Research On Intelligent Diagnosis Method For Bearing Fault Of In-wheel Motor Based On Time-frequency Analysis

Against the background of the national strategy of "carbon peak and carbon neutrality",new energy vehicle industry has shown a booming trend.With the rise of distributed drive technology,in-wheel motor driven electric vehicles have attracted widespread attention in the industry.To avoid vehicle safety hazards caused by key component failures of in-wheel motor and solve the problem of difficult fault diagnosis caused by high complexity of monitoring signals in noisy environments,the typical faults of in-wheel motor bearing are regarded as a starting point,feature extraction and fault diagnosis methods are developed based on time-frequency analysis to achieve intelligent management of the safety of in-wheel motor driven electric vehicles.The main contents are as follows:Firstly,the operational situation and common faults of in-wheel motor is analyzed,and the characteristic performance of bearing faults for in-wheel motor are discussed.The in-wheel motor test bench is rebuilt to simulate its actual operating conditions,providing data assurance for subsequent research.Secondly,to address the issue of poor early weak fault feature extraction performance of in-wheel motor bearing,a fault feature extraction method based on component weighted reconstruction and sparse non-negative matrix factorization(SNMF)is proposed.The implementation process of local mean decomposition is analyzed,and a new fusion index CIHis proposed to evaluate the component signals from multiple angles and adaptively weight the reconstruction,improving both the reconstruction effect of the component signals and the enhanced expression of fault features.The smoothing coefficient of the reconstructed signal is defined using the singular value variance ratio,and the implicit subspace of its time-frequency energy matrix is discriminatively estimated to solve the dimension selection problem.The Itakura-Saito distance and sparse constraint are introduced to construct an SNMF algorithm with feature learning ability,which decomposes the time-frequency energy matrix and complete fault feature extraction.Simulation and experimental verification proved the superior performance of the proposed method compared to traditional methods.Thirdly,in consideration of the difficulty in extracting fault features under strong background noise,a fault feature extraction method based on optimized singular spectrum decomposition and enhance multipoint optimal minimum entropy deconvolution adjusted(EMOMEDA)is proposed.To overcome the defect of determining the number of components in the iterative process of singular spectrum decomposition,a time-frequency synthesis index TCI is constructed to optimize the frequency band partitioning rules,and the envelope spectral peak indicator is used to adaptively select sensitive singular spectrum components,completing the preliminary extraction of fault features.To solve the problem of point number attenuation in deconvolved signals,the waveform extension strategy is improved to achieve boundary compensation of signals,and quadratic deconvolution operation is used to enhance the extraction of fault features,laying the foundation for intelligent fault diagnosis.Simulation and experimental data comparison analysis demonstrate that the proposed method has good robustness and practicality in fault feature extraction for in-wheel motor bearing.Finally,to satisfy the input requirements of state recognition,the synthetic weight detection index is used to select characteristic parameters that can represent the faults of in-wheel motor bearing with high sensitivity in both the time and frequency domains.The energy values obtained by wavelet packet decomposition in the time-frequency domain are calculated as time-frequency characteristic parameters,and a multi-domain feature fusion characteristic parameter set for in-wheel motor fault is constructed to achieve multi-angle description of fault features.The change trends of weights and thresholds during the learning process of extreme learning machines are analyzed,and the sparrow search algorithm is used to optimize and correct them.A fault diagnosis method based on multi-domain feature fusion for in-wheel motor bearing is established to achieve intelligent fault diagnosis in noisy environments.The effectiveness and accuracy of the proposed method are verified through open source and in-wheel motor experiment data.