摘要
在经验模式分解(Empirical Mode Decomposition,EMD)的基础上,结合人工神经网络技术提出一种超声缺陷信号的分类方法.首先利用EMD对信号进行分解获得多个平稳的IMF(Intriusic Mode Functions)分量,然后对各IMF分量分别在时域和频域求出一组特征值,由这些特征值构造向量,作为识别信号的特征向量.采用BP神经网络作为诊断决策分类器,神经网络模型中输入节点对应信号特征向量,输出节点对应缺陷类型.详细地论述了这种方法的基本原理及实现过程.通过典型人工缺陷样品的回波信号的试验,表明该方法对于检测此类缺陷信号有较好的效果.
The original ultrasonic flaw signals are decomposed into a finite number of stationary intrinsic mode functions(IMF) by empirical mode decomposition(EMD),and then a set of eigenvalues are obtained in time domain and frequency domain from the IMF components.The signal eigenvector is constructed by the eigenvalues for identification.BP neural network is used as diagnosis decision-making classifier.In neural network model,the input node corresponds to the signal eigenvector for identification and the output node corresponds to the flaw type.The basic theory and the course of implementing of this method are discussed in this paper.The experimental results by typical artificial flaw echo signals show that the method has better performance in detecting such flaw signals.
出处
《中北大学学报(自然科学版)》
CAS
北大核心
2012年第5期598-602,616,共6页
Journal of North University of China(Natural Science Edition)
关键词
超声信号
EMD
特征向量
神经网络
ultrasonic signal
empirical mode decomposition
eigenvector
neural network