转子全周碰摩与局部碰摩的识别方法研究  

Recognition method for rotor system under full annular rub and partial rub

作　　者：熊炘[1] 杨世锡[1] 甘春标[1] 周晓峰[1] 

机构地区：[1]浙江大学机械工程学系流体传动及控制国家重点实验室,杭州310027

出　　处：《振动与冲击》2012年第16期13-17,共5页Journal of Vibration and Shock

基　　金：国家自然科学基金(11072214);国家"863"高技术研究发展计划(2008AA04Z410)

摘　　要：超临界汽轮发电机组的结构和工况复杂,容易引起转、静子间的碰摩。根据碰摩诱发因素的不同,可将其分为全周碰摩与局部碰摩。由于两种碰摩故障的时、频特征相似,传统的时、频域分析方法很难准确提取它们的故障特征。针对这一不足,提出一种基于经验模式分解-奇异值分解(EMD-SVD)与支持向量机(SVM)的碰摩故障识别方法,用于对转子全周碰摩与局部碰摩故障进行识别。首先,通过EMD获取碰摩信号的固有模式函数(IMF);然后,提取表征信号主要能量的前四阶IMF组成特征矩阵并进行SVD分解,得到关于原信号的一组特征值;最后,将特征值输入SVM,对原信号进行分类识别。转子试验台全周碰摩与局部碰摩试验结果表明,该方法对转子全周碰摩与局部碰摩故障的分类准确率高,其中以径向基函数作为核函数的SVM分类准确率达到96.0%。Supercritical steam turbosets are highly complex in their structure and always run under various complicated working conditions. They are prone to rub between static and dynamic parts. Induced by different factors, rubbing faults can be divided into full annular rub and partial rub. Since the time-frequency characteristics of both are similar to each other, discrimination of the two kinds of rub faults is hard to proceed by using the traditional spectrum analysis methods. In response to make up for this shortage, an intelligent recognition method based on the EMD-SVD and SVM was proposed. IMFs were collected through EMD and the first-four-order IMFs, which contain the main power of the original signal, were extracted to form the characteristic matrix. The SVD was applied to obtain a series of eigenvalues, which were then inputted to train the SVM in order to classify rub faults. The newly developed intelligent reeogonition method was used to analyze the signals collected from rotor test-bed under both full annular rub and partial rub conditions. The experiment results show that, classification accuracy of the method is high, especially for the SVM using radial basis as kernel function, where the classification accuracy is up to 96.0%.

关 键 词：转子 全周碰摩 局部碰摩 经验模式分解 奇异值分解 支持向量机 

分 类 号：TN911.7[电子电信—通信与信息系统] TH165.3[电子电信—信息与通信工程]