Research On Motor Rolling Bearing Fault Classification Method Based On CEEMDAN And GWO-SVM

As an important component in mechanical equipment,bearing is also a frequently-occurring component in mechanical equipment.Therefore,the early diagnosis of bearing fault status has great theoretical research value and practical significance.The purpose of the paper is bearing fault diagnosis.It mainly analyzes and researches the two parts of bearing fault information feature extraction and pattern recognition in fault diagnosis.The main work is as follows:(1)Firstly,the physical structure,fault mechanism and frequency characteristics of the bearing are analyzed,as well as the common time-frequency analysis methods for processing unsteady signals.(2)Secondly,for the non-flat characteristic of the bearing fault vibration signal,the improved CEEMDAN is introduced to extract the fault information of the motor rolling bearing.Adaptive white noise sequence is added at each stage of decomposition,and each IMF component is obtained by calculating a unique residual signal,which improves the modal aliasing phenomenon in EEMD and the completeness of the reconstructed signal after decomposition.Through the simulation of the digital-analog signal and the measured signal,the comparison of EMD and EEMD with the completeness,orthogonality and other indicators of the algorithm decomposition verify the superiority of the CEEMDAN decomposition method.At the same time,according to the vibration characteristics of the bearing signal,using the energy values of each IMF obtained by the decomposition of CEEMDAN,the characteristic data set for the diagnosis of the rolling bearing of the motor is formed.(3)Finally,based on the knowledge of statistical theory,the extracted feature data is analyzed.Combined with the SVM classification principle,GWO is used to optimize the penalty factor parameters and kernel function parameters of the SVM.A well-designed support vector machine network is used to train the bearing fault feature data,and the unknown state signal features are classified and predicted.Through the simulation analysis of the measured data of the rolling bearing faults of the motor,the CEEMDAN-SVM based predictive combination model is superior to the EEMD-SVM in the recognition accuracy rate,indicating that the CEEMDAN algorithm feature extraction method is superior to the EEMD under the same recognition network.The prediction combination model based on CEEMDAN-GWO-SVM is superior to CEEMDAN-SVM in the recognition accuracy and efficiency,indicating that under the same feature extraction conditions,GWO optimizes SVM for better network classification and recognition.The CEEMDAN-GWO-SVM hybrid fault diagnosis model proposed in this paper has better feasibility and high efficiency.