Information fusion-based fault diagnosis method via hierarchical evidence reasoning

In the framework of information fusion, this paper presents a hierarchical evidence reasoning (ER)-based fusion method to deal with accurate acquirement problem of diagnosis evidence and the information limitation problem in online diagnosis. In the process of diagnosis evidence acquirement, the casting strategy using reference values of fault features is proposed to proportionally calculate the similarity degree between feature sample and its neighboring reference values. In this way, the reference evidence matrix (REM) with the form of point values can be obtained and then the accurate diagnosis evidence of online fault feature sample can be generated by the REM. In the process of evidence fusion, the hierarchical ER fusion model is designed which includes two-level fusion operations. In the first level operation, k-NN algorithm is used to search for the k historical samples close to the online sample and then ER rule is used to fuse the k+1 pieces of diagnosis evidence of k historical samples and the online sample. In the second level operation, the multiple first level fusion results coming from different features can be fused again. Thus the diagnosis decision can be made according to the second level fusion results. Furthermore, the objective fusion is constructed based on evidential Euclidean distance to optimize the importance weights of evidence. Final, some diagnosis experiments on a rotor test bed are conducted to compare the proposed hierarchical ER fusion method with the previous single level ER fusion method. The experimental results show that the new method can effectively promote diagnosis rate by enhancing the evidential accuracy and adding historical sample for information expansion. © 2020, Editorial Department of Control Theory & Applications South China University of Technology. All right reserved.