一种基于相邻模块化加权D-S的融合诊断方法

常规D-S (Dempster-Shafter)决策融合方法由于其自身理论不足,不能很好直接处理决策结果偏差大、冲突大的传感器融合问题,因而对于信息高冲突情况下的转子微弱故障融合诊断存在着失效问题。针对该类问题与不足,借鉴复杂网络的舆论传播、社会学习理论及多智能体一致性决策的相关概念与思路,从避免决策结果冲突大的传感器直接进行融合的角度进行改进,提出相邻模块化加权D-S融合方法。该方法首先根据初步结果进行相邻节点与模块划分,只有决策距离在相邻界限值范围内的相邻模块节点才能进行决策融合;对于同一模块内相邻节点,根据各节点决策权重及初步决策结果采用加权D-S融合方法进行决策融合;针对融合结果再进行相邻节点模块划分与融合,依此步骤进行循环划分与融合,直到所有模块与节点均不相邻;最后采用专家权威决策方法确定权重和最大的模块融合结果作为最终的传感器网络一致性决策结果。通过多传感器网络的转子故障模拟实验对所提方法进行验证,应用结果表明:所提方法可以较好解决少数传感器诊断正确、而多数诊断错误的信息高冲突条件下的局部微弱故障融合诊断问题。

Fusion diagnosis method based on neighboring modular weighted D-S

The traditional Dempster-Shafter (D-S) method has disadvantages in dealing with the sensors fusion problems whose decision results are highly deviated and contradicted; therefore, it will also fail in the fusion diagnosis of rotor weak fault in the situation of high information conflict. According to the related ideas and concepts of opinion propagation of complex network, social learning theory and consensus decision of multi-agent, and with a focus on avoiding the direct fusion of the sensors which have great conflict, this paper proposes a new fusion diagnosis method based on neighboring modular weighted D-S. Firstly, we judge whether the sensors are adjacent to each other based on the preliminary results. Only for the neighboring sensors whose distance is within the limited range can the fusion be carried out. For the neighboring sensors which are in the same module, we carry out the fusion based on weighted D-S fusion method according to the preliminary decision results and the weights of different sensors. Subsequently, according to the obtained fusion result, we carry out the module division and fusion again. We repeat the above steps until all the nodes and modules are not neighboring, then output the decision result of a module, whose weights' summation is the maximum. Finally, we use expert authority's decision method to determine the final weight and fusion result, which are used as the consensus decision result of the sensor network. The method is tested through a simulation test of rotor fault and it proves that the presented method can effectively deal with the fusion diagnosis of rotor weak fault in the situation of fusion diagnosis of rotor weak fault with high information conflict.