Application Of CEEMDAN Method In Feature Extraction And Pattern Recognition Of Rolling Bearing Faults

Rolling bearing plays an extremely important role in mechanical equipments.Once the running state of the bearing deteriorates,it will cause equipment failure.The chain reaction will further affect the safe operation of the equipment.Therefore,it is necessary to carry out the section of the running bearing.The vibration signals of rolling bearings is analyzed to diagnosis faults in this thesis.The main contents are as follows.(1)The vibration signal processing methods of rolling bearing is investigated comparatively.The empirical mode decomposition(EMD)method is often used for non-stationary signal analysis with its unique advantages,but there exists the modal aliasing.Ensemble empirical mode decomposition(EEMD)is proposed to overcome the modal aliasing to some degree.The above-mentioned defect will occur in EEMD due to inappropriate selection of parameters and this will affect the accuracy of fault diagnosis.Therefore,complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)method is developed.The signal decomposition results from the three methods are compared and analyzed using simulation signals.The comparative analysis shows that the CEEMDAN method can effectively overcome the above-mentioned defect.(2)The fault feature of rolling bearing is extracted.CEEMDAN is used to decompose the collected vibration signal and intrinsic mode functions(IMFs)are obtained.The evaluation indicators of kurtosis and correlation coefficient are used to filter IMF components with a large amount of fault information.Independent component analysis(ICA)is performed on the selected components.The Fast ICA algorithm is used to process the filtered IMF components,and then the filtered signal is reconstructed.The envelope spectrum is analyzed and the fault frequency is extracted.The CEEMDAN-ICA signal processing method is verified by using the bearing fault signal from Western Reserve University.The results show that the method can isolate the characteristic frequency when the bearing fails,which realizes the fault diagnosis of the bearing.(3)The fault pattern recognition of rolling bearing is carried out.CEEMDAN is used to decompose the data collected by the acceleration sensor under different states and IMFs are obtained.Then,the IMF components with high correlation with the fault signal is filtered.The sample entropy of the IMF is calculated to construct the feature vector.Then the Fisher score is used to reduce the dimension of l feature vector.The reduced feature vector is input into the support vector machine(SVM)optimized by particle swarm optimization(PSO)for pattern recognition under different fault conditions.the vibration signals of different bearing states from the laboratory of Western Reserve University are used to verify the above method.The results show that this method can show a high recognition rate in the pattern recognition under different bearing states.