Research On Fault Diagnosis Method Of Wind Power Gearbox Based On Deep Learning

With the rapid development of wind power industry,the reliability of wind turbines is increasingly demanded.Gear box is an important transmission component widely used in wind turbine transmission system.In actual operation,the wind power gearbox bears dynamic heavy load,and the operation environment is complex,which makes the bearing and other parts prone to failure.Research on fault diagnosis technology for wind power gearbox has important theoretical significance and application value for improving the safety and reliability of wind power equipment and reducing maintenance costs.This paper mainly studies the fault diagnosis technology of wind power gearbox based on deep learning.The main research work is as follows:Firstly,the application of traditional fault diagnosis methods and deep learning methods in the field of fault diagnosis is analyzed,and the existing problems of fault diagnosis of wind power gearbox are studied.The characteristics of fault diagnosis classification algorithms based on signal processing and pattern recognition are analyzed in detail.On the basis of in-depth study of several typical deep learning algorithms,the fault diagnosis ability of deep learning algorithms under different hyperparameters is analyzed and discussed,and the effect of different hyperparameters on the network is studied.Secondly,aiming at the deficiency of feature extraction ability of denoising autoencoder in small sample data set,the basic theory of denoising autoencoder is analyzed,a new data preprocessing method is studied,and the selection of hyperparameters in the network is optimized by decreasing the hyperparameters layer-by-layer.By analyzing the advantages and disadvantages of L1 regularization and L2 regularization,the two regularization methods are combined to increase the sparsity of the network and improve the generalization ability.Using the powerful feature extraction ability of denoising autoencoder and the conciseness of k-nearest neighbor algorithm,the extracted deep feature is directly applied to k-nearest neighbor algorithm,which improves the fault recognition rate.Finally,aiming at the slow convergence rate of the training process of convolution neural network,an adaptive regularization coefficient method is proposed and applied to convolution neural network.The regularization coefficient is inversely proportional to the gradient of loss function,and the adjustment of regularization coefficient can be adapted to the gradient of current iteration objective loss function in stochastic gradient descent method.The experimental results show that the convolutional neural network based on adaptive regularization coefficient method reduces the time required for network training and improves the accuracy of fault diagnosis.