Diagnosis Of Bearing Faults Of Wind Turbine Generator Based On Mathematical Morphology

The clean energy represented by wind power technology is rapidly emerging globally.Compared with traditional fossil fuels,the biggest advantage of wind power generation is that it causes less pollution to the environment.The cumulative installed capacity of wind turbines worldwide is increasing year by year.Bearings,as the core components of the wind turbine drive system,play a vital role in the safe and stable operation of the entire wind turbine drive system.Among the mechanical transmission system failures of wind turbines,many faults are caused by bearing damage and failure.Due to the long-term service of wind turbines under complex alternating loads,coupled with the effects of strong background noise,electromagnetic interference and complex transmission paths,the measured generator bearing fault signals will show nonlinear and unsteady characteristics.This brings great difficulties to the fault diagnosis and intelligent monitoring of generator bearings.Commonly used bearing fault diagnosis signal processing methods are time domain method,frequency domain method and time-frequency domain analysis method.However,the time domain method and the frequency domain method alone cannot effectively detect the weak fault information of the bearing,and the traditional time-frequency analysis method has certain limitations.Therefore,how to find a new fault diagnosis research method to detect the early weak faults of the bearing is very necessary.Mathematical morphology theory has now been widely used in the field of mechanical vibration signal processing.It extracts fault characteristic information by directly acting on the time-domain signal through pre-set structural elements.It is an effective nonlinear and unsteady signal processing method.This paper takes wind turbine generator bearings as the research object,and improves and extends the theoretical methods of mathematical morphology to extract weak fault characteristic information of generator bearings.It provides a theoretical basis for early accurate fault detection and fault classification of generator bearings.The specific research contents of this paper are as follows:(1)Through the basic theoretical research of mathematical morphology,the filtering rules of feature extraction operators and noise reduction operators are revealed.In order to be able to extract the fault impact feature information of the generator bearing,an enhanced morphological difference operator(EMDO)is constructed based on the product of the difference gradient operator(MG)and the closed-open difference gradient operator(DIF).When selecting the optimal structural element scale of EMDO operator,the feature energy factor(FEF)is used as the evaluation index.Next,the fault signal model of the bearing outer ring is established,and the filtering performance of the MG,DIF and EMDO operators is quantitatively analyzed.The research results show that the EMDO operator can enhance the fault impact information of the bearing.Finally,the fault experiment verification of the wind turbine generator bearing is carried out,and the experimental comparison and analysis with the MGPO and MHPO operators are carried out.The research results show that the proposed EMDO operator has superior fault feature extraction ability and is suitable for the fault diagnosis of the generator bearing.(2)The generator bearing fault signal often suffers from the interference of strong background noise,multi-coupling harmonic signal and random impact signal.It is difficult to suppress these interference components using EMDF filtering,so in order to make up for the lack of EMDF's de-noising ability,the probabilistic principal component analysis(PPCA)method is introduced.In the traditional PPCA algorithm,the parameters for decomposing the principal component k and the original variable n are usually set manually.In order to solve this problem,a parameter adaptive PPCA method based on grasshopper optimization algorithm(GOA)is proposed.Subsequently,a new dimensionless comprehensive evaluation index KSP was constructed to comprehensively and quantitatively detect the noise reduction performance of PPCA,and the maximum KSP value was used as the objective function of the GOA algorithm to optimize PPCA parameters.Finally,an adaptive PPCA-EMDF fault diagnosis method for generator bearings is proposed.Simulation and engineering application results show that the proposed method can effectively analyze and diagnose the fault information of generator bearings.The comparison results with the ACDIF and VMD methods show that the algorithm has certain advantages.(3)Based on the construction mechanism of the morphological top hat transform operator,four harmonic signal extraction operators(enhanced average filtering,EAVG_DC-EO,EAVG_DC-OE,EAVG_CD-EO and EAVG_CD-OE)are constructed.Then it is proved by simulation that the EAVG_CD-OE operator's ability to restore harmonic signals is better than the other three operators.In order to maintain the integrity of the positive and negative pulses of the time domain signal,an enhanced top-hat morphological filtering(EAVGH)is defined.The comparison results with other four top hat transform operators BTH,WTH,AVGH and CMFH show the superiority of the EAVGH operator.On the basis of this research,in order to further solve the problem that the generator bearing fault signal is affected by the nonlinear modulation frequency components,the cyclic spectrum coherence function(CSC)is introduced.Finally,an EAVGH-CSC method for generator bearing fault diagnosis based on the combination of enhanced top hat transformation and cyclic spectrum coherence is proposed.The experimental results prove that the EAVGH-CSC method has high fault feature extraction ability.(4)In order to analyze and extract the fault feature information of generator bearings from multiple scales and angles,a multiscale enhanced top-hat morphological filtering(MEAVGH)operator is defined.In order to further enhance the fault feature information,a new multi-scale feature extraction filtering operator(MFEO)is constructed by multiplying MEAVGH and MCMFH.Subsequently,in order to solve the problem that the kurtosis index is easily affected by large-scale signals,a comprehensive non-dimensional evaluation index of kurtosis fault feature ratio(KFFR)is proposed.On the basis of this research,in order to solve the problem of fault feature extraction caused by the unreasonable selection of scale-weighted interval in the traditional MMF method and MCMFH method,a multi-scale morphological signal reconstruction method with iterative weighted threshold is proposed.Simulation and experimental results verify the effectiveness and engineering applicability of the proposed MFEO method.Because two-dimensional images contain more characteristic information than one-dimensional signals.In order to solve the problem that traditional time-frequency analysis methods need to rely on expert knowledge when converting one-dimensional vibration signals into two-dimensional images,a signal image conversion method based on multi-scale mathematical morphological transformation is proposed.Then the generated image samples are regarded as the input of Convolutional Neural Network(CNN).In order to improve the generalization ability of the CNN model,the Batch Normalization(BN)method was introduced.Finally,an intelligent recognition classification method called MFEO-CNN was proposed.The fault data set of the generator bearing verifies the effectiveness of the MFEO-CNN method.The comparative analysis results with EMD-CNN,EWT-CNN,CNN,SVM and ANN methods show that the MFEO-CNN method has higher classification accuracy.