基于SVM和证据理论的多数据融合故障诊断方法  

Multi-data fusion fault diagnosis method based on SVM and evidence theory

作　　者：姜万录[1] 吴胜强[1,2] 

机构地区：[1]燕山大学机械工程学院,秦皇岛066004 [2]邢台职业技术学院,邢台054035

出　　处：《仪器仪表学报》2010年第8期1738-1743,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(50775198);河北省自然科学基金(E2008000812)资助项目

摘　　要：针对D-S证据理论很难确定基本概率分配(BPA)及支持向量机(SVM)的硬判决难获得概率输出的缺陷,融合D-S证据理论和SVM算法提出了一种多数据融合故障诊断新方法:利用"一对一"多类SVM分配了BPA,引入基于矩阵分析的融合算法解决了证据理论存在的计算瓶颈问题。对液压泵进行了试验,首先,采集了柱塞泵松靴、缸体与配流盘磨损等故障信号,应用小波包对采集的信号进行了预处理,提取了12个时频特征量;最后,用所提出的基于SVM和证据理论的多数据融合新方法进行了诊断。试验结果表明,新方法故障确诊率高,诊断有效。Aiming at the difficulty that evidence theory can hardly determine basic probability assignment （BPA） and SVM can hardly obtain probability output,a new multi-data fusion fault diagnosis method is proposed,which is based on SVM and D-S evidence theory.In the method,BPA is assigned based on one-versus-one multi-class SVM;a fusion arithmetic based on matrix analysis is presented to solve the calculation bottleneck problem of evidence theory.The method is tested on a hydraulic pump;at first,the fault signals of slipper looseness and wearing between cylinder body and valve plate are collected;the measured signals are preprocessed using wavelet cluster;12 fault features are picked up in time domain and frequency domain;at last,the fault is diagnosed using the proposed new method.Experiment results show that the new method features high correct diagnosis rate and is effective in fault diagnosis.

关 键 词：故障诊断 SVM D-S证据理论 多数据融合 小波包 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]