Research On Fault Diagnosis Of Wind Turbine Main Bearing Based On Generative Adversarial Network

As a clean energy source,wind energy has developed rapidly in recent years.As large-scale wind turbines are put into operation,the number of operational failures has also increased.The main bearing failure is one of the most costly failures in wind turbine maintenance,and a quick and effective diagnosis is an effective measure to improve the economic benefits of wind farms.But the number of main bearing failures is relatively low in all kinds of faults of wind turbines,there is a serious data imbalance problem,which brings great difficulties to the use of data mining methods to determine the type of failure.The data enhancement method based on the Generative Adversarial Network(GAN)currently shows strong ability of data generation.It learns the distribution characteristics of the original samples and generates new fake samples by game training.Based on GAN,this paper implements an enhanced network of vibration signal data for main bearing faults of wind turbine.The main research work is as follows:(1)By improving the adaptability of the Auxiliary classifier Generative Adversarial Network(ACGAN),an improved ACGAN data enhancement framework based on one-dimensional deep convolution is proposed.In this framework,gradient penalty is introduced to improve its stability during training;A pooling layer is introduced in the discriminator network to improve its ability to extract data features in multi-class fault scenarios.The simulation results show that the proposed improved ACGAN framework can effectively learn the distribution characteristics of the main bearing fault samples of wind turbines.Compared with the original framework,the training process is more stable and the quality of the generated data is improved.The generated data is used to enhance the original unbalanced sample set,which effectively improves the fault diagnosis accuracy of the main bearing of the wind turbine.(2)Considering that the vibration signal of the main bearing failure of the wind turbine has timing characteristics,in order to better adapt to the time series information,on the basis of the improved ACGAN data enhancement framework,an improved LSTM-ACGAN data enhancement framework is proposed.In the discriminator network,a combination structure of deep convolutional neural network(DCNN)and long short-term memory network(LSTM)is used,DCNN is used to extract fault feature information,and combined with the ability of LSTM to effectively model time series information to learn fault features.The simulation results show that,compared with the improved ACGAN framework,the convergence speed is faster in the model training process.The classification ability of the discriminator and the generation ability of the generator have been further improved;When the training set size is decreasing,the generator has better generation ability,which can effectively improve the quality of generated data caused by insufficient training samples.The improvement of the LSTM-ACGAN framework improves the fault diagnosis accuracy of the main bearing of the wind turbine.