The Application Of Stacked Manifold Autoencoder In Fault Diagnosis Of Wind Turbine Gearbox

Wind turbines are the most important production equipment in the wind industry.Since the "Twelfth Five-Year Plan",China has strongly supported the development of large-scale wind turbines,which has led to rapid development of wind power equipment.In the meantime,the failure rate of wind turbine has also increased.In order to improve the reliability of wind turbine and reduce the possible economic losses,it is of great significance to monitor the working condition of wind turbine and detect the existed faults in time.This is also one of the hot issues in the field of Industrial Engineering.An accurate and stable fault diagnosis model can provide reliable data for enterprises to eliminate the existed faults,prolong the life cycle of equipment,and also reduce production costs.In order to improve the efficiency of equipment management and maximizes the economic benefits of enterprises,an efficient fault diagnosis scheme is proposed for wind turbine gearbox based on the background of wind industry.The wind turbine is driven by wind energy,and the wind direction and wind speed are always changing in the natural environment.Wind turbines have to work under multiple working conditions,and the vibration transmission inside the gearbox is non-stationary and multi-component due to the uncertain disturbances and measurement noise of wind turbine in time series signal.However,the vibration signals of the same fault are collected from the same working conditions in most existing literature.In this regard,a new manifold learning algorithm called SMAE(Stacked Manifold Autoencoder)is proposed to construct a stable fault diagnosis model for wind turbine gearbox.Firstly,vibration signals from multiple acceleration sensors are used to collect more reliable and accurate equipment information.Secondly,the Fast Fourier Transform(FFT)is applied to transform the original time-domain vibration signal into the frequency domain.Finally,the frequency domain signal is directly fed to the proposed SMAE deep network for fault diagnosis.SMAE is an improved method that combines deep learning and manifold learning.The loss function of tSNE(t-distributed stochastic neighbor embedding)is applied into autoencoder(AE)to update the parameters.Then,the discrepancy of probability distribution between the input layer and the hidden layer is reduced,and the hidden layer is supposed to retain more information of the input data.The proposed fault diagnosis method is applied to wind turbine fault diagnosis under the multiple working conditions,and the experimental results are compared with the peer methods.Experimental results show that the method proposed in this paper has better robustness and accuracy.