Research On Fault Diagnosis Of Wind Turbine Gearbox Based On Neural Network

Energy shortage and environmental pollution are two important issues in our times.Renewable-clean new energy can effectively solve them.Therefore,in recent years,countries are actively developing new energy.Wind power is one of the new energy with a vision of commercial development.However,with the rapid development of the wind power market,the double-fed wind turbine(DFWT)have frequent failures due to harsh working conditions.gearbox is a compnent of frequent failures and the longest time-consuming troubleshooting,which principally affects the power generation efficiency and causes high operation maintenance(OM)cost.Therefore,the research on fault diagnosis of the gearbox,accurately determining failure situation as soon as possible are significative.This paper takes DFWT gearbox as the research object.Firstly,it introduces the common fault types and causes of gearbox components,analyzes the vibration mechanism of the gearbox,and gives the frequency characteristics of the gearbox gear and bearing fault vibration signals.Secondly,in order to separate gearbox gear composite fault,accurately determine the gearbox fault situation,this paper will use wavelet packet(WP)and ensemble empirical mode decomposition(EEMD)to process vibration signal,namely,WP is used firstly to decompose the signal into multiple frequency bands.EEMD is performed secondly for each frequency band,and finally the intrinsic mode function(IMF)is selected according to the correlation coefficient criterion.The above method is used to process a simulation signal containing multiple fault conditions,the obtained IMFs are respectively subjected to spectrum and envelope analysis,the result show that main frequency components related to each fault in the simulation signal are decomposed into different IMFs,the envelope only contains the characteristic frequencies associated with its own fault condition.Comparative analysis,using WP or EEMD to process the simulated signal seperatly,each frequency band or IMF does not completely separate the main frequency components,and the envelope spectrum often contains multiple characteristic frequencies.It is not conducive to a clear diagnosis of various fault conditions.Finally,the QPZZ-II Gearbox Simulation Platform was used as the experimental equipment to obtain the actual vibration data,and the application analysis and simulation verification were carried out.In order to realize the automatic diagnosis of the gearbox failure,the actual vibration data of the gearbox containing various fault conditions is firstly analyzed by WP and EEMD,then the energy values,root mean square values and kurtosis of each IMF are calculated as the quantitative feature index of gearbox fault information,which is sent to RBF neural network with particle swarm optimization for training to establish a system that can be used for fault classification.Compared with the normal RBF,the optimized RBF neural network has higher accuracy.