基于多传感器信息融合和神经网络的汽轮机故障诊断研究  

Research on fault diagnosis of turbine based on multi-sensor information fusion and neural network

作　　者：凌六一[1] 黄友锐[1] 魏圆圆[2,3] 

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001 [2]中国科学院合肥智能机械研究所,安徽合肥230031 [3]中国科学技术大学信息学院,安徽合肥230027

出　　处：《中国电力》2010年第3期46-50,共5页Electric Power

基　　金：国家863计划资助项目(2006AA10Z237);国家自然科学基金资助项目(60774096)

摘　　要：针对传统故障诊断方法存在的诊断准确性不高的问题,提出了基于D-S证据理论的多传感器信息融合技术与BP神经网络相结合的方法,实现对汽轮机的机械故障诊断。由多个传感器采集振动信号,分别经小波变换特征提取后获得故障特征值,再经BP神经网络进行故障局部诊断,得到相应传感器对故障类型的基本可信任分配函数值,即获得彼此独立的多个证据,然后运用D-S证据理论对各证据进行融合,最终完成对汽轮机机械故障的准确诊断。实验结果表明,该方法克服了单个传感器的局限性和不确定性,是一种有效的故障诊断方法。For the reasons of low fault diagnosis accuracy of traditional diagnosis methods, a fault diagnosis method fusing BP neural network and muhi-sensor information fusion technique based on D-S evidence theory was presented to realize machinery fault diagnosis of turbine. The fault features of the vibration signals multi sensors sample were extracted by using wavelet transform, and after these fault features were locally diagnosed through BP neural network the basic reliability distribution values of corresponding fault were got, namely multi independent evidences were got. Then all the evidences were fused using D-S evidence theory and veracious machinery fault diagnosis of turbine was realized. Experiment result shows that the presented method of fauh diagnosis overcomes the limitation and uncertainty of single sensor and it is a valid method.

关 键 词：故障诊断 信息融合 BP神经网络 证据理论 汽轮机故障 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TK267[自动化与计算机技术—控制科学与工程]