydroelectric Generating Units Based On Empirical Mode Decomposition

The vibration stability of the hydroelectric generating set is the most important factor to determine the safe operation of the unit.To analyze and extract the characteristics of the vibration signal effectively and to diagnose the fault accurately is very important to improve the safety and reliability of the hydroelectric generating set.In this paper,the vibration fault of hydroelectric unit is taken as the research object,and the vibration signal pretreatment,feature extraction and fault diagnosis method are studied deeply.The main work and research results are:Firstly,this paper presents a singular data rejection reduction method based on wavelet transform for the non-stationary of the vibration signal of hydroelectric generating units.It plays a good performance of the wavelet in singular point analysis,reconstructs the clear axis trajectory pattern,and improves the accuracy of the fault diagnosis.Secondly,according to the low signal-to-noise ratio of the early fault signal of the hydroelectric generating unit,the multi-wavelet adaptive threshold denoising method is introduced to effectively remove the noise of the vibration signal.Compared with the db2 wavelet,the multi-wavelet denoising has better performance and Improve the accuracy of vibration signal feature extraction.Thirdly,in order to solve the problem of false component in the Emprical Mode Decomposition(EMD)method,the energy fluctuation method is proposed to remove the false component according to the law of energy distribution and the threshold of the fluctuation ratio,and improve the accuracy of the vibration signal analysis and fault feature extraction of the hydroelectric unit.Fourthly,we propose an Improved Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise(I-CEEMDAN)algorithm for the problem of aliasing in EMD method.This method effectively eliminates the problem of modal aliasing and ensures the reliability of signal feature extraction.Fifthly,the diagnosis method of I-CEEMDAN singular spectrum entropy and SOM(Self-Organizing Feature Maps,SOM)neural network is proposed to identify the vibration fault of hydroelectric generating unit.The method can extract the non-stationary signal feature simply and effectively higher accuracy,faster operation.Finally,the diagnosis method of I-CEEMDAN complexity and least squares support vector machine is proposed to identify the vibration fault of hydroelectric generating unit.The method can effectively judge the fault type and have high accuracy for the operation,which applicable to the case of small fault samples,improve the scope of application of fault diagnosis of hydropower units.