Information Fusion Fault Diagnosis Based On Rough Set And Evidence Theory

The diagnosis result gotten by information fusion,which is to deal with multi-source fault features with uncertainty,is more accurate and reliable than the result gotten from any single source feature information.There are two problems need to be focused on when using this method in practice.The first is how to reduce the multi fault attribute and use fewest features to diagnose all types of fault,at the same time,such that the real-time requirement of the equipment diagnosis should also to be satisfied;and the second is how to model and fuse the features data considering the accuracy of fault diagnosis,that is to say the fusion result could reflect the real state of equipment.This paper makes full use of rough set theory and evidence theory to solve the above problems.For the diagnosis of rotating machinery,we propose an integrated information fusion method,includes feature extraction,reduction,the modeling of uncertain fault feature,fusion and fault decision.The main works are follows:(1)The fault feature data discretization based on k-means and random fuzzy variable.The k-means is used to divide all historical fault sample data of a fault feature into several clusters.Each cluster corresponds to a discrete value,and RFV model of each cluster is established.In the sense of evidence theory,the discrete confidence expression of each sample can be obtained by matching the historical sample data with the RFV models,and the corresponding discrete values are determined according to the principle of maximizing belief.Finally,the fault feature is regarded as the condition attribute and the fault type is regarded as the decision attribute,and the decision information system about the fault history sample is set up.(2)Fault features reduction based on belief interval.In view of the decision information system from(1),the compressed binary matrix is set up to obtain the core attribute set.Each of rest attribute and the core attribute will make up a new attribute set,each new attribute set is calculated to get upper and lower approximation of all equivalence classes of decision attribute,then get the corresponding belief interval and its width by combining the approximation relationship and belief measure degree of evidence theory on rough set.Then we will put the attribute with minimum interval width into the core attribute set,thus the attributes reduction result will be obtained.(3)Fault diagnosis method based on KNN evidence fusion.According to the reduction result from(2),the historical data in different fault type are used to model fault template patterns with form of RFV,the KNN algorithm is used to find out K historical samples nearest to a testing sample and the RFV-type fault testing patterns of the K samples are presented to describe the testing sample.The matching degree between both of them can be calculated to generate the K pieces of diagnosis evidence,and then all evidence coming from the different fault features can be fused and diagnosis decision can be made based on the fused result.