基于神经网络的旋转机械监测参数的信息融合技术  

Information Fusion Technique of Monitoring Parameters in Rotary Based on Neural Networks

作　　者：张淑清[1] 靳世久[1] 吕江涛[2] 

机构地区：[1]天津大学仪器科学与技术博士后,天津300072 [2]燕山大学电气工程学院

出　　处：《电子测量与仪器学报》2005年第3期15-17,共3页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金资助项目(60102002);河北省自然科学基金资助项目(601224);霍英东基金资助项目(81057)。

摘　　要：本文考虑到旋转机械故障诊断中监测多个参数,且各参量之间又有主次之分,提出将传统神经网络分解为多个子网络方法。不但可以突出各参量对不同故障的敏感程度,还可防止不同征兆域的诊断结论不一致甚至相反的情况。该融合结构可通过子网络连接权值的修正实现针对不同诊断对象的信息最优分化和融合。最后用DS推理融合得到最终的诊断结果。该融合结构被应用在试验中取得了良好的效果,进一步验证了其可行性和优越性。Considering various monitoring parameters in the fault diagnosis of rotary, and the distinction between the significance of those parameters, a method of substituting the traditional neural networks to several subnets are given in this paper. Not only giving prominence to the sensitivity of parameters to differem kinds of malfunction, but also disaccord of diagnostic conclusion in different symptom domains was avoided. Through the correction of the linked weight parameters of subnets, the optimum differentiation and fusion are implemented according to different objects to be diagnosed. Furthermore, D-S inference fusion is applied to obtain the ultimate result. Experiments verified the practicability and superiority of the fusion structure.

关 键 词：信息融合 故障诊断 神经网络 征兆域 信息融合技术 监测参数 旋转机械 机械故障诊断 融合结构 网络方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP277[自动化与计算机技术—计算机科学与技术]