Research On Fault Diagnosis Method Of Wind Turbine Generator Bearing Based On Deep Transfer Learning

In order to achieve the goal of national 2030 carbon peak 2060 carbon neutrality,the development of wind power industry is in full swing,and the safety of wind turbines is becoming more and more important.The generator bearing is one of the core components of the wind turbine.Due to the complex and changeable operating environment of the wind turbine,the failure of the component occurs frequently,which seriously affects the stable operation of the wind turbine and causes economic losses and even safety accidents in the wind farm.Therefore,it is of great significance to study the fault diagnosis method of wind turbine generator bearing to improve the utilization rate of wind turbine,optimize the maintenance strategy and reduce the maintenance cost.This paper first analyzes the failure mechanism of wind turbine generator bearings,and calculates the fault characteristic frequencies of the two types of bearings used in the experiment to prepare for subsequent experiments.Then,a vibration signal processing method IVMD-MOMEDA combining variational mode decomposition(VMD)and multi-objective minimum entropy deconvolution adjustment(MOMEDA)is proposed for signal noise problems.IVMD-MOMEDA uses a composite evaluation index to measure the quality of the reconstructed signal,and uses genetic algorithm to optimize the parameters of VMD.And the reconstructed signal is processed by MOMEDA to further reduce noise.In this paper,the bearing data set of Case Western Reserve University(CWRU)and the measured data set of wind turbines are used for experimental analysis.The results show that the method can accurately highlight the fault impact components in the collected signals and effectively identify the fault types of wind turbine generator bearings.Finally,this paper designs a deep convolutional neural network model with multi-scale convolutional module(MDCNN)based on deep transfer learning,and proposes a fault diagnosis method based on MDCNN.Through the design of data enhancement,wide convolution layer,multi-scale convolution module,adaptive batch normalization(AdaBN)and multiple kernel maximum mean discrepancy(MK-MMD),the generalization performance of the model is improved,and high-precision and high-stability fault diagnosis is realized.The CWRU bearing data set is selected to carry out anti-noise experimental analysis,small sample experimental analysis and domain adaptive experimental analysis by adding noise and establishing transfer learning tasks.The results show that MDCNN has excellent performance in strong noise,small sample and crossdomain fault diagnosis.In addition,this paper also changes the main parameters of the key structure of the model for comparative experiments,analyzes the influence of the main parameters and verifies the effectiveness of the key structure design.