Fusion Decision Methods Based On Evidence Reasoning And Its Application In Fault Diagnosis

The classifier based on evidential reasoning(Evidence Reasoning,ER)rule can handle the mapping relationship between input features and output values(class tags).In view of the optimization of structure and parameters in ER classification model,most of the existing studies consider the optimization of ER model structure and parameters separately,and can not realize the collaborative optimization of both.However,model parameters and structure determine the complexity and accuracy of ER model respectively,so it is more necessary to solve them simultaneously in practice in order to find a balanced solution between model complexity and accuracy.Based on this ideal,this thesis proposes ageneralized ER classifier under joint structure and parameter optimization,andapplies this classifier to the fault mode classification of rotating machinery equipment.The main work is as follows:(1)The design of generalized ER classifier under the joint optimization of structure and parameters.First of all,the reference value of the input feature is initialized by k-means,and the input point matrix and reference evidence matrix are constructed by using the sample casting method;then,the referenceevidence is fused with ER rule,and the final decision is made according to the fusion result to obtain the class label of the input feature.In the process of fusion,parallel multi-population strategy and redundant gene strategy are adopted to jointly optimize the structure and parameters,and several groups of parameter individuals of different length(a set of ER parameters)and redundant genes are constructed to ensure that these individuals of different length can carryout cross-mutation operation smoothly.Finally,the classical benchmark data sets of University of California Irvine(UCI)are selected for classification experiments.Making comparison of other typical classifiers,the performance advantages of the new classifier are verified.(2)The Fault diagnosis method of gearbox based on ER fusion model.In order to solve the problem of gearbox fault diagnosis,the fault mode classification method of gearbox is given based on the new classifier in(1).Firstly,the features of collected multi-source vibration signal are extracted,the reference evidence matrices are constructed through the fault feature sample input points,and the reliability factor and importance weight of diagnostic evidence are obtained;then,Using the ER rule to diagnostic evidence fusion and fault decision-making;in the fusion process,the best model parameters are obtained by using the strategy of joint optimization of structure and parameters.Finally,by the comprehensive comparison of the diagnosis results,the advantages of the method in gearbox fault mode classification is verified.(3)Probability transformation method of decision information based on COWA operator.Aiming at the problem of decision uncertainty in fault diagnosis,based on the research of(2),a method of reliability allocation and probability transformation of uncertain decision based on COWA operator is proposed.The reliability matrix is constructed through the trust interval to describe the reliability of decision-making information and measure the information loss of the decision-making layer.Through the combination between this method and the proposed ER fusion model,the accuracy of fault mode classification decision can be effectively improved.Finally,the effectiveness of the method is verified by comparing with the experimental results in (2).