Research On Feature Extraction And Intelligent Diagnosis Methods Of Rolling Bearing Fault Signal Of Wind Turbine

With the rapid development of wind power generation technology in China,the monitoring and fault diagnosis of wind turbine generator bearings is the trend of the times.This paper takes the wind turbine bearing as the research object,carries out the research from three aspects of signal processing,feature recognition and intelligent diagnosis,and proposes an Adaptive Multivariable Variational Mode Decomposition(AMVMD)model based on slime mold algorithm(SMA),Kernel limit learning machine(KELM)based on improved Sparrow Search Algorithm(SSA)and an intelligent di agnosis method based on AMVMD and improved Deep Discriminative Transfer Learning Network(IDDTLN).The applicability of the proposed methods is verified by the actual wind turbine bearing fault data and multiple public data sets of rolling bearings.The article studies the following three aspects of work:(1)To address the shortcomings of multivariate variational modal decomposition,it is necessary to manually set the decomposition parameter K and penalty factorα,and propose an adaptive multivariate variational modal decomposition algorithm to achieve adapt ive parallel decomposition of multi-channel signals.The specific steps are as follows:Taking kurtosis-mutual information-envelope entropy(EKM)as the o bjective function,and using the slime mold algorithm to find the optimal solution of the decomposition parameter K and penalty factorαin the search space,the optimal mode is selected by reconstruction judgment basis.Finally,the fault frequency is ide ntified by the Teager energy operator demodulation.The availability of this model has been verified by simulation signals and actual bearing signals of wind turbines,and compared with other similar models,this method has the best decomposition effect an d can accurately find the fault frequency.(2)The two parameters of the kernel limit learning machine algorithm which are the penalty factor C and the kernel function widthσ~2 are difficult to effectively select,a fault diagnosis method of the kernel limit learning machine optimized by the Von Neumann sparrow search algorithm is proposed.The AMVMD algorithm is used to decompose the fault data o f the wind turbine bearing,and the mixed entropy of the optimal modal component is calculated to construct the eigenvector.The Von Neumann topology is introduced into the sparrow search algorithm to optimize the kernel limit learning machine as a classifier.The effectiveness of the model was verified through actual bearing data of wind turbines and comparison with other models,with an accuracy of99%.(3)A bearing fault diagnosis based on AMVMD and improved depth discrimination transfer learning networ k(IDDTLN)is proposed to solve the problems that are difficult to accurately diagnosis under variable operating conditions.A feature extraction method combining attention mechanism module and depth residual convolution neural network is proposed and appl ied to the model.The distance metric function combined with maximum mean discrepancy(MMD)and correlation alignment(CORAL)is used to align the marginal distribution difference and conditional distribution difference between the source domain and the ta rget domain.Experimental data have proved the advantages of the signals processed by the AMVMD algorithm as the input of the improved depth discrimination transfer learning network,and the feasibility of bearing fault diagnosis under variable operating c onditions is also verified by the bearing datasets and actual bearing signals from wind turbines.