Study On Condition Monitoring And Diagnosis Of Wind Turbine Gearboxes Based On Deep Fusion Of Multi-dimensional Features

Wind turbines are usually installed in remote areas with harsh environments.They have been affected by rain,snow,sandstorms,and lightning strikes for a long time.They are prone to failure,and their health during operation is critical.As a key component of wind turbine equipment,wind power gearbox occupies an important position in the fan drive system,and its health status directly affects the service life of wind turbine.Therefore,condition monitoring and diagnosis of wind power gearboxes is an important research topic.At present,the condition monitoring and diagnosis of wind power gearboxes mainly uses a combination of Data Acquisition and Detection System(SCADA)and Condition Monitoring System(CMS).The overall health status of the box is used for condition monitoring,and then the CMS vibration data is used for fault diagnosis.With the expansion of wind farm scale and the accumulation of wind turbine operation time,the scale of wind power SCADA and CMS data has grown rapidly.The wind power industry has entered the era of big data operation and maintenance.The traditional wind power gearbox condition monitoring based on signal processing and machine learning The method of diagnosis has been difficult to deal with.The deep learning method is a kind of intelligent analysis method with great potential.By constructing a deep network structure,features can be automatically learned from massive data,which provides new ideas for wind power gearbox condition monitoring and diagnosis.However,due to dense wind turbine sensors,complex gearbox structure,and variable working conditions,the information contained in wind power monitoring data is extremely rich and complex.The current deep learning methods are difficult to analyze the multi-dimensional information existing in the data due to the limitation of a single structure Comprehensive utilization makes it difficult to make accurate judgments on the health status of wind power gearboxes,and still cannot meet the needs of wind power gearbox condition monitoring and diagnosis.In view of the above problems,this paper proposes method of condition monitoring and diagnosis of wind turbine gearboxes based on fusion of multi-dimensional features.The deep learning method is used to deeply integrate the multi-dimensional information such as space,time and size in the SCADA monitoring data of the wind power and the vibration data of the CMS to achieve effective monitoring and diagnosis of the wind power gear box status.The main research of the paper is as follows:(1)Aiming at the characteristics of wind power SCADA monitoring data with many variables and long time,a wind power gearbox condition monitoring method based on the fusion of spatio-temporal characteristics of SCADA data was proposed.This method uses the Pearson correlation coefficient method to screen SCADA variables,and then uses deep learning methods to construct a convolutional neural network to fuse the spatial features in SCADA data,establish a gated recurrent network fusion time feature,and propose an overall residual as a condition monitoring Indicators,use the EWMA(Exponentially Weighted Moving-Average)control chart to monitor the condition of the wind power gearbox.The validity of this method is verified by comparing the SCADA data measured in the wind field with other deep learning methods.(2)Aiming at the characteristics of rich and time-varying CMS vibration data segments,a fault diagnosis method for wind power gearboxes based on time-frequency fusion and attention mechanism of CMS data is proposed.This method first characterizes the original vibration signal as a time-frequency map by wavelet packet transformation,then incorporates the deep learning method into a two-layer convolutional neural network to fuse the segmented features of the CMS data,and fuses the transformed features through a two-way gated recurrent network.Then the mechanism is dynamically extended and fused by introducing a mechanism,and the validity of the method is verified by comparing the power transmission test bench data and wind field measured CMS vibration data with other deep learning methods.(3)On the basis of the wind turbine transmission system detection and diagnosis system developed earlier in this group,a wind power gearbox condition monitoring and diagnosis module based on multi-dimensional information deep fusion was developed.This system module can be used to manage,display and analyze wind power monitoring data,to realize the wind power gear box condition monitoring of SCADA data spatiotemporal feature fusion and CMS data time-frequency integration and attention mechanism fault diagnosis of wind power gearbox proposed in this paper.The effectiveness of the system module is verified by measured wind field data.Finally,this research work is summarized and the future research directions are prospected.