Research On Wind Turbine Fault Diagnosis Based On Deep Transfer Learning

Since the wind turbine works in the harsh environment for a long time,and its internal structure is complex,especially the transmission components in the core gearbox are often subjected to alternating load,which is more likely to cause a series of faults.Once the gearbox fails,the performance of the whole wind turbine will be greatly affected,resulting in an increase of maintenance costs and loss of benefits during downtime.Therefore,it is necessary to monitor and diagnose the wind turbine in real time.The research object in this paper is the gearbox of wind turbine transmission system.The main contents of the research are as follows:1.This paper summarizes various fault diagnosis methods of wind turbine and puts forward the research direction.The concept of deep transfer learning is introduced,and the technology roadmap and research scheme of wind turbine fault diagnosis method based on deep transfer learning are determined.2.Aimed at the problem that traditional feature extraction methods are difficult to extract fault features when the vibration signal of the wind turbine gearbox is disturbed by background noise,a composite variational mode entropy fault feature extraction method is proposed.In this method,the original signal is subjected to variational mode decomposition,and components with high correlation with the original signal are selected for multi-scale analysis to extract fuzzy entropy at different scales.Composite variational mode entropy can effectively suppress the complex interference noise,and provide sufficient distinguishability for fault identification.3.The irregular direction and load change of wind lead to the complicated internal working conditions of wind turbine,and it is difficult to obtain a large number of vibration signal data with labels of wind turbine.In this paper,the concept of transfer learning is introduced,and a wind turbine fault diagnosis method based on correlative feature domain adaptation is proposed.This method improves the accuracy of fault identification by minimizing the data distribution discrepancy of the vibration signal feature sample sets of wind turbine under different working conditions.4.Aimed at the limited ability of traditional transfer learning method to extract the deep features of vibration signals,a wind turbine fault diagnosis method based on one-dimensional convolutional neural networks domain adaptation is proposed.This method uses the data features of the vibration signal under auxiliary working conditions through one-dimensional convolutional neural networks to help learn the data features of the vibration signal under the target working conditions.It can effectively solve the problem of the low accuracy of wind turbine fault identification under variable working conditions,and the effectiveness of the proposed method is verified by experiments.