Research On Bearing Fault Diagnosis Based On Improved Empirical Mode Decomposition

At present,rotating machinery is widely used as the main equipment of production development.As an indispensable precision component in rotating machinery,whether the working state of bearings is normal will directly affect the normal operation of machinery,even cause the interruption of the production line,resulting in huge economic losses.Because of the bad working environment and complex working conditions,the bearing is very prone to failure.Failure to find and deal with the faults in the early stage will lead to a sharp reduction in bearing life and affect the efficiency and reliability of rotating machinery.According to the characteristics of convenient vibration signal acquisition and analysis,this thesis studies the bearing fault diagnosis method based on singular value decomposition,collective empirical mode decomposition and support vector machine,and uses the bearing data set of Case Western Reserve University to verify the proposed diagnosis method.Firstly,the mechanism and failure forms of rolling bearings are analyzed,the causes of each failure form are summarized,and the fault characteristic frequencies of different components are calculated theoretically.The signal preprocessing is studied by means of time-frequency analysis,focusing on the singular value decomposition(SVD)and ensemble empirical mode decomposition(EEMD)algorithms,and an improved EEMDSVD method is proposed for vibration signal preprocessing.The IMF component decomposed by EEMD is denoised twice by SVD to complete the reconstruction of the whole signal frequency band,thus reducing the impact of noise on the diagnostic accuracy of vibration signal.Then the classification model of support vector machine is used to identify the types of bearing fault samples.The model uses signal sensitive eigenvalues as inputs to identify fault modes.The improved EEMD-SVD method is used to preprocess the bearing vibration data.Combined with the characteristics of bearing fault,the kurtosis,singular spectrum entropy,energy entropy and fuzzy entropy of signal sensitivity are analyzed and extracted.These four eigenvalues are used as the input vectors of support vector machine to train the SVM model and complete the fault type classification.Genetic algorithm,particle swarm optimization algorithm and improved whale algorithm are respectively applied to optimize the SVM classifier,and the classification accuracy obtained is used as the objective function to judge the performance of the SVM classifier optimized by the three algorithms.The research shows that the classification accuracy of the support vector machine model optimized by the improved whale algorithm for ten bearing states reaches 98.6667%,which is significantly higher than the SVM optimized by the other two algorithms.Finally,a bearing fault diagnosis platform is designed based on MATLAB software.The above time-frequency domain analysis,collective empirical mode decomposition algorithm and SVM classification algorithm are integrated into the software design.The software realizes a human-computer friendly interface,which facilitates the user to calculate and analyze the collected vibration data,and can use the software platform to complete the identification of bearing fault types...