Research On Fault Diagnosis Of Wind Turbine Gearbox Based On Multi-Scale Learning

Today the world is facing climate change,energy security and other problems.The development of clean energy industry,is the focus of the industrial configuration of countries.Wind energy as one of the main clean energy,the development of wind power industry is imperative.As a complex large rotating electromechanical equipment,wind turbine has many internal transmission parts.Its operating environment is extremely harsh,so the chance of failure of the unit components is extremely high.The smooth even safe operation of the unit is easily effected by the failure of the unit components,so the fault diagnosis of wind turbine is an essential part of the wind power industry.Therefore,the fault diagnosis of wind turbines is an essential part of the wind power industry.It is extremely important to study the fault diagnosis of gearbox,which bear most of the load of wind turbine,and to carry out fault diagnosis of wind turbine gearbox with high precision and high efficiency.Generally speaking,the gearbox fault characteristics in the vibration signal will be interfered by numerous noises,especially the early faults.Therefore,it is especially important to study how to reduce the noise interference in the signal and enhance the fault characteristics.Similarly,to improve the diagnostic accuracy and efficiency and reduce the maintenance cost,it is also necessary to study the application of artificial intelligence in the field of fault diagnosis.To address the above problems,this thesis takes wind turbine gearboxes as the research object and proposes a fault diagnosis method based on multi-scale adaptive signal processing method and multi-channel fusion multi-scale dynamic adaptive residual learning model to achieve high precision intelligent fault diagnosis of wind turbine gearbox faults,as follows:(1)The common fault forms of wind turbine gearboxes are analyzed.The vibration signal characteristics of wind turbine gearbox drive process are studied,and the signal modulation phenomenon is the main focus.This provides a theoretical basis for wind turbine gearbox signal processing and fault diagnosis.(2)Through the basic theoretical study of variational mode decomposition(VMD)method and singular value decomposition(SVD),the multi-scale adaptive signal processing(MSASP)method is proposed.The multi-scale adaptive signal processing(MSASP)method is proposed to remove the multi-scale noise distributed at different frequency scales.Firstly,in order to remove the strong background noise in the vibration signal and decompose the original signal at different scales,a signal denoising method based on permutation entropy variational mode decomposition(PE-VMD)is proposed.Secondly,in order to reduce the multi-scale noise interference and enhance the fault characteristics,the signal denoising method based on adaptive singular value decomposition(ASVD)is proposed,which further adaptively denoises each group of sub-signals obtained by the PE-VMD method.The reconstructed signal is obtained after noise reduction.The effectiveness of the MSASP method is verified by conducting noise reduction experiments on the simulated signal and the vibration signal of the in-service wind turbine gearbox with early wear of the parallel stage low-speed shaft gear.The two experiments show that the MSASP method has better performance in reducing the strong background noise and multi-scale noise in the signal compared with other traditional signal processing methods,and is applicable to the fault diagnosis of wind turbine gearboxes.(3)In the vibration signal of the wind turbine gearbox,the vibration signal in one direction does not provide comprehensive fault information,because the focus of the fault information reflected by the vibration in different directions is different.In order to learn gear fault characteristics comprehensively and accurately with multi-scale reference to multi-directional wind turbine gearbox vibration signals,the basic theoretical research on convolutional neural network(CNN)and residual learning(RL)is carried out,and multi-channel multi-scale dynamic adaptive residual learning(MC-MSDARL)model is established in this thesis.Firstly,the multi-scale dynamic adaptive convolutional neural network(MSDACNN)model is established in order to dynamically and adaptively adjust the weights of convolutional kernels at different scales and improve the multi-scale feature extraction capability.Secondly,based on the MSDACNN,a multi-scale dynamic adaptive residual learning(MSDARL)model is further proposed to accelerate the training speed of the model.Finally,in order to integrate the multi-directional data organically,the MC-MSDARL model is established based on the MSDARL model.The MC-MSDARL fault identification model is established based on the MSDARL model in order to organically fuse the multi-directional data and extract the comprehensive fault feature information from the gearbox.In the wind turbine gearbox gear fault identification and classification experiments,the fault identification accuracy of MC-MSDARL model reaches 97%,which is better than the existing methods.The experimental results demonstrate that the MC-MSDARL model has superior fault identification performance by fusing multiple time-scale information on multiple channels.(4)To further improve the fault diagnosis performance of the model,the MSASP-MC-MSDARL fault diagnosis model is established by combining the MSASP method with the MC-MSDARL model based on its excellent denoising performance and enhanced fault features.Firstly,the signal is noise-reduced based on the MSASP method to enhance the gearbox fault features.Secondly,the MC-MSDARL model is used to extract and fuse multi-channel and multi-scale features of the noise-reduced signal to achieve accurate fault diagnosis.Finally,in the diagnosis experiment of single fault type,the MSASP-MC-MSDARL model diagnoses the single fault of wind turbine gearbox with98.8% accuracy,which is nearly two-thirds less than the single MC-MSDARL model.In the experiment of compound fault diagnosis,the MSASP-MC-MSDARL model diagnoses the compound fault of wind turbine gearbox with 95% accuracy,which is much higher than other fault diagnosis methods.From the above experimental results,it can be concluded that the MSASP-MC-MSDARL model further optimizes the model fault diagnosis performance by integrating multi-frequency scale and multi-time scale on the basis of MC-MSDARL model.