Bearing Fault Diagnosis Based On Local Mean Decomposition Sample Entropy And Parameter Transfer Learning

In the operation of rotating machinery,rolling bearing,as a vulnerable part,is prone to be damaged.Therefore,the study of fault diagnosis of rolling bearing is meaningful.Based on the analysis of the research status of rolling bearing faults,a fault diagnosis method for rolling bearings based on local mean decomposition(LMD)sample entropy and parameter transfer learning is proposed in this paper,aiming at the three aspects of vibration signal decomposition,signal component screening,eigenvector matrix construction and classification and recognition.First of all,the characteristics of Empirical Mode Decomposition(EMD)method are analyzed,and its defects such as endpoint effect and frequency aliasing are pointed out.A feature extraction method of vibration signals based on LMD is proposed.LMD method adaptively decomposes vibration signals into the sum of product function(PF)components with physical significance,which reduces the influence of endpoint effect on decomposition accuracy and overcomes the disadvantage of Hilbert decomposition for each signal,so that the decomposition has better adaptability and accuracy.By respectively implementing EMD and LMD methods on the same signal.The time-frequency characteristics of the decomposition results show that LMD method can better suppress the endpoint effect and frequency aliasing.Then,spearman correlation coefficient is used to screene the PF components obtained from the decomposition of vibration signals by LMD,and screene out the components sufficient to characterize the characteristics of the original signals.Sample Entropy(SampEn,SE)is calculated to form a feature matrix to highlight the fault information in the original vibration signals and provide feature vectors for bearing fault classification and recognition.Next,by analyzing the limitations of traditional machine learning in classifying and recognizing feature signals,a new method of classifying and recognizing feature signals based on parameter transfer learning is proposed.By specifying the source domain and target domain,the mapping function is constructed by analyzing the source domain and target domain.Then the source domain and target domain feature vectors are migrated and mapped to get the re-mapping of the target domain.Finally,the target domain parameters are classified and identified.The simulation results show that the fault identification method based on parameter transfer learning has better accuracy.Finally,experimental verification and practical application of the proposed method are carried out.Firstly,the rolling bearing fault signal of Case Western Reserve University(CWRU)is adopted,SVM and the proposed method are implemented on the fault signal to diagnose the different position faults and the different degree faults of the same position of rolling bearings,respectively.The experiment results show the effectiveness and superiority of this method.Then the method was applied to the data collected by BaoGang in Shanghai,and good diagnostic results were obtained.