Fault diagnosis method for machinery based on multi-source conflict information fusion

Abstract

Multi-source information fusion diagnosis is usually more reliable than fault diagnosis with a single source employed. However, fusion results may be absurd when fusing highly conflicting information. To address this problem, the Dempster–Shafer (DS) evidence theory is updated by weighting each piece of evidence according to the corresponding contribution to the decision, and a novel fault diagnosis method based on multi-source conflict information fusion is proposed. First, the basic probability assignment of evidence corresponding to the sensor information is given by introducing the feature parameters of electromyographic signals and using the back-propagation neural network. Then, the importance of each piece of evidence is determined by solving the difference degree and exclusion degree among the evidence, and the evidence is assigned weights according to the degree of importance of each piece of evidence in the fusion decision-making process. Next, the weighted evidence is combined for making decisions and further diagnosis after weighted averaging of the evidence with different weights. Finally, the performance of the proposed method is assessed using receiver operating characteristic (ROC) curves. The experimental results show that the areas under the ROC curves for the proposed method are 0.3229, 0.0729 and 0.9271 higher than those of the traditional DS method, Murphy’s method and Yager’s method, respectively, which proves that the proposed method has better diagnostic performance and reliability.