• CN:11-2187/TH
  • ISSN:0577-6686

›› 2010, Vol. 46 ›› Issue (15): 52-58.

• 论文 • 上一篇    下一篇

基于高斯混合模型的转子碰摩声发射识别方法

邓艾东;包永强;赵力   

  1. 东南大学能源与环境学院;南京工程学院通信工程学院;东南大学信息科学与工程学院
  • 发布日期:2010-08-05

Rub-impact Acoustic Emission Signal Recognition of Rotating Machinery Based on Gaussian Mixture Model

DENG Aidong;BAO Yongqiang;ZHAO Li   

  1. School of Energy & Environment, Southeast University School of Information Engineering, Nanjing Institute of Technology School of Information Science and Engineering, Southeast University
  • Published:2010-08-05

摘要: 基于模态声发射和窄带信号理论,给出描述多模态特征的声发射信号(Acoustic emission,AE)的表达式,提出基于倒谱系数和分形维相结合作为特征参数的高斯混合模型(Gaussian mixture model,GMM)的碰摩声发射识别方法。该识别模型对碰摩声发射信号中不同模态波的特征矢量所具有的概率密度函数进行建模,将这些特征矢量进行聚类,每一类均作为一个多维高斯分布函数,以每一类的均值、协方差矩阵和出现的概率作为每种模态波的训练模板,识别时将待测碰摩声发射信号的特征矢量代入每个模板,采用最大比合并的方法对高斯模型似然概率进行加权得到总似然概率,当该值大于设定的门限,即可判定存在碰摩声发射。在转子试验台上获得碰摩AE信号,根据AE信号在传播过程中的波型并结合其分形曲线进行分类,由此确定GMM的模型输入类型;然后对所有测试数据叠加不同信噪比的高斯白噪声和非平稳噪声,再利用上述模型进行识别。试验结果表明,该模型具有较高的识别率,并具有较好的抗噪声能力。

关键词: 倒谱系数, 分形维, 高斯混合模型, 声发射识别

Abstract: On the basis of modal acoustic emission and narrow-band signal theory, the mathematical expression of multi-modal AE signal is given. A mixed parameter composed of logarithm cepstral coefficients and fractal dimension is presented as the characteristic coefficients of rub-impact AE signal, and the AE recognition system based on Gaussian mixture model is also established. Because the eigenvector of each modal wave in rub-impact AE signal is different, so a model of probability density function of eigenvector is built and it is regarded as a clustering. Training model is established with mean, covariance matrix and probability of each clustering. Sum likelihood probability weighted by maximal ratio combining to likelihood probability of Gaussian model will be obtained when the tested signal is recognized, if the probability is larger than a set threshold, it can be confirmed that rub-impact AE is available. Rub impact AE signals are sampled from rotor test stand, the types of input model of Gaussian are sorted by AE wavrform shapes and its fractal dimensions, then white noise and non-stationary noise are added to simulate the real AE signal. In the end, this Gaussian mixture model is used for AE recognition. The results indicate that the model has high recognition rate and good anti-noise ability.

Key words: Acoustic emission recognition, Cepstral coefficients, Fractal dimension, Gaussian mixture model

中图分类号: