Research On Fault Diagnosis Method For Wind Turbine Gearbox Under Complex Working Conditions

With the implementation and promotion of clean energy strategy in China,wind power generation has been rapidly developed and widely used.Because most wind turbines work in complex and changeable harsh environment,the gears and bearings,the core components of the gearbox of its transmission system,are affected by high cycle fluctuation load for a long time,which easily leads to local faults and long-term shutdown of the wind turbine.Therefore,it is very important to strengthen the research on fault diagnosis methods of wind turbine gearbox under complex working conditions to ensure the stable operation of wind turbines and reduce economic losses.In addition,the wind turbine gearbox is affected by strong noise and changeable working conditions,and the fault signal has the characteristics of strong interference,high coupling and high dimension,so it is difficult to identify the fault effectively by traditional diagnosis methods.Based on the collected vibration signal data,this paper takes the gears and bearings in the wind turbine gearbox as the research objects,and studies the fault diagnosis methods of key components of the wind turbine gearbox under complex working conditions by using signal processing technology and deep learning technology.The specific contents are as follows:(1)Aiming at the weak fault signal characteristics of wind turbine gearbox and the overlapping interference among various fault characteristics,this paper proposes a compound fault diagnosis method of wind turbine gearbox gear and bearing based on RWSSA-VMD-RSSD.Firstly,VMD is used to decompose the original signal,and a new weighted index F is constructed to select the optimal modal component for reconstruction.Then the reconstructed signal is decomposed by RSSD,and the high and low resonance components including fault impact components are obtained.Then,the envelope analysis of resonance component is carried out and 1.5-dimensional spectrum is made,so as to identify the fault type.In addition,RWSSA algorithm is used to optimize the key parameters of the two algorithms adaptively,so as to avoid the uncertainty of manual experience selection.The simulation and experimental results show that the proposed method can accurately separate the compound fault features of gears and bearings of wind power gearbox and successfully diagnose the fault types.(2)Aiming at the problems of low accuracy of fault identification in traditional CNN models under strong noise environments and the complexity of model calculation caused by high-dimensional characteristic data,this paper proposes a fault diagnosis method of rolling bearing of wind power gearbox based on enhanced SVD combined with SAE-CNN.Firstly,the SVD reconstruction order is determined by combining singular value difference spectrum and correlation coefficient.Then,the original signal is preprocessed by enhanced SVD to filter out a large amount of noise and effectively retain sensitive fault features.Then,the high-dimensional signal data after preprocessing is transformed into low-dimensional feature data by SEA network,which reduces the complexity of subsequent calculation of the model.Finally,the rolling bearing fault is accurately identified and classified by CNN model.The experimental results and comparative analysis of the public bearing dataset in Paderborn University show that the fault identification accuracy of the proposed network model reaches97.2%,and the model has strong diagnosis stability and identification accuracy.(3)Aiming at the problem of single scale of feature extraction and poor generalization and anti-noise performance under working conditions,a fault diagnosis method of rolling bearing of wind turbine gearbox under working conditions based on two-dimensional image and MSCNN-BiLSTM is proposed in this paper.Firstly,the synchronous compressed generalized wavelet transform is used to transform onedimensional fault vibration signals into time-frequency two-dimensional images,which reduces the similarity of signal features among multiple faults.Secondly,MSCNN network with channel attention mechanism is used to deeply mine and extract fault feature information of different dimensions,which improves the generalization and stability of fault diagnosis model.Finally,BiLSTM model is used to extract time series feature information from data.Through the experimental analysis and comparison of CWRU bearing public dataset,the fault diagnosis accuracy of the proposed method reaches 100% under steady-state conditions,and has higher diagnosis accuracy,stronger generalization and noise resistance in the diagnosis of different intensity extra noise and cross-load conditions.