The Similarity Measure Method Of Conflicting Evidence And Its Application In Information Fusion Fault Diagnosis

As an information fusion method, Dempster-Shafer evidence theory taking the advantages of representing, measuring and combining the uncertain information, has been widely used in fault diagnosis field. The processing of solving the diagnosis problem based on the evidence theory can be divided into four steps: deciding the fault frame of discernment, obtaining the diagnosis evidence from information the sensors collected, combining the evidence and making decision. Since the sensors are disturbed commonly by the environmental noise or human disturbance, it makes the diagnosis evidence conflict with each other, when the evidence is combined by using the classic Dempster combination rule, there will be an unreasonable result. Therefore, it is very important to give a reasonable measure of conflicting eviedence and combine them in the fusion diagnosis.In fact, the diagnosis evidence can be seen as a vector that measures the credibility that the faults happen, the vector is obtained by converting the original evidence into the specific space of credibility.So,the problem of conflicting evidence can be transformed into a measure of vector similarity problem in the space of credibility. This thesis starts from the similarity of vectors in the space of credibility, gives the method for measuring conflicting evidence, and it is applied to solve the fusion problem of conflicting evidence in the fault diagnosis. The main contributions are as follow:(1) Give an overview of theory evidence, summarize the procedure of completing the fusion fault based on evidence theory; analysis the main causes of generating the conflicting evidence,introduce the existing methods of the measurement of conflicting evidence and combination them in fault diagnosis, point out some drawbacks of these methods.(2) The fusion method of conflicting evidence based on the measurement of similarity of evidence in the Pignistic probability space (a type of space of credibility). The body of evidence is transformed to Pignistic evidence vector (PEV) in the Pignistic space. The vector cosine is used to measure similarity between two PEVs. The method for measuring the degree of conflicting evidence is given on this opinion. Finally, numerical examples illustrate the effectivity of the proposed method. (3) The information fusion algorithm of fault diagnosis based on the reliability evaluation of evidence. The reliability of the fault evidence is considered from the dynamic and static aspects, a static discounting factor is given by a process of optimizing indicator function about Pignistic probability measure, a dynamic discounting factor is presented using cosine similarity measure of the revised evidence based on Pignistic vector, this can solve the problems that the reliability evaluation of diagnosis evidence and combination.(4) The fusion method of diagnosis evidence based on the dynamic updateing of evidence. In order to deal with the problems existed in dynamic fusion of diagnosis evidence, the conflict factor about evidence generated at continuous time are obtained based on the proposed measurement of similarity between the evidence,the updating rule is selected according to the values of the factors (class Jeffery or the conditioning and updating rule) to achieve fault-line diagnosis. The effectivity of the proposed method is verified through the motor rotor fault diagnosis experiment.