Research On Fault Diagnosis Of Rolling Bearing Based On Entropy Algorithm

As the core component of rotating machinery,rolling bearing is prone to damage in practical work due to the harsh working environment and improper use,which in turn affects the operation of the entire equipment.Therefore,it is urgent to carry out intelligent fault diagnosis research on bearings,the essence of intelligent fault diagnosis technology is to use the fault features extracted from vibration signals and combine machine learning algorithms for pattern recognition.The bearing vibration signal usually presents nonlinear and non-stationary characteristics,and the feature extraction is difficult,but the entropy algorithm has unique advantages for the analysis of such signal.In this thesis,under the background of fault diagnosis of rolling bearing,multiscale diversity entropy(MDE)and multiscale permutation entropy(MPE)are improved and applied to feature extraction to improve the diagnostic accuracy.The specific research is as follows:(1)Aiming at the problem that the MDE algorithm cannot effectively extract features in the fault diagnosis of rolling bearing,a MDE algorithm based on coarse-grained and similarity optimization is proposed by this thesis.First,by setting the range of the sliding factor,the overlapping area of the sequence in the coarse-grained process is reduced,and then the sequence entropy value is calculated to extract the feature with better separability at multiple scales.Secondly,the modified cosine similarity is introduced to optimize the similarity measurement method in the MDE algorithm to improve the sensitivity to the amplitude difference of vibration signals of different fault types.Finally,the improved MDE algorithm is used to extract the features of the vibration signal,and the Relief-F algorithm is used to filter the feature.The feature with larger weights is input into the Support Vector Machine(SVM)to complete the classification.The Paderborn University(PU)dataset and Case Western Reserve University(CWRU)dataset are used to carry out simulation experiments.The experimental result shows that the proposed method has higher diagnostic accuracy and robustness.(2)Aiming at the problem that permutation entropy(PE)does not consider the amplitude difference of vibration signals,which leads to low fault diagnosis accuracy,a PE algorithm based on amplitude information optimization is proposed by this thesis.Firstly,the boxplot algorithm is introduced to calculate the abnormal threshold of the vibration signal,and the phase space reconstruction of the vibration signal is carried out to obtain a series of components,the elements in the components are compared with the abnormal threshold to generate the abnormal flags.Secondly,the symbol sequence is constructed by using the arrangement of components and the abnormal flags,and then the statistical probability of the symbol sequence is calculated to obtain the entropy value,so as to improve the recognition ability of the PE algorithm to the abnormal amplitude.Finally,the coarse-grained process is added to the improved PE algorithm to extract the feature of the vibration signal and input to the Extreme Learning Machine(ELM)to complete the classification and realize fault diagnosis.The Logistic system is used to verify that the entropy calculated by the improved PE algorithm has a high consistency with the complexity of the system.The fault diagnosis simulation experiment is carried out on the PU data set,and the experimental result verifies the effectiveness of the proposed method.(3)Aiming at the lack of completeness of single entropy feature set,a feature integration model based on improved MDE and improved MPE is proposed by this thesis to further improve the accuracy of fault diagnosis.Firstly,based on the ensemble learning framework,the improved MDE and the improved MPE are used as feature extractors to extract two kinds of different feature respectively.Secondly,the concat strategy is used to stitch the two types of feature to obtain integrated feature.Finally,it is input into SVM to complete classification and realize fault diagnosis.Simulation experiments are carried out using PU data sets.The experimental result shows that the proposed integrated model can effectively improve the accuracy of fault diagnosis.