Information-fusion method for fault diagnosis based on reliability evaluation of evidence
Chenglin Wen · Control theory & applications · 2011
In fault diagnosis methods based on evidence theory with information fusion, the reliabilities of evidences will affect the accuracy of diagnosis results. However, most existing fusion diagnosis methods do not take the reliabilities of the evidences into account comprehensively. The main factors which determine the reliability of evidence are the precision of individual sensor and the performance of the method in obtaining the evidence, as well as the uncertainties in the observation environment. They are considered static factors and dynamic factors. The original evidence is first modified by a static discount-factor obtained by optimizing the indication function of Pignistic probability measure. This result is further modified by a dynamic discount-factor which is obtained by applying the measurement method to evidence similarity in Pignistic vectors. Double-modified evidences are combined by Dempster combination rule to obtain the final diagnosis results. Experiments on the multi-functional flexible rotor-testing validate the effectiveness of the proposed method.