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

As a key component of the wind turbine,the bearing is one of the components with a high failure rate in the wind turbine due to its poor operating environment.It is estimated that about 80%of mechanical faults in generators,gearboxes,and drive shafts are caused by bearing failures.Therefore,it is very important to carry out condition monitoring and fault diagnosis for wind turbine bearings.This paper mainly analyzes the shortcomings of the existing traditional diagnostic methods for wind turbine bearings.Based on the time-frequency analysis of bearing vibration data and the principle of deep learning neural network method,a fault diagnosis plan based on deep learning is proposed.Firstly,study on the basic composition and working principle of the wind turbine is carried out,the fault modes and characteristic frequency of the wind turbine bearing are introduced.The bearing vibration data is analyzed in time domain,frequency domain and time-frequency domain,and the support vector machine model is used to identify the classification.Secondly,this paper investigates the principle of deep belief network,by which a bearing fault diagnosis modelis developed.The time-frequency spectrum of the bearing vibration data subjected to the short-time Fourier transform is input to the deep belief network model,the model will respondly output the identification code of the bearing various fault types.In order to study the anti-noise performance of the deep belief network diagnostic model,The bearing vibration data samples under different SNR and the bearing vibration data samples after wavelet denoising are input into the deep belief network model respectively after taking the operating conditions and working environment of the wind turbine bearings into consideration.Proctically the faults detected by vibration monitoring are mixed fault types,it is necessary to input a mixed fault data sample,formed by mixing the vibration data with different fault characteristics of the bearing,to the deep belief network fault diagnosis model,so as to judge the recognition performance of the deep belief network fault diagnostic model for mixed fault samples.Finally,the working principle of convolutional neural network is studied,and a bearing fault diagnosis model based on convolutional neural network is established.The time-frequency spectrum of the bearing vibration data is input to the convolutional neural network model,the model will respondly give the identification code of the bearing various fault types as its output.Similarly the bearing vibration data under different SNR and the bearing vibration data after wavelet denoising are respectively input into the convolutional neural network model and to study the anti-noise performance of the convolutional neural network model.Then the mixed fault data samples with different fault characteristics of the bearing are input into the convolutional neural network model,so as to study the recognition performance of the convolutional neural network model for the mixed fault samples.The results show that the convolutional neural network model can automatically find the fault characteristics according to the bearing vibration data under different loads and different speeds,and the model can identify the faults of ball,inter race and outer race of bearing;it is helpful to avoid the expert experience and feature engineering,which makes it more versatile and practical,Thus the model has the great potential for efficient on-site bearing fault diagnosis.The anti-noise performance and hybrid fault identification performance of the convolutional neural network model is also superior to the deep belief network model.