Application Of Improved ATVF In Fault Diagnosis Of Wind Turbine Drive Gear

As the transmission and acceleration device of wind turbine,the gearbox plays an important role in the safe operation of wind turbine.Affected by variable wind direction,wind speed and other factors,the gearbox usually has the characteristics of variable load and speed,so its failure rate is high.However,the failure of gearbox usually causes the wind turbine to shut down for overhaul,and even causes great losses in serious cases.Therefore,it is of great practical significance to monitor and diagnose the operation of wind turbine gearbox.Compared with gears in other mechanical equipment,speed and load of wind turbine drive gear will change due to factors such as wind speed and wind direction instability,starting and stopping,etc.,and it is usually in variable operating conditions.Vibration signals of wind turbine drive gear show non-linear and non-stationary characteristics.At the same time,due to complex operating environment,a large number of noise disturbances often exist in the collected gear vibration signals,weak fault features are easily submerged in noise and cannot be identified.However,traditional time-frequency analysis methods have some limitations in extracting fault characteristics of wind turbine gearbox under variable operating conditions.Therefore,based on the above problems,the following research work is carried out in this paper:(1)The dynamic model of gear engagement is established,the gear vibration signal model is given according to the gear vibration characteristics at variable speed,the gear engagement characteristics and the characteristics of modulation side frequency band are elaborated,the gear fault type and corresponding vibration signal characteristics are introduced,and the difficulties in fault diagnosis of wind turbine drive gear are analyzed.(2)Aiming at the disadvantage of adaptive time-varying filter(ATVF)which only filters the noise in the stopband and does not effectively deal with the noise in the passband,a method of feature extraction of non-stationary signal based on EEMD and ATVF is proposed.This method can not only effectively filter the noise in the stopband,but also suppress the noise in the passband.The validity of the proposed method is verified by the analysis of simulated variable speed gear fault signal and actual gear signal.At the same time,the method is compared with several common time-frequency analysis methods,which highlights the superiority of this method.(3)Aiming at the problems of low calculation accuracy and long calculation time of ATVF,an adaptive time-varying comb filter(ATVCF)method is put forward.This method combines ATVF,comb filter and kernel filter ideas,and according to the characteristics of gear fault vibration signal,a comb filter is set at each moment,which can adaptively change the filter parameters with the change of gear speed,and targeted filtering at fault modulation frequency point can effectively improve signal-to-noise ratio of analysis signal.On this basis,Viterbi algorithm(VA)is introduced into gear meshing frequency estimation of ATVCF,and a feature extraction method of variable speed gear based on VA and ATVCF is proposed.The calculation efficiency of the improved algorithm is significantly improved.The simulation and WTDS testbed data analysis show that this method can extract gear signal fault characteristics very well and is very suitable for wind turbine gear fault diagnosis.(4)For diagnosing gear faults in engineering practice,a new fault identification method suitable for variable speed is put forward by combining VA,ATVCF and Fuzzy C-Means(FCM)clustering algorithm,and it is applied to wind turbine drive gear fault identification.This method firstly uses the filtering methods based on VA and ATCF to filter the signal,and calculates the eigenvalues of the filtered signal;then the eigenvalues form the eigenvector matrix;secondly,FCM is used to classify the eigenvector matrix and obtain several clustering centers.Finally,the minimum Euclidean distance between the eigenvalues of the signal to be identified and several clustering centers is calculated,gear fault identification can be realized.The validity and superiority of pattern recognition methods based on ATVCF and FCM are highlighted by comparing them with those based on EMD and FCM.