Rolling Bearing Fault Diagnosis Based On Improved VMD And Graph Attention Neural Network

Rolling bearings are widely used in wind power generation,railway transportation,aerospace and other fields that are closely related to people’s production and life.As machinery and equipment are often in a long working condition,once the bearing failure occurs,it will cause significant economic losses or even threaten the safety of the entire machinery and equipment,so timely fault diagnosis of the bearing can greatly avoid the safety of equipment and personnel problems.In this paper,we will take rolling bearings as the object of research,and carry out research on several aspects such as fault mechanism analysis,signal decomposition and reconstruction and fault diagnosis respectively,the main contents are as follows.Firstly,in order to improve the accuracy of the identification of rolling bearing faults,an explicit kinetic method is applied to analyse the rolling bearing fault mechanism.Based on the rolling bearing failure mechanism model,a finite element model of the bearing inner ring and outer ring with a single point failure is established,followed by the simulation of the established finite element model using ANSYS/LS-DYNA explicit dynamics module and obtaining its simulation signal,the signals obtained are then analysed in an envelope to determine the initial type of rolling bearing fault and to provide a theoretical basis for rolling bearing fault diagnosis.Secondly,a Whale Optimization Algorithm(WOA)based Variational Mode Decomposition(VMD)signal decomposition and reconstruction method is proposed to address the problem of endpoint effect and modal confusion in signal decomposition.The vibration signal is decomposed by VMD,and the penalty factor and the number of decomposition layers of the VMD decomposition parameter are determined adaptively to obtain multiple intrinsic modal components(IMF);the different IMF components obtained from the decomposition are analysed by Pearson correlation coefficient method,and the signal with less noise component is selected from the multiple IMF components for reconstruction.By comparing the simulated data,we know that the method has a large signal-to-noise ratio and small mean square deviation compared with other noise reduction algorithms,and can retain the effective information to a great extent.Thirdly,in order to mine the relationship and dependence between data to improve the accuracy of fault diagnosis,this paper proposes a fault diagnosis method of rolling bearings based on graph attention neural network(GAT)model.The network structure model of GAT fault diagnosis was constructed,and the Attention architecture was used to perform node classification of graph structure data,improve the weight allocation of sensitive information,and select reasonable network structure parameters.The experimental results show that compared with other prediction models,the fault diagnosis accuracy and accuracy rate of GAT are significantly higher than other prediction models.Finally,to verify the effectiveness of the proposed method,an experimental bench for rolling bearing fault diagnosis is set up,and the WOA-VMD signal decomposition and reconstruction method and the GAT fault diagnosis method are experimentally verified in turn to prove the effectiveness and superiority of the proposed method.