Fault Diagnosis Method Based On Interval-valued Belief Structure

The traditional D-S evidence theory requires single precise values as supportevidences, but for various reasons, a single-valued form of evidence is difficult toobtain in the real world, even if available, they are often unreliability. Moreover,single-valued evidence is not perfect for the information expression; for this reason,some scholars proposed an improved Interval-valued Belief Structures (IBS) theory todeal with this situation. Such evidence can fully meet the ordinary human-thinkingand quantify uncertainty information. But with refinement of this theory, itsnormalization problems and update issues need further research. To the end, this paperpresents a new method to deal with them.The current fault diagnosis system based on multi-sensor information fusion hasbecome a mainstream, that is, troubleshooting applications under the framework ofD-S evidence theory has become a major trend. With the complexity of modernindustrial equipments increasing, fault diagnosis technology has been put forward tohigher requirements, the traditional evidence-based static diagnostic methods havebeen hardly adapt to the future needs of real-time diagnostics, this paper presents anew real-time updating strategy based on IBS framework.In this paper, I studied several areas, as follows below:1. Introduce the basic knowledge and research status of traditional D-S evidencetheory and Interval-valued Belief Structures, and with a detailed analysis of theshortcomings of these theory. Finally, the problem of dynamically updatedapplications for real-time fault diagnosis elaborated existing update methods andrelated progress.2. This paper presented a normalization method by optimizing its discount togenerate standard IBSs based on the Interval-valued Belief Structures (IBS) ofreliability value. This method works by solving the optimal discount factor, so thegenerated normalized standard IBSs contain the maximum amount of informationfrom the original IBSs. Also we defined similarity measure between two IBSs basedon Euclidean distance interval vectors to measure similarity degree, with its measureof the extent of the information contained common information between them. Bytypical numerical examples and large-scale statistic randomized experiments, weproved the effectiveness of the proposed optimization methods.3. On the basis of Interval-valued Belief Structures, we propose a new updating strategy applied in on-line fault diagnosis. This approach extends the traditionalJeffrey-like rules and linear combination rules, according to the changes of newdiagnostic evidence in real time, and selecting the appropriate adaptive updating rulesautomatically; and diagnose the evidence before the new diagnostic evidence andbased on a short update the relationship between the combined adaptive linearcombination weights. As can we seen from the last series of fault simulationexperiments, and compared to other strategys, our strategy has the best performanceoverall.