Research On Multiple Faults Diagnosis Based On Information Fusion

In complex industrial process(such as rolling process),because of equipments running for a long time and the internal and external conditions changing over time,multiple faults may occur simultaneously,among which,there are always notorious characteristics,such as coupling.And because of the harsh working environment of equipments,the fault characteristic parameters obtained by a single sensor are often vague,uncertain,which will lead to an error diagnosis result.In order to achieve the comprehensive and accurate multiple faults diagnosis,it needs to make full use of the fault information from multi-source synthetically.In view of the complex relationship between multi-faults and symptoms,this thesis uses a three-stage information fusion system for multi-faults diagnosis,and focuses on researching the multi-faults diagnosis approach,which can draw an accurate conclusion aiming at the complicated situation of independent faults and coupling faults occurring simultaneously.The main research work is as follows:Firstly,applying BP(Back Propagation)network approach obtains the diagnostic result of feature level.According to the problem of BP network that is easy to fall into local extremum in the network learning and training process,this thesis adopts genetic algorithm to optimize the original weights and thresholds.At last,it verifies that the method of improved BP network is more effective by a numerical example.Secondly,research on the theory of classical DSm(Dezert-Smarandache)to obtain the diagnostic result of decision-making level.Compared to the theory of D-S evidence,DSmT not only can represent and deal with uncertain information,but also can represent and deal with high conflict information,and what is most important is that DSmT can deal with coupling information.At last,it verifies that the method of classical DSmT is fit for the pure coupling multiple faults diagnosis by the test data.Thirdly,in view of the problem of classical DSmT that can't handle the situation with constraint condition,hybrid DSmT is studied.The prior constraints in hybrid DSmT,which are summarized according to historical data,the knowledge of specialist experience,and so on,make the diagnosis model more accuracy in the line with the actual situation.And the generalized basic probability assignment will be redistributed based on the constraint model,which make the diagnostic result more obvious.Through the simulation experiments and analysis compared with classical DSmT,it verifies the superiority of hybrid DSmT.Last but not least,research on the hierarchical DSmT to obtain the diagnostic result of decision-making level.According to the problem of hierarchical DSmT that is lack of the hierarchical rules at present,which include the physical meaning,a method is proposed based on the principal components hierarchical clustering.Then,the diagnostic tests show that the method of multiple faults diagnosis based on hierarchical DSmT not only reduces the calculation amount,but also recognizes independent faults and coupling faults from framework,which can make contribution to reducing diagnostic uncertainty.