摘要
针对汽轮机转子振动故障的特点,根据Bently实验台所采集的4种典型汽轮机转子振动故障数据,运用分形盒维数、ARMA自谱函数、ARMA模型的二维双隐层神经网络和小波包分析方法研究了振动故障的非线性特征,进行故障诊断。诊断结果表明:不同故障盒维数不同,采用盒维数能够较好的对故障类型进行判别;各种故障的自谱函数幅值分布在不同的频段,有较好地区分度;采用ARMA模型的二维双隐层神经网络进行故障诊断,可以得到各种故障检验样本与目标函数在欧氏空间的最小距离,有较高的故障辨识力;运用小波包分析方法,可以获得汽轮机转子振动的故障状况,根据不同故障发生时的频谱特征,识别出不同的故障。
According to the characteristics of turbine rotor vibration, fauhs nonlinear characteristics are studied by the methods of fi'aetal box counting dimension, ARMA self-spectral function, Euclidean space dual hidden layers neural network of ARMA model and wavelet packet analysis based on the four typical vibration faults data of turbine rotor colleeted from the Bently experiment set. The results show that different faults have different box counting dimension which can be used to diagnose faults. The value of self-spectral function for each fault distributes in different frequency hand and has better ctiscrimination. The minimal distance in the two-dimensional Euclidean space between exam sample and objective function for each fauh can be obtained by Euclidean spaee dual hidden layers neural network of ARMA model, whieh has good tault identification capability. Turbine rotor vibration fauhs station can be obtained by wavelet packet analysis method. According to the characteristics in both the time domain anti the frequency domain of fauhs, character of faults can be identified.
出处
《水利电力机械》
CAS
2006年第6期5-9,共5页
Water Conservancy & Electric Power Machinery
关键词
汽轮机转子
故障诊断
分形
自谱函数
小波包分析
turbine rotor
taults diagnosis
fractal
self - spectral function
wavelet packet analysis