Research On Intelligent Diagnosis Of Wind Turbine Gearbox Bearings Based On Deep Learning

Under the background of the comprehensive development and utilization of clean energy,as a representative product of wind energy development technology,wind turbines have been widely used.However,the wind turbines work in complex conditions perennially,the gearbox part of the transmission system increases the risk of damage inevitably.As one of the most easily damaged parts in the gearbox,the fault of the rolling bearing will not only aggravate the damage process of the gearbox,but also lead to the accident.Therefore,the research on fault diagnosis of rolling bearing in gearbox of wind turbine is of great significance.In recent years,intelligent diagnosis methods represented by deep learning have developed rapidly.However,there are some problems in the application of deep learning algorithm to the fault diagnosis of wind turbine gearbox bearing,such as the wind turbine signal samples are insufficient,and it is also difficult to obtain wind turbine fault labels,and because the fault features of the collected data are not clear,it is difficult to carry out multi data fusion.In order to optimize the intelligent diagnosis algorithm of wind turbine gearbox bearing.The specific research contents of this paper are as follows:(1)Aiming at the problem that it is difficult to obtain the fault data of wind turbine gearbox bearing in practical work,a fault diagnosis method based on wasserstein generative adversarial networks(WGAN)and semi-pooling convolutional neural networks(SPCNN)is proposed.Firstly,in order to adapt the signal samples to the input dimension of the model,the one-dimensional time-domain signal is transformed into two-dimensional matrix by dimension transformation.Then,training generator to generate two-dimensional matrix with similar features to the original sample,and adds the generated samples to the original training sample to expand the training sample set.Finally,the proposed algorithm is verified by experimental data.The results show that the data expansion method based on WGAN can improve the feature diversity of the training set.In the classification process,SPCNN effectively reduces the training time.(2)Aiming at the problem that it is difficult to obtain the label of wind turbine gearbox bearing samples,a cross equipment fault diagnosis method based on improved domain adaptive neural network(DANN)is proposed.Firstly,the data expansion method based on WGAN is used to generate the data of source domain and target domain.Then,in order to extract the features that can be transferred more efficiently,a convolutional neural network(CNN)with residual structure is constructed as the feature extractor of the transfer learning model.Finally,the proposed algorithm is verified by experimental data.The results show that the proposed algorithm can effectively complete the cross domain diagnosis task after data expansion.(3)On the basis of single sensor data,multi-sensor data with multi angle information is further considered.In order to avoid the influence of signals with unclear features in multi-sensor data on the overall model.At the same time,aiming at the problem of large signal noise of wind turbine gearbox bearing and high robustness of data fusion model,a multi-sensor data fusion diagnosis model based on improved multi-scale CNN(MSCNN)is proposed.The proposed model introduces multi-scale convolution kernel in feature extraction,which improves the robustness in the process of data fusion.In addition,in terms of multi-scale fusion,global average pooling(GAP)is adopted to retain the spatial structure of the feature map and better adapt to the fusion process of each branch structure.Then,the proposed algorithm is verified by experimental data.The results show that the proposed model has strong robustness and can effectively solve the problem of multi-sensor data fusion diagnosis of wind turbine gearbox bearing.