Research On Fault Diagnosis Method Of Wind Turbine Gearbox

In recent years,China’s wind power industry has entered a stage of rapid development.China’s total installed capacity of wind turbines and annual new installed capacity rank first in the world,but at the same time,a large number of wind turbine operation and maintenance problems have emerged.For megawatt wind turbines,gearbox is one of the most frequent fault components.Fault of gearbox is which causes the longest downtime of the wind turbine and the largest economic loss,while fault of gear and bearing account for 80% of all types of gearbox fault.Therefore,conduct early fault diagnosis research on gearbox gears and bearings,analyze their vibration signals and analyze the cause and location of fault is important to reduce the operation and maintenance costs of wind farms,improve the stability of wind turbine operation and increase economic benefits significance.Taking the grid-connected double-fed wind turbine gearbox as the research object,the signal analysis is based on the monitoring data of SCADA system and wind turbines.The main types of fault are gear fracture,wear,fatigue,gluing,bearing outer ring fault,inner ring fault and rolling body fault.The main research methods are mathematical diagnosis methods and intelligent diagnosis methods.The main analysis methods are signal noise reduction,signal decomposition and reconstruction,feature extraction,spectrum analysis,classifier recognition,etc.The main work of the paper is as follows:(1)The basic structure and working principle of wind turbines are introduced,and the common failure types of wind turbines are analyzed.Among them,the fault rate of electrical systems,control systems,hydraulic systems and sensors is relatively high,while the failures of gear boxes,drive chains,and generators have the greatest impact.Focus on the inherent characteristics of gearbox gears and bearings,fault modes and characteristics of fault signal,analyze in detail the corresponding fault characteristics and theoretical fault characteristic frequencies of the gears and bearings.(2)Introduced the data acquisition process of the vibration signals of gearboxgears and bearings and the signal noise reduction,signal reconstruction method and feature extraction method of vibration signal.Aiming at the non-stationary and non-stationary vibration signals of gears and bearings,two adaptive signal denoising and reconstruction methods are proposed,which are stagewise orthogonal matching pursuit(St OMP)based on the approximate conjugate gradient pursuit(ACGP)and complementary ensemble empirical mode decomposition(CEEMD)based on EEMD.Aiming at the fault performance characteristics of gear and bearing fault signals in time domain,frequency domain and time-frequency domain,two fault feature extraction methods are proposed,which are multi-domain feature extraction method and multiscale entropy(MSE)construction method.By observing the results of spectrum analysis and feature extraction,we know that the above method has obvious noise reduction effect on the signal and high reconstruction accuracy.The constructed features reflect the main fault characteristics of the signal and do not lose the main information of the original signal.(3)According to the fault characteristics of gears and bearings,different feature screening methods and pattern recognition methods were proposed.Neighbor rough set theory(NRS)and correlation coefficient method are used to filter fault feature vectors,and a random forest(RF)classifier based on genetic algorithm(GA)parameter optimization and wavelet kernel extreme learning machine(WKELM)are used on fault classification and recognition of gear and bearing data.The above diagnostic models are applied to the analysis of actual fault data.And compared with the classification effects of other fault diagnosis models,the results show that in the identification of gear and bearing fault data,the performance of the fault diagnosis model constructed in this paper is outstanding,the accuracy of fault diagnosis is high,and the recognition speed of the diagnosis model is fast.