Research On Fault Diagnosis Method Of Large Wind Turbine Bearing Based On Improved Generation Adversarial Networks

As a key component of wind turbine generator,bearing has become one of the components with high failure rate due to its bad operating environment.Therefore,it is of great significance to diagnose the fault of wind turbine generator bearing.Aiming at the shortcomings of the traditional feature extraction and machine learning based mechanical fault diagnosis models,this paper takes wind turbine generator bearing as the research object,and takes the emerging deep learning technology as the theoretical basis in recent years,the research directions are from the following aspects:applicability and stability of the deep network model,fault classification of sensor information,improvement of feature learning ability and diagnosis recognition rate of fault diagnosis model,and feature learning of small sample data,in this way,the influence of manual participation is reduced,the operation efficiency of the system and the accuracy of fault identification are improved,and the actual data of the wind farm are verified.This paper focuses on the following four aspects:(1)Aiming at that in the huge bearing vibration data obtained by the sensor,there are data that are not conducive to bearing fault diagnosis,such as null value,repeated value and record redundancy,The data preprocessing technology is selected to filter and extract data suitable for analysis and mining from a large number of original data,and label such data,in order to facilitate the smooth follow-up fault diagnosis.(2)To solve the problem of insufficient fault data acquisition and imbalance of fault categories,a network model based on improved generation adversarial network is proposed,by using the feature learning and data generation capability of generating adversarial network,the feature information helpful to fault diagnosis can be learned from the pre-processed original fault data,and the feature learning and data generation can be combined in a network structure,the network model is gradually adjusted and optimized through the generation function and loss function of the network,so that the network model is more intelligent and the generated data is more representative and random.(3)Aiming at the bearing fault diagnosis model driven by data,a fault diagnosis model based on two-dimensional convolutional neural network is proposed,the powerful local feature extraction capability of convolutional neural network is used to capture fault features,and the fault mode of bearings is learned to realize accurate fault identification,comparing different parameters under the two-dimensional convolution neural network diagnostic accuracy,the selection of the highest accuracy network model,according to the training precision and training error of the model,the parameters selected by the model are judged to be suitable for bearing fault diagnosis,so as to maximize the accuracy of bearing fault diagnosis.(4)Based on the actual complex wind farm generator bearing data,get the training model,the two models are suitable for bearing direction.The real generator bearing data is taken as the input of the model,and the output of the model is matrix data,according to the number in the output matrix and the fault label added during the pretreatment,the fault diagnosis of the fan generator bearing at different positions,such as rolling body,inner ring and outer ring,is realized,which effectively avoids expert experience and characteristic engineering.