Data Classification Based On Attribute Vectorization And Evidence Fusion And Its Application In Fault Diagnosis

Evidence reasoning(ER)rule is an important extension of Dempster-Shafer(DS)evidence theory,it provides a rigorous probabilistic reasoning process and clearly distinguish the difference between reliability and weight of evidence,therefore,it is one of the representative methods to deal with the uncertainty in various areas.Classifiers based on ER rule can well handle the uncertainty in the mapping relationship between input attributes and output classes.However,with the growing number of input attributes,the parameters of original ER(O-ER)classifiers increase significantly,which adds the computational burden in inference process and model optimization.So,a classifier based on attribute vectorization and evidence reasoning(AV-ER)is proposed,and is applied to deal with the classification problems for specific equipment failure(fault diagnosis),The main work is following:(1)General data classification based on attribute vectorization and evidence fusion.Firstly,different input attributes are combined into attribute vectors by using principal component analysis(PCA),and the reference evidence matrix is generated based on the attribute vector.After that,ER rule is used to combine all pieces of activated reference evidence and to make the classification decision according to the fusion result.In the fusion process,genetic algorithm(GA)is used to optimize the parameters of initial classifier model.Finally,the typical University of California Irvine(UCI)benchmark datasets are applied to verify the proposed classification model as the same time the implementation process of the classifier(AV-ER)is described in detail,and then Akaike information criterion(AIC)is used to evaluate the model performance.And the performance of the AV-ER classifier is compared with that of the O-ER classifier,the results indicate that advantages of the proposed classifier.(2)Fault diagnosis for rotating machinery method based on fault feature vectorization and evidence fusion.Aiming at the specific fault diagnosis for rotating machinery,the corresponding fault classification method is given based on the AV-ER classifier in(1).Firstly,the vibration signals collected from multiple information sources are extracted,then the reference evidence matrix of fault feature vector is generated and the corresponding evidence reliability based on the PCA method is given.Secondly,ER rule is used to combine all pieces of evidence,and make the decision according to the fusion result.In the fusion process,genetic algorithm is used to optimize the parameters of initial model.Finally,some diagnosis experiments on a rotor test bed are conducted,and the effectiveness of the method is verified through a comprehensive evaluation of the experimental results.(3)Fault diagnosis for rotating machinery under imbalanced sample conditions.In order to solve the typical problem of imbalance between the number of normal samples and the number of fault samples in fault diagnosis,a hierarchical ER fusion method is introduced on the basis of the diagnosis method in(2).In this way,more relevant historical samples are involved in fusion process to increasing the amount of the effective information,so as to improve the model performance.At last,some diagnosis experiments on a rotor test bed are conducted to compare the proposed hierarchical ER fusion method with the single level ER fusion method.The experimental results show that the new method can effectively promote the diagnosis rate.