A New Method for Decision Fusion and Its Application in Multi-Model Fault Diagnosis

Multi-model decision fusion is an effective method for machinery fault diagnosis. In this paper, a new fusion formula combining two parameters of decision reliability and sample recognizability is presented. These parameters can reflect the performances of recognizers, and they assign different weights to each recognizer. Between them, the decision reliability is approximated by the accuracy rate of the recognizer, and the sample recognizability is obtained through modelling by support vector data description (SVDD). Data shows that this method can effectively avoid fusion failure while using highly conflict evidences, and it has stronger reliability and robustness than some other rules in practice.