Study On Vibration Fault Diagnosis And Health Performance Trend Prediction For Hydropower Generator Unit

As the key equipment in the energy conversion of hydropower station,hydropower generator unit(HGU)tends to be large and intelligent.Due to the adverse operating environment of HGU and the coupling influence such as hydraulic,mechanical and electrical factors,the risk of abnormal vibration,compound fault,fatigue degradation and even structural damage are becoming increasingly prominent.Thus,the issues of conditionbased maintenance(CBM)of HGU have attracted extensive attention from researchers.The thesis focuses on the key scientific issues of CBM in engineering,such as vibration signal de-noising,fault diagnosis and performance prediction of HGU.Taking the advanced methods like signal analysis approach,intelligent optimization algorithm,machine learning as the research tool,the method of analyzing nonlinear vibration signal for HGU is put forward.On the basis,we construct a fault diagnosis model with typical vibration signal of HGU,which utilizes intelligent algorithm to collaboratively realize feature reduction and parameters optimization.Besides,considering the multiple operating parameters of HGU,a health performance trend prediction model is developed.Further,we design a multiobjective optimization strategy of health performance interval with information fusing.Consequently,the relevant achievements are of great application value to ensure the safe,efficient and stable operation for HGU.The main contents and innovative achievements of the thesis are as follows.(1)Under the interference of hydraulic-mechanical-electrical factors and the background noise,the vibration signal cannot effectively characterize the state of HGU.A de-noising method based on mode function reconstruction and adaptive singular value decomposition(SVD)is proposed.With the selection strategy of VMD mode function based on correlation analysis,the primary denoising of original signal is completed.Then,based on analyizing effect of singular value sequence on filtering performance,an adaptive SVD secondary filtering method based on kurtosis theory is developed to effectively eliminate the non-periodic random noise.Further,the method can provide data basis for analyzing the condition of HGU.(2)Since the redundant information in multi-dimensional feature space and parameters of support vector machine(SVM)both have great influence on the diagnosis accuracy,a fault diagnosis model based on intelligence algorithm is proposed for typical vibration signal.It can collaboratively realize the feature reduction and model parameters optimization.A multi-dimensional feature extraction method is firstly designed to obtain the fault information of HGU.On the basis,the binary optimization algorithm is used to realize multi-dimensional features reduction,which break through the limitations of traditional feature reduction.Meanwhile,the decimal optimization algorithm is applied for adaptively searching the best combination of feature subsets and model parameters.Consequencely,th accuracy of fault diagnosis can be improved.(3)Since the traditional prediction methods only utilize the single-dimensional data,a health performance tendency prediction model of HGU based on convolution neural network-long short-term memory neural network(CNN-LSTM)is proposed to make up for the insufficiency of the post-analysis.With analyzing the correlation between the vibration signal and operating parameters,a health state model based on Gaussian process regression is developed,which can effectively describe the operating characteristic of HGU.Then,the health performance index(HPI)is designed to quantify the health performance level.Furthermore,the CNN-LSTM model is proposed with the ability of local feature extraction and nonlinear learning,which can timely and accurately predict the health performance tendency of HGU.Further,the evolution law of health performance is effectively revealed.(4)Aiming at the defect that the prediction method cannot convey the uncertainty information in the unit’s health performance,a health performance interval prediction model based on information fusing and multi-objective optimization is proposed,where the influence of the health levels from different components of HGU is analyzed.The entropy weight theory is introduced to effectively fuse the key information of health performance.Then,an integrate health index(IHI)that can represent the health level of HGU is constructed.On this basis,considering the uncertainty of interval prediction,the global optimization strategy for the prediction model is developed based on the multi-objective algorithm.In the final,a more accurate and reliable range of health performance is obtained,which can better provide decision support for analyzing the risk on the health state of HGU.