基于主元-双谱支持向量机的旋转机械故障诊断方法  

FAULT DIAGNOSIS APPROACH BASED ON PRINCIPAL-COMPONENT -BISPECTRUM ANALYSIS AND SUPPORT VECTOR MACHINE FOR ROTATING MACHINERY

作　　者：刘占生[1] 窦唯[1] 王晓伟[1] 

机构地区：[1]哈尔滨工业大学,哈尔滨150001

出　　处：《振动与冲击》2007年第12期23-27,共5页Journal of Vibration and Shock

摘　　要：提出一种基于主元-双谱支持向量机的旋转机械故障诊断方法,该方法具有对故障在线监测分类和故障分类能力强的优点,能有效地从汽轮机故障的非平稳、非高斯时变信号中提取状态特征。首先研究了直接法、间接法及ARMA模型法的双谱估计特征,然后利用双谱分析主元数字特征提取技术提取这些特征,最后利用支持向量机对汽轮机小样本故障进行分类,较好地解决了故障样本不足的实际问题。诊断结果表明与常用的BP神经网络相比可以获得较高的诊断精度。The method presented has the ability of online fault monitoring and high classification veracity and can effectively extract the state characters from the non-steady and non-Gaussian time-varying signal of turbine fault. The bis-pectrum estimations by using direct method, indirect method and ARMA modal method are investigated. Then the principal component numeric character technology based on bispectrum analysis is used to extract the characteristis. At last, support vector machine is introduced to classify sample faults of turbine and to solve fairly well the actual problem of inadequate number of fault samples. The diagnosis results show that the method can reach higher diagnosis veracity compared equations of shallow shell were transformed to the nonlinear integral-differential equations by using the method of Green's function. To solve the integral-differential equations, expansion method was used to obtain Green's function. Then the integral-differential equations were reduced to the form with degenerate core by expanding Green's function into series of characteristic functions. Therefore, the integral-differential equations became nonlinear ordinary differential equations with respect to time. The amplitude-frequency response under harmonic force was solved by considering the vibration shape of single mode. As a numerical example, forced vibration phenomena of shallow spherical shells with sinusoidal corrugation were studied. The obtained solutions are available for reference to design corrugated shells

关 键 词：故障诊断 特征提取 双谱 支持向量机 旋转机械 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TP206.2[自动化与计算机技术—控制科学与工程]