Research On Wind Turbine Fault Diagnosis Method Based On Empirical Wavelet And Improved Convolutional Neural Network

With the gradual development of wind power generation at national and international,the operation and maintenance of wind turbines has become a vital part of wind power development,which plays a very important role in both the efficiency of wind power generation capacity and investment costs.The biggest problem in wind turbine faults is mechanical faults,which are mainly difficult to identify and have long maintenance cycles.Therefore,for the characteristics of wind turbine vibration acceleration signals,the following studies are conducted in this paper for the fault diagnosis of wind turbines.(1)A new fault diagnosis scheme for wind turbine vibration acceleration signals is proposed.A signal resampling method is proposed to reduce the influence of the fan speed and sampling frequency on the fault characteristics in the fan sampling data.Empirical wavelet transform(EWT)is used to reduce noise of wind turbine vibration acceleration signals,and to process the sampled signals from the time-frequency domain can obtain better results and achieve the effect of enhancing the fault characteristics.Sensitivity analysis is performed on the characteristic parameters of the signal after Resampling Denoising(RD)processing,and the characteristic parameters with higher sensitivity are extracted and input into the Probabilistic Neural Network(PNN)belonging to the feedforward network.Realize the fault classification of the fan efficiently and quickly.(2)In order to avoid the uncertainty of the extraction of fan characteristic parameters,a method combining the resampling method,empirical wavelet denoising method and one-dimensional convolutional neural network is proposed to form a resampling denoising convolutional neural network(RDCNN).CNN is used for automatic learning of fault feature space information of wind turbine data,replacing the extraction of fault feature parameters and classifiers,which improves the accuracy and sensitivity of fault diagnosis.(3)In view of the non-linear classification problems that are common in the analysis of fan failure degree,a linear and visualized fan failure degree analysis method is proposed.For the vibration acceleration signal of the wind turbine,the empirical wavelet transform method is also used to reduce the noise of the signal and remove the redundant component information.Then,through the phase space reconstruction(PSR)method,the characteristic information of the signal is extended to a high-dimensional space that can show the dynamics and internal characteristics of the signal.Finally,the extended information is subjected to a covariance matrix singular value decomposition(SVD),and the linear analysis and visualization of the fault degree of the wind turbine are realized through the EWT-PSR-SVD method.