Fault Diagnosis Of Wind Turbine Gearboxes Based On Deep Domain Adaptation Neural Networks

With the rapid development of the global wind power industry,a large number of wind turbines have gradually come out of the warranty period and are prone to frequent failures.The operation and maintenance work of wind turbines have increasingly become the focus of development in the wind power industry.,wind turbines are widely distributed and remotely located.The traditional operation and maintenance mode is characterized with high costs and low efficiency,which makes it difficult to meet the current operation and maintenance needs of the wind power industry.As an intelligent analysis method with great potential,deep learning method can effectively mine the rich potential information from big data,and provide new solutions for intelligent operation and maintenance of wind turbines.The structure of wind power gearboxes is complicated which may lead to strong noise in the vibration signal.Under strong noise interference,the fault characteristics are weak and strongly coupled,which seriously affects the accuracy of deep learning fault diagnosis.In addition,the disassembly and installation of wind turbines are usually money and time consuming,and some wind farms have no long-term systematic collection of fault samples.While for the collected fault samples,there are scarce samples of labeled fault data for wind power gearboxes,which makes deep learning models prone to overfitting and poor generalization performance.Targeting the problems listed above,this paper combines the characteristics of wind turbine gearbox with weak and strong coupling,a few labeled fault samples and the advantages of deep learning,and carries out research on fault diagnosis method of wind turbine gearbox based on deep learning.The main research of the paper is listed as follows:(1)Aiming at the problem of complex structure,weak fault characteristics and strong coupling of wind power gearbox,a fault diagnosis method of wind turbine gearbox with enhanced capsule network is proposed.In this method,the original vibration data is transformed by wavelet packet into the time-frequency coefficient matrix as the input for the model;then the capsule network of vector calculation is employed as the main frame,and the convolutional kernel filtering is conducted to increase the perception field of view of the convolution kernel filter,and the overlap is introduced.The coefficient enriches the correlation information between the capsule channel layers,and further enhances the model’s ability to extract fault features.(2)Aiming at the problem of over-fitting and low fault diagnosis accuracy of traditional deep learning methods due to the scarcity of labeled fault samples in some wind farms,a fault diagnosis method for the wind power gearbox adapted to the network in the field of deep balance is proposed.This method deploys the traditional convolutional neural network as the main framework and combines the ideas of domain adaptation.By minimizing the difference in the distribution of source and target data in the high-dimensional feature space,the source and target data are approximately independent and identically distributed in the feature space,thus learning the feature information of a large amount of data from other wind farms is applied to the fault diagnosis of wind turbines in the target wind farm.(3)Combined with the wind power gearbox fault diagnosis method elaborated in this article,Py QT5 is introduced to design and develop a wind power gearbox intelligent fault diagnosis module,and the module function is verified by an example application.Finally,the article summarizes the work of this article and looks forward to the next research direction.