It is difficult for the evidence in a D-S evidence theory application to determine the distribution of target-mode confidence function. In the light of this problem the authors have proposed a typical swatch-based method for the acquisition of confidence function distribution. This method utilizes the Hamming distance between the evidence and the typical swatch of each target mode to construct the distribution of confidence function, thus meeting the definition of confidence function distribution and reducing its subjectivity. By using this data-fusion method in the fault diagnosis of a boiler milling system it is feasible to identify such pulverizer malfunctions and faults as pulverized coal self- ignition, pulverizer being full of coal and empty of coal, etc. As verified by historical data, the method under discussion can effectively recognize various types of faults and make an early prediction and diagnosis of ensuing malfunctions.
Journal of Engineering for Thermal Energy and Power
evidence theory, typical swatch, fault diagnosis, coal milling system